{"data":[{"id":"10.48448/p1jj-7t40","type":"dois","attributes":{"doi":"10.48448/p1jj-7t40","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2025","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Xin Luna","familyName":"Dong","name":"Dong, Xin Luna","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Mohammad","familyName":"Kachuee","name":"Kachuee, Mohammad","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Mohammad","familyName":"Kachuee","name":"Kachuee, Mohammad","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Minseok","familyName":"Kim","name":"Kim, Minseok","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Minseok","familyName":"Kim","name":"Kim, Minseok","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Anuj","familyName":"Kumar","name":"Kumar, Anuj","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Meta"}],"givenName":"Anuj","familyName":"Kumar","name":"Kumar, Anuj","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Yue","familyName":"Liu","name":"Liu, Yue","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Xin","familyName":"Luna Dong","name":"Luna Dong, Xin","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Meta"}],"givenName":"Kai","familyName":"Sun","name":"Sun, Kai","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Jiaqi","familyName":"Wang","name":"Wang, Jiaqi","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Wen-tau","familyName":"Yih","name":"Yih, Wen-tau","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Wen-tau","familyName":"Yih","name":"Yih, Wen-tau","nameIdentifiers":[]}],"titles":[{"title":"PrismRAG: Boosting RAG Factuality with Distractor Resilience and Strategized Reasoning"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2025,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Artificial Intelligence"}],"contributors":[],"dates":[{"date":"2025-10-10T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Retrieval-augmented generation (RAG) often falls short when retrieved context includes confusing semi-relevant passages, or when answering questions require deep contextual understanding and reasoning. We propose an efficient fine-tuning framework, called PrismRAG, that (i) trains the model with distractor-aware QA pairs mixing gold evidence with subtle distractor passages, and (ii) instills reasoning-centric habits that make the LLM plan, rationalize, and synthesize without relying on extensive human engineered instructions. Evaluated across 12 open-book RAG QA benchmarks spanning diverse application domains and scenarios, PrismRAG improves average factuality by 5.4%, outperforming state-of-the-art solutions. Our method is being deployed in production."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/131733-prismrag-boosting-rag-factuality-with-distractor-resilience-and-strategized-reasoning","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2025-11-26T17:08:24Z","registered":"2025-11-26T17:08:24Z","published":null,"updated":"2026-10-17T10:01:50Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/ea7f-dv16","type":"dois","attributes":{"doi":"10.48448/ea7f-dv16","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2026","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Agency for Science, Technology and Research"}],"givenName":"Nancy F.","familyName":"Chen","name":"Chen, Nancy F.","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Hwaran","familyName":"Lee","name":"Lee, Hwaran","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Roy Ka-Wei","familyName":"Lee","name":"Lee, Roy Ka-Wei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Agency for Science, Technology and Research (A*STAR), Singpoare"}],"givenName":"Zhengyuan","familyName":"Liu","name":"Liu, Zhengyuan","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Singapore University of Technology and Design"}],"givenName":"Bryan","familyName":"Tan","name":"Tan, Bryan","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Kenny","familyName":"Tsu Wei Choo","name":"Tsu Wei Choo, Kenny","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Zheng","familyName":"Weihua","name":"Weihua, Zheng","nameIdentifiers":[]}],"titles":[{"title":"BLEnD-Vis: Benchmarking Multimodal Cultural Understanding in Vision Language Models"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2026,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Artificial Intelligence"}],"contributors":[],"dates":[{"date":"2026-03-06T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"As vision-language models (VLMs) are deployed globally, their ability to understand culturally situated knowledge becomes essential. Yet, existing evaluations largely assess static recall or isolated visual grounding, leaving unanswered whether VLMs possess robust and transferable cultural understanding. We introduce BLEnD-Vis, a multimodal, multicultural benchmark designed to evaluate the robustness of everyday cultural knowledge in VLMs across linguistic rephrasings and visual modalities. Building on the BLEnD dataset, BLEnD-Vis constructs 313 culturally grounded question templates spanning 16 regions and generates three aligned multiple-choice formats: (i) a text-only baseline querying from Region Entity, (ii) an inverted text-only variant (Entity Region), and (iii) a VQA-style version of (ii) with generated images. The resulting benchmark comprises 4,916 images and over 21,000 multiple-choice questions (MCQ) instances, validated through human annotation. BLEnD-Vis reveals significant fragility in current VLM cultural knowledge; models exhibit performance drops under linguistic rephrasing. While visual cues often aid performance, low cross-modal consistency highlights the challenges of robustly integrating textual and visual understanding, particularly in lower-resource regions. BLEnD-Vis thus provides a crucial testbed for systematically analysing cultural robustness and multimodal grounding, exposing limitations and guiding the development of more culturally competent VLMs. Code is available at https://github.com/Social-AI-Studio/BLEnD-Vis."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/147365-blend-vis-benchmarking-multimodal-cultural-understanding-in-vision-language-models","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-04-23T16:24:33Z","registered":"2026-04-23T16:24:33Z","published":null,"updated":"2026-10-17T10:01:50Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/sbas-jb17","type":"dois","attributes":{"doi":"10.48448/sbas-jb17","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2026","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Drexel University"}],"givenName":"Liancheng","familyName":"Gong","name":"Gong, Liancheng","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"University of Southern California"}],"givenName":"Jesse","familyName":"Thomason","name":"Thomason, Jesse","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Drexel University"}],"givenName":"Li","familyName":"Zhang","name":"Zhang, Li","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"University of Southern California"}],"givenName":"Wang","familyName":"Zhu","name":"Zhu, Wang","nameIdentifiers":[]}],"titles":[{"title":"Iterative Formalization and Planning in Partially Observable Environments"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2026,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Artificial Intelligence"},{"subject":"Software Intelligent Agents"}],"contributors":[],"dates":[{"date":"2026-06-03T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Using LLMs not to predict plans but to formalize an environment into the Planning Domain Definition Language (PDDL) has been shown to improve performance and control. While most existing methodology only applies to fully observable environments, we adapt to the more realistic and challenging partially observable environments without sufficient information to make a complete plan. We propose PDDLego+, a framework to iteratively formalize, plan, grow, and refine PDDL representations by decomposing the environment and the goal into fully observable episodes. Without fine‑tuning, in-context exemplars, or trajectories, PDDLego+ improves planning success and exhibits robustness against problem complexity compared to end-to-end approaches. We also show that the domain knowledge captured after a successful trial can benefit future tasks."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/152027-iterative-formalization-and-planning-in-partially-observable-environments","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-07-16T04:47:53Z","registered":"2026-07-16T04:47:53Z","published":null,"updated":"2026-10-17T10:01:50Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/2e3r-6115","type":"dois","attributes":{"doi":"10.48448/2e3r-6115","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2025","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Roy Ka-Wei","familyName":"Lee","name":"Lee, Roy Ka-Wei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Singapore University of Technology and Design"}],"givenName":"Bryan","familyName":"Tan","name":"Tan, Bryan","nameIdentifiers":[]}],"titles":[{"title":"Unmasking Implicit Bias: Evaluating Persona-Prompted LLM Responses in Power-Disparate Social Scenarios"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2025,"subjects":[{"subject":"Artificial Intelligence"},{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Philosophy and Ethics of AI"}],"contributors":[],"dates":[{"date":"2025-04-14T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Large language models (LLMs) have demonstrated remarkable capabilities in simulating human behaviour and social intelligence. However, they risk perpetuating societal biases, especially when demographic information is involved. We introduce a novel framework using cosine distance to measure semantic shifts in responses and an LLM-judged Preference Win Rate (WR) to assess how demographic prompts affect response quality across power-disparate social scenarios. Evaluating five LLMs over 100 diverse social scenarios and nine demographic axes, our findings suggest a ''default persona'' bias toward middle-aged, able-bodied, native-born, Caucasian, atheistic males with centrist views. Moreover, interactions involving specific demographics are associated with lower-quality responses. Lastly, the presence of power disparities increases variability in response semantics and quality across demographic groups, suggesting that implicit biases may be heightened under power-imbalanced conditions. These insights expose the demographic biases inherent in LLMs and offer potential paths toward future bias mitigation efforts in LLMs."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/117358-unmasking-implicit-bias-evaluating-persona-prompted-llm-responses-in-power-disparate-social-scenarios","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2025-08-26T14:27:51Z","registered":"2025-08-26T14:27:51Z","published":null,"updated":"2026-10-17T10:01:49Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/wgy2-zc08","type":"dois","attributes":{"doi":"10.48448/wgy2-zc08","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2026","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Bibo","familyName":"Cai","name":"Cai, Bibo","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Harbin Institute of Technology"}],"givenName":"Xiao","familyName":"Ding","name":"Ding, Xiao","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Sichuan University"}],"givenName":"Yuxiang","familyName":"He","name":"He, Yuxiang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Yutai","familyName":"Hou","name":"Hou, Yutai","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Yufei","familyName":"Liu","name":"Liu, Yufei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Ting","familyName":"Liu","name":"Liu, Ting","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Wu","familyName":"Ning","name":"Ning, Wu","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Haonan","familyName":"Song","name":"Song, Haonan","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Dandan","familyName":"Tu","name":"Tu, Dandan","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Yuxian","familyName":"Wang","name":"Wang, Yuxian","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"HIT"}],"givenName":"Yirong","familyName":"Zeng","name":"Zeng, Yirong","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Qixun","familyName":"Zhang","name":"Zhang, Qixun","nameIdentifiers":[]}],"titles":[{"title":"TinyJudge: Unverifiable Constraint Alignment via Lightweight Specialist Ensembles"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2026,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Artificial Intelligence"}],"contributors":[],"dates":[{"date":"2026-06-03T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Instruction Following (IF) is a core capability of LLMs, requiring strict adherence to diverse constraints, ranging from verifiable ones (e.g., output length) to unverifiable ones (e.g., tone). Reinforcement learning with verifiable rewards has emerged as a paradigm for IF tasks, leveraging LLM-as-a-judge to assess unverifiable constraints. However, we empirically find that this approach remains a significant bottleneck, suffering from severe reward hacking and higher computational overhead. In this work, we first analyze the generalization capabilities of unverifiable constraints and discover that specific constraints exhibit distinct, high-generalization patterns. Motivated by this, we propose TinyJudge, a framework that employs an ensemble of specialized tiny language models (e.g., 0.6B) to provide rewards for soft constraints. By distilling expertise from frontier models into these tiny models, it achieves high-precision, lightweight evaluation. Extensive evaluations across five benchmarks demonstrate that TinyJudge outperforms the baselines by ~10% in average performance and 12% in reward precision. Crucially, it also achieves a 3× speedup in total training time. Our work provides a scalable and robust path for aligning LLMs with unverifiable human instructions."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/149978-tinyjudge-unverifiable-constraint-alignment-via-lightweight-specialist-ensembles","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-07-16T00:07:29Z","registered":"2026-07-16T00:07:29Z","published":null,"updated":"2026-10-17T10:01:49Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/cqj3-yb45","type":"dois","attributes":{"doi":"10.48448/cqj3-yb45","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2024","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Nara Institute of Science and Technology"}],"givenName":"Hidetaka","familyName":"Kamigaito","name":"Kamigaito, Hidetaka","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Nara Institute of Science and Technology"}],"givenName":"Mana","familyName":"Makinae","name":"Makinae, Mana","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Nara Institute of Science and Technology"}],"givenName":"Yusuke","familyName":"Sakai","name":"Sakai, Yusuke","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Nara Institute of Science and Technology"}],"givenName":"Taro","familyName":"Watanabe","name":"Watanabe, Taro","nameIdentifiers":[]}],"titles":[{"title":"Simul-MuST-C: Simultaneous Multilingual Speech Translation Corpus Using Large Language Model"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2024,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Machine translation"}],"contributors":[],"dates":[{"date":"2024-10-28T19:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Simultaneous Speech Translation (SiST) begins translating before the entire source input is received, making it crucial to balance quality and latency. In real interpreting situations, interpreters manage this simultaneity by breaking sentences into smaller segments and translating them while maintaining the source order as much as possible. SiST could benefit from this approach to balance quality and latency. However, current corpora used for simultaneous tasks often involve significant word reordering in translation, which is not ideal given that interpreters faithfully follow source syntax as much as possible. Inspired by conference interpreting by humans utilizing the salami technique, we introduce the Simul-MuST-C, a dataset created by leveraging the Large Language Model (LLM), specifically GPT-4o, which aligns the target text as closely as possible to the source text by using minimal chunks that contain enough information to be interpreted. Experiments on three language pairs show that the effectiveness of segmented-base monotonicity in training data varies with the grammatical distance between the source and the target, with grammatically distant language pairs benefiting the most in achieving quality while minimizing latency."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/106334-simul-must-c-simultaneous-multilingual-speech-translation-corpus-using-large-language-model","contentUrl":[],"metadataVersion":8,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2024-11-29T13:50:51Z","registered":"2024-11-29T13:50:51Z","published":null,"updated":"2026-10-17T10:01:48Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.15468/dl.hcbkzw","type":"dois","attributes":{"doi":"10.15468/dl.hcbkzw","identifiers":[{"identifier":"0032033-260928105237408","identifierType":"GBIF"}],"creators":[{"nameType":"Organizational","name":"GBIF.org User","affiliation":[],"nameIdentifiers":[]}],"titles":[{"title":"Occurrence Download"}],"publisher":{"name":"The Global Biodiversity Information Facility"},"container":{},"publicationYear":2026,"subjects":[{"subject":"GBIF","lang":"eng"},{"subject":"biodiversity","lang":"eng"},{"subject":"species occurrences","lang":"eng"}],"contributors":[],"dates":[{"date":"2026-10-17","dateType":"Created"},{"date":"2026-10-17","dateType":"Updated"}],"language":null,"types":{"schemaOrg":"Dataset","resourceTypeGeneral":"Dataset","citeproc":"dataset","bibtex":"misc","ris":"DATA"},"relatedIdentifiers":[{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/9us6fb","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/ypoair","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/44r5e4","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/ly60bx","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/nc6rxy","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/o3pvnh","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/irnwew","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/kvfygp","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/pui83d","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/r2la8h","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/hja69f","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/7gudyo","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/hnhrg3","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/k64rgi","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/69ybth","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/liiipc","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/ib5ypt","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/bkzv1l","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/wrthhx","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/nl8bvi","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/wem0v1","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/dk5gko","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/tnj8wm","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/k5j0ep","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/iuxeth","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/gyvkrn","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/yo3mmu","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/7xc4hy","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/zdeuix","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/c664ha","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/7m584w","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/rxbp4w","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/o1q1w3","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/jb9tdf","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/0aqglb","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/dlwwhz","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/xtrjuf","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.5886/f6tdd5ap","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/vo3d2x","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/inygc6","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/tjvqci","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/t4lkyc","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/0bhhip","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/sft3wv","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/cchsv3","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/loyiz6","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/s5auru","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/a7ujmj","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.5519/qd.k6ev1cri","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/a3lnmd","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/wqcp7j","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/dkhtjv","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/dhouv6","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/7xnzpt","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/xvz2nj","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/npk1o5","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/yefl6d","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/uc1apo","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/ydqvcq","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.5886/uxx7hzy3","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/nxnqzf","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/u8bzai","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/ufmslw","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/5ensp7","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/wbjrjg","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/x5ucvh","relatedIdentifierType":"DOI"},{"relationType":"IsDerivedFrom","relatedIdentifier":"10.15468/rvjdu1","relatedIdentifierType":"DOI"},{"relationType":"HasMetadata","relatedIdentifier":"https://api.gbif.org/v1/occurrence/download/0032033-260928105237408","relatedIdentifierType":"URL"},{"relationType":"HasMetadata","relatedIdentifier":"https://api.gbif.org/v1/occurrence/download/0032033-260928105237408/datasets","relatedIdentifierType":"URL"}],"relatedItems":[],"sizes":["627294"],"formats":["Darwin Core Archive","text/tab-separated-values","application/zip"],"version":null,"rightsList":[{"rightsIdentifierScheme":"SPDX","rightsUri":"https://creativecommons.org/licenses/by-nc/4.0/legalcode","schemeUri":"https://spdx.org/licenses/","rights":"Creative Commons Attribution Non Commercial 4.0 International","rightsIdentifier":"cc-by-nc-4.0"}],"descriptions":[{"descriptionType":"Abstract","description":"A dataset containing 3135 species occurrences available in GBIF matching the query:\n{\n \"TaxonKey\" : [\n \"is one of (Anemone rivularis [Catalogue of Life])\"\n ]\n}\n\n The dataset includes 3135 records from 67 constituent datasets; see https://api.gbif.org/v1/occurrence/download/0032033-260928105237408/datasets/export for details.\n \nData from some individual datasets included in this download may be licensed under less restrictive terms.","lang":"eng"}],"geoLocations":[],"fundingReferences":[],"url":"https://www.gbif.org/occurrence/download/0032033-260928105237408","contentUrl":null,"metadataVersion":0,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-10-17T10:01:46Z","registered":"2026-10-17T10:01:47Z","published":null,"updated":"2026-10-17T10:01:47Z"},"relationships":{"client":{"data":{"id":"gbif.gbif","type":"clients"}}}},{"id":"10.48448/kfrc-wj44","type":"dois","attributes":{"doi":"10.48448/kfrc-wj44","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2025","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"University of Virginia"}],"givenName":"Zihan","familyName":"Chen","name":"Chen, Zihan","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Florida State University"}],"givenName":"Yushun","familyName":"Dong","name":"Dong, Yushun","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Xi'an Jiaotong University"}],"givenName":"Zhenyu","familyName":"Lei","name":"Lei, Zhenyu","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"University Of Virginia"}],"givenName":"Jundong","familyName":"Li","name":"Li, Jundong","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Arizona State University"}],"givenName":"Zhen","familyName":"Tan","name":"Tan, Zhen","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"University Of Virginia"}],"givenName":"Song","familyName":"Wang","name":"Wang, Song","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Yaochen","familyName":"Zhu","name":"Zhu, Yaochen","nameIdentifiers":[]}],"titles":[{"title":"Learning from Diverse Reasoning Paths with Routing and Collaboration"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2025,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Artificial Intelligence"},{"subject":"Machine Learning"}],"contributors":[],"dates":[{"date":"2025-10-10T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Advances in large language models (LLMs) significantly enhance reasoning capabilities but their deployment is restricted in resource-constrained scenarios. Knowledge distillation addresses this by transferring knowledge from powerful teacher models to compact and transparent students. However, effectively capturing the teacher's comprehensive reasoning is challenging due to conventional token-level supervision's limited scope. Using multiple reasoning paths per query alleviates this problem, but treating each path identically is suboptimal as paths vary widely in quality and suitability across tasks and models. We propose Quality-filtered Routing with Cooperative Distillation (QR-Distill), combining path quality filtering, conditional routing, and cooperative peer teaching. First, quality filtering retains only correct reasoning paths scored by an LLM-based evaluation. Second, conditional routing dynamically assigns paths tailored to each student's current learning state. Finally, cooperative peer teaching enables students to mutually distill diverse insights, addressing knowledge gaps and biases toward specific reasoning styles. Experiments demonstrate QR-Distill's superiority over traditional single- and multi-path distillation methods. Ablation studies further highlight the importance of each component—quality filtering, conditional routing, and peer teaching—in effective knowledge transfer."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/129893-learning-from-diverse-reasoning-paths-with-routing-and-collaboration","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2025-11-26T15:27:04Z","registered":"2025-11-26T15:27:05Z","published":null,"updated":"2026-10-17T10:01:47Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/qatv-7a70","type":"dois","attributes":{"doi":"10.48448/qatv-7a70","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Artificial Intelligence 2025","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Chen","familyName":"Chen","name":"Chen, Chen","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Florida State University"}],"givenName":"Yushun","familyName":"Dong","name":"Dong, Yushun","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Xi'an Jiaotong University"}],"givenName":"Zhenyu","familyName":"Lei","name":"Lei, Zhenyu","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"University Of Virginia"}],"givenName":"Jundong","familyName":"Li","name":"Li, Jundong","nameIdentifiers":[]}],"titles":[{"title":"ST-FiT: Inductive Spatial-Temporal Forecasting with Limited Training Data"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2025,"subjects":[{"subject":"Artificial Intelligence"},{"subject":"Data Mining"}],"contributors":[],"dates":[{"date":"2025-02-02T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[{"relationType":"References","resourceTypeGeneral":"Text","relatedIdentifier":"https://doi.org/10.1609/aaai.v39i11.33310","relatedIdentifierType":"DOI"}],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Spatial-temporal graphs are widely used in a variety of real-world applications. Spatial-Temporal Graph Neural Networks (STGNNs) have emerged as a powerful tool to extract meaningful insights from this data. However, in real-world applications, most nodes may not possess any available temporal data during training. For example, the pandemic dynamics of most cities on a geographical graph may not be available due to the asynchronous nature of outbreaks. Such a phenomenon disagrees with the training requirements of most existing spatial-temporal forecasting methods, which jeopardizes their effectiveness and thus blocks broader deployment. In this paper, we propose to formulate a novel problem of inductive forecasting with limited training data. In particular, given a spatial-temporal graph, we aim to learn a spatial-temporal forecasting model that can be easily generalized onto those nodes without any available temporal training data. To handle this problem, we propose a principled framework named ST-FiT. ST-FiT consists of two key learning components: temporal data augmentation and spatial graph topology learning. With such a design, ST-FiT can be used on top of any existing STGNNs to achieve superior performance on the nodes without training data. Extensive experiments verify the effectiveness of ST-FiT in multiple key perspectives."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/110815-st-fit-inductive-spatial-temporal-forecasting-with-limited-training-data","contentUrl":[],"metadataVersion":3,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2025-03-24T13:35:14Z","registered":"2025-03-24T13:35:14Z","published":null,"updated":"2026-10-17T10:01:47Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.25405/ncl.34641537","type":"dois","attributes":{"doi":"10.25405/ncl.34641537","identifiers":[],"creators":[{"givenName":"S","familyName":"Coulthard","name":"S Coulthard","affiliation":[],"nameIdentifiers":[]},{"givenName":"C","familyName":"Howell","name":"C Howell","affiliation":[],"nameIdentifiers":[]},{"givenName":"J","familyName":"Robson","name":"J Robson","affiliation":[],"nameIdentifiers":[]},{"affiliation":[{"affiliationIdentifier":"https://ror.org/01kj2bm70","schemeUri":"https://ror.org/","name":"Newcastle University","affiliationIdentifierScheme":"ROR"}],"givenName":"Andrew","familyName":"Hall","name":"Andrew Hall","nameIdentifiers":[]}],"titles":[{"title":"Thiopurine methyltransferase genotype and phenotype in normal subjects and patients with acute leukaemia"}],"publisher":{"name":"Newcastle University"},"container":{},"publicationYear":2026,"subjects":[],"contributors":[],"dates":[{"date":"2026-10-17","dateType":"Created"},{"date":"2026-10-17","dateType":"Updated"},{"date":"1998-01-01","dateType":"Issued"}],"language":null,"types":{"schemaOrg":"ScholarlyArticle","resourceTypeGeneral":"JournalArticle","citeproc":"article-journal","bibtex":"article","ris":"JOUR","resourceType":"Journal contribution"},"relatedIdentifiers":[],"relatedItems":[],"sizes":["0 Bytes"],"formats":[],"version":null,"rightsList":[{"rightsUri":"https://rightsstatements.org/page/InC/1.0/?language=en","rights":"Copyrighted"}],"descriptions":[{"descriptionType":"Abstract","description":"No description provided"}],"geoLocations":[],"fundingReferences":[],"url":"https://outputs.ncl.ac.uk/articles/journal_contribution/Thiopurine_methyltransferase_genotype_and_phenotype_in_normal_subjects_and_patients_with_acute_leukaemia/34641537","contentUrl":null,"metadataVersion":0,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"mds","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-10-17T10:01:45Z","registered":"2026-10-17T10:01:46Z","published":null,"updated":"2026-10-17T10:01:47Z"},"relationships":{"client":{"data":{"id":"figshare.ncl","type":"clients"}}}},{"id":"10.48448/akc6-6k23","type":"dois","attributes":{"doi":"10.48448/akc6-6k23","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2026","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Guiyang","familyName":"Hou","name":"Hou, Guiyang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Yiwei","familyName":"Jiang","name":"Jiang, Yiwei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Xin","familyName":"Li","name":"Li, Xin","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Peng","familyName":"Li","name":"Li, Peng","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Gangao","familyName":"Liu","name":"Liu, Gangao","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Jiajun","familyName":"Liu","name":"Liu, Jiajun","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Weiming","familyName":"Lu","name":"Lu, Weiming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yongliang","familyName":"Shen","name":"Shen, Yongliang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Mengna","familyName":"Wang","name":"Wang, Mengna","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Huixin","familyName":"Xu","name":"Xu, Huixin","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Hang","familyName":"Zhang","name":"Zhang, Hang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Wenqi","familyName":"Zhang","name":"Zhang, Wenqi","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Zhe","familyName":"Zheng","name":"Zheng, Zhe","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yueting","familyName":"Zhuang","name":"Zhuang, Yueting","nameIdentifiers":[]}],"titles":[{"title":"Embodied-Reasoner: Synergizing Visual Search, Reasoning, and Action for Embodied Interactive Tasks"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2026,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Artificial Intelligence"},{"subject":"Information and Knowledge Engineering"}],"contributors":[],"dates":[{"date":"2026-06-03T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Recent advances in reasoning models have demonstrated remarkable capabilities on mathematical and coding tasks. However, their effectiveness in embodied domains, where the agent must continuously interact with environments and process observation-action interleaved trajectories, remains largely unexplored. We present Embodied-Reasoner, a reasoning model for interactive embodied tasks. Unlike mathematical reasoning that relies primarily on logical deduction, embodied scenarios demand spatial understanding, temporal reasoning, and ongoing self-reflection based on interaction history. To address these challenges, we synthesize 9.3k coherent Observation-Thought-Action trajectories containing 64k ego-centric images and 90k diverse reasoning processes (analysis, spatial reasoning, reflection, planning, and verification). We develop a three-stage training recipe that progressively enhances the model's capabilities through imitation learning, rejection sampling tuning on self-exploration trajectories, and reflection tuning. The evaluation shows that our model significantly outperforms advanced visual reasoning models, e.g., exceeds OpenAI o1, o3-mini, and Claude-3.7 by +9%, 24%, and +13%. Analysis reveals that our model exhibits fewer repeated searches and logical inconsistencies, with particular advantages in complex long-horizon tasks. Real-world testing further validates the effectiveness of our approach."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/150708-embodied-reasoner-synergizing-visual-search-reasoning-and-action-for-embodied-interactive-tasks","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-07-16T02:07:21Z","registered":"2026-07-16T02:07:21Z","published":null,"updated":"2026-10-17T10:01:46Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/jj6j-bc78","type":"dois","attributes":{"doi":"10.48448/jj6j-bc78","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2026","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Guangyi","familyName":"Liu","name":"Liu, Guangyi","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Zhengxi","familyName":"Lu","name":"Lu, Zhengxi","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Weiming","familyName":"Lu","name":"Lu, Weiming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yongliang","familyName":"Shen","name":"Shen, Yongliang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Kaitao","familyName":"Song","name":"Song, Kaitao","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Xu","familyName":"Tan","name":"Tan, Xu","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Fei","familyName":"Tang","name":"Tang, Fei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Daniell","familyName":"Wang","name":"Wang, Daniell","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Jun","familyName":"Xiao","name":"Xiao, Jun","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Wenqi","familyName":"Zhang","name":"Zhang, Wenqi","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yueting","familyName":"Zhuang","name":"Zhuang, Yueting","nameIdentifiers":[]}],"titles":[{"title":"UI-Copilot: Advancing Long-Horizon GUI Automation via Tool-Integrated Policy Optimization"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2026,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Artificial Intelligence"},{"subject":"Software Intelligent Agents"}],"contributors":[],"dates":[{"date":"2026-06-03T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"MLLM-based GUI agents have demonstrated strong capabilities in complex user interface interaction tasks. However, long-horizon scenarios remain challenging, as these agents are burdened with tasks beyond their intrinsic capabilities, suffering from memory degradation, progress confusion, and math hallucination. To address these challenges, we present $\\textbf{UI-Copilot}$, a collaborative framework where the GUI agent focuses on task execution while a lightweight copilot provides on-demand assistance for memory retrieval and numerical computation. We introduce memory decoupling to separate persistent observations from transient execution context, and train the policy agent to selectively invoke the copilot as $\\texttt{Retriever}$ or $\\texttt{Calculator}$ based on task demands. To enable effective tool invocation learning, we propose $\\underline{\\textbf{T}}$ool-$\\underline{\\textbf{I}}$ntegrated $\\underline{\\textbf{P}}$olicy $\\underline{\\textbf{O}}$ptimization ($\\textbf{TIPO}$), which separately optimizes tool selection through single-turn prediction and task execution through on-policy multi-turn rollouts. Experimental results show that UI-Copilot-7B achieves state-of-the-art performance on challenging MemGUI-Bench, outperforming strong 7B-scale GUI agents such as GUI-Owl-7B and UI-TARS-1.5-7B. Moreover, UI-Copilot-7B delivers a 17.1% absolute improvement on AndroidWorld over the base Qwen model, highlighting UI-Copilot's strong generalization to real-world GUI tasks. Code website: https://anonymous.4open.science/r/UI-Copilot-0535."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/149672-ui-copilot-advancing-long-horizon-gui-automation-via-tool-integrated-policy-optimization","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-07-15T23:16:58Z","registered":"2026-07-15T23:16:58Z","published":null,"updated":"2026-10-17T10:01:45Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.5281/zenodo.23434742","type":"dois","attributes":{"doi":"10.5281/zenodo.23434742","identifiers":[],"creators":[{"nameType":"Personal","givenName":"Jincheng","familyName":"Zhang","name":"Zhang, Jincheng","nameIdentifiers":[{"nameIdentifierScheme":"ORCID","nameIdentifier":"0009-0005-1860-0009"}],"affiliation":[]}],"titles":[{"title":"Adaptive Switching Between Self-Mechanisms, Mutual Mechanisms, and Goal Mechanisms in Multiple Large Language Models: A Goal-Directed Framework for Dynamic Intelligence Coordination"}],"publisher":{"name":"Zenodo"},"container":{},"publicationYear":2026,"subjects":[],"contributors":[],"dates":[{"date":"2026-10-17","dateType":"Issued"}],"language":null,"types":{"schemaOrg":"CreativeWork","resourceTypeGeneral":"Preprint","citeproc":"article","bibtex":"misc","ris":"GEN","resourceType":""},"relatedIdentifiers":[{"relationType":"HasVersion","relatedIdentifier":"10.5281/zenodo.23434743","relatedIdentifierType":"DOI"}],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[{"rightsIdentifierScheme":"SPDX","rightsUri":"https://creativecommons.org/licenses/by/4.0/legalcode","schemeUri":"https://spdx.org/licenses/","rights":"Creative Commons Attribution 4.0 International","rightsIdentifier":"cc-by-4.0"}],"descriptions":[{"descriptionType":"Abstract","description":"Large language models (LLMs) can improve task performance through individual processing mechanisms, including self-reflection, self-evaluation, self-correction, self-verification, and self-planning. Multiple LLMs can also interact through mutual evaluation, collaborative planning, cross-model verification, mutual correction, and distributed reasoning. However, neither individual processing nor multi-model collaboration is necessarily sufficient to ensure that a system consistently pursues its intended objectives. A system may produce locally convincing responses while deviating from the task goal, optimize an easily measured intermediate objective at the expense of the final outcome, or continue generating and revising outputs without making meaningful progress.\n\n \n\nThis paper proposes Adaptive Switching between Self-Mechanisms, Mutual Mechanisms, and Goal Mechanisms (ASSM-MM-G), a conceptual framework that integrates individual-model processing, cross-model interaction, and explicit goal-oriented control. The framework represents goals as operational structures containing desired outcomes, constraints, priorities, intermediate milestones, evaluation criteria, and termination conditions. A goal-aware controller dynamically selects among individual processing, multi-model collaboration, goal analysis, goal decomposition, goal evaluation, goal revision, and goal-progress monitoring. Its decisions depend on the current task state, estimated goal satisfaction, uncertainty, model disagreement, expected improvement, and computational cost. The framework distinguishes changes in processing mechanisms from changes in the objectives those mechanisms serve, allowing goal-directed behavior to remain stable while processing strategies adapt. It also permits goal revision when requirements change, evidence invalidates assumptions, or objectives conflict, subject to explicit constraints on permissible goal changes. This paper defines the framework, formalizes its decision problem, presents a reference algorithm, identifies failure modes, and proposes research questions, hypotheses, and an experimental protocol. The framework is a research proposal rather than an empirically validated system. Its potential contribution is the investigation of whether goal representation and goal-state feedback can improve adaptive mechanism selection beyond systems that select individual or collaborative processing mechanisms without an explicit, dynamically evaluated goal structure."}],"geoLocations":[],"fundingReferences":[],"url":"https://zenodo.org/doi/10.5281/zenodo.23434742","contentUrl":null,"metadataVersion":0,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":1,"versionOfCount":0,"created":"2026-10-17T10:01:45Z","registered":null,"published":null,"updated":"2026-10-17T10:01:45Z"},"relationships":{"client":{"data":{"id":"cern.zenodo","type":"clients"}}}},{"id":"10.48448/etvv-0b63","type":"dois","attributes":{"doi":"10.48448/etvv-0b63","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2026","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Fei","familyName":"Huang","name":"Huang, Fei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Weiming","familyName":"Lu","name":"Lu, Weiming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Zhengxi","familyName":"Lu","name":"Lu, Zhengxi","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yongliang","familyName":"Shen","name":"Shen, Yongliang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Fei","familyName":"Tang","name":"Tang, Fei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Jun","familyName":"Xiao","name":"Xiao, Jun","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Alibaba Cloud"}],"givenName":"Haiyang","familyName":"Xu","name":"Xu, Haiyang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Alibaba Group"}],"givenName":"Ming","familyName":"Yan","name":"Yan, Ming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"East China Normal University, Alibaba Cloud"}],"givenName":"Jiabo","familyName":"Ye","name":"Ye, Jiabo","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Ziwei","familyName":"Zheng","name":"Zheng, Ziwei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yueting","familyName":"Zhuang","name":"Zhuang, Yueting","nameIdentifiers":[]}],"titles":[{"title":"Experience-driven Multi-turn Reinforcement Learning for GUI Agents"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2026,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Artificial Intelligence"},{"subject":"Software Intelligent Agents"}],"contributors":[],"dates":[{"date":"2026-06-03T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"GUI agents have demonstrated remarkable progress in automating complex user interface interactions. However, training such agents for long-horizon tasks remains challenging. Single-turn reinforcement learning conditions on expert histories during training but self-generated histories during deployment, causing distribution mismatch. Online multi-turn methods eliminate this gap via environment interaction but suffer from sparse rewards and prohibitive costs. We propose $\\underline{\\textbf{E}}$xperience-driven $\\underline{\\textbf{M}}$ulti-turn $\\underline{\\textbf{P}}$olicy $\\underline{\\textbf{O}}$ptimization ($\\textbf{EMPO}$), which leverages expert trajectories as environment experiences for on-policy multi-turn training. The agent constructs self-generated history throughout rollouts; when actions match expert experiences, the trajectory provides valid state transitions, and a Patch Module recovers mismatched steps to maintain on-policy rollouts. EMPO further incorporates discounted future rewards and dual-level advantage estimation to capture long-horizon dependencies. We also propose $\\textbf{AndroidControl-Real}$, an evaluation metric strongly correlated with real-world performance (R$^2$=0.934). With only 1K public trajectories as RL experiences, our method achieves substantial gains over the base model (e.g., +12.0% on AndroidWorld and +23.8% on AITW) and achieves competitive performance against strong baselines such as UI-TARS-7B and GPT-4o, demonstrating better generalization than prior single-turn RL approaches. Code available: \\url{https://anonymous.4open.science/r/UI-S1-0DAF}."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/149184-experience-driven-multi-turn-reinforcement-learning-for-gui-agents","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-07-15T21:57:17Z","registered":"2026-07-15T21:57:17Z","published":null,"updated":"2026-10-17T10:01:45Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.5281/zenodo.23434743","type":"dois","attributes":{"doi":"10.5281/zenodo.23434743","identifiers":[{"identifier":"oai:zenodo.org:23434743","identifierType":"oai"}],"creators":[{"nameType":"Personal","givenName":"Jincheng","familyName":"Zhang","name":"Zhang, Jincheng","nameIdentifiers":[{"nameIdentifierScheme":"ORCID","nameIdentifier":"0009-0005-1860-0009"}],"affiliation":[]}],"titles":[{"title":"Adaptive Switching Between Self-Mechanisms, Mutual Mechanisms, and Goal Mechanisms in Multiple Large Language Models: A Goal-Directed Framework for Dynamic Intelligence Coordination"}],"publisher":{"name":"Zenodo"},"container":{},"publicationYear":2026,"subjects":[],"contributors":[],"dates":[{"date":"2026-10-17","dateType":"Issued"}],"language":null,"types":{"schemaOrg":"CreativeWork","resourceTypeGeneral":"Preprint","citeproc":"article","bibtex":"misc","ris":"GEN","resourceType":""},"relatedIdentifiers":[{"relationType":"IsVersionOf","relatedIdentifier":"10.5281/zenodo.23434742","relatedIdentifierType":"DOI"}],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[{"rightsIdentifierScheme":"SPDX","rightsUri":"https://creativecommons.org/licenses/by/4.0/legalcode","schemeUri":"https://spdx.org/licenses/","rights":"Creative Commons Attribution 4.0 International","rightsIdentifier":"cc-by-4.0"}],"descriptions":[{"descriptionType":"Abstract","description":"Large language models (LLMs) can improve task performance through individual processing mechanisms, including self-reflection, self-evaluation, self-correction, self-verification, and self-planning. Multiple LLMs can also interact through mutual evaluation, collaborative planning, cross-model verification, mutual correction, and distributed reasoning. However, neither individual processing nor multi-model collaboration is necessarily sufficient to ensure that a system consistently pursues its intended objectives. A system may produce locally convincing responses while deviating from the task goal, optimize an easily measured intermediate objective at the expense of the final outcome, or continue generating and revising outputs without making meaningful progress.\n\n \n\nThis paper proposes Adaptive Switching between Self-Mechanisms, Mutual Mechanisms, and Goal Mechanisms (ASSM-MM-G), a conceptual framework that integrates individual-model processing, cross-model interaction, and explicit goal-oriented control. The framework represents goals as operational structures containing desired outcomes, constraints, priorities, intermediate milestones, evaluation criteria, and termination conditions. A goal-aware controller dynamically selects among individual processing, multi-model collaboration, goal analysis, goal decomposition, goal evaluation, goal revision, and goal-progress monitoring. Its decisions depend on the current task state, estimated goal satisfaction, uncertainty, model disagreement, expected improvement, and computational cost. The framework distinguishes changes in processing mechanisms from changes in the objectives those mechanisms serve, allowing goal-directed behavior to remain stable while processing strategies adapt. It also permits goal revision when requirements change, evidence invalidates assumptions, or objectives conflict, subject to explicit constraints on permissible goal changes. This paper defines the framework, formalizes its decision problem, presents a reference algorithm, identifies failure modes, and proposes research questions, hypotheses, and an experimental protocol. The framework is a research proposal rather than an empirically validated system. Its potential contribution is the investigation of whether goal representation and goal-state feedback can improve adaptive mechanism selection beyond systems that select individual or collaborative processing mechanisms without an explicit, dynamically evaluated goal structure."}],"geoLocations":[],"fundingReferences":[],"url":"https://zenodo.org/doi/10.5281/zenodo.23434743","contentUrl":null,"metadataVersion":0,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":1,"created":"2026-10-17T10:01:44Z","registered":"2026-10-17T10:01:45Z","published":null,"updated":"2026-10-17T10:01:45Z"},"relationships":{"client":{"data":{"id":"cern.zenodo","type":"clients"}}}},{"id":"10.48448/90ye-q982","type":"dois","attributes":{"doi":"10.48448/90ye-q982","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2026","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Weiming","familyName":"Lu","name":"Lu, Weiming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yongliang","familyName":"Shen","name":"Shen, Yongliang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Xing","familyName":"Yue","name":"Yue, Xing","nameIdentifiers":[]}],"titles":[{"title":"\\texttt{Phun-Bench}: Evaluating LLMs on Phonological Understanding in Chinese"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2026,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Artificial Intelligence"},{"subject":"Semantic Evaluation"}],"contributors":[],"dates":[{"date":"2026-06-03T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Language is a vehicle for thought, intricately tied to sounds, symbols, and meaning. However, most large language model (LLM) research focuses on meaning (semantics) and symbols (spelling) while largely overlooking sounds. Existing benchmarks on LLMs' phonological abilities are either solvable through rote memorization or intertwined with other abilities, making them inadequate to measure LLMs' genuine ability in phonological understanding. Here, we present $\\texttt{Phun-Bench}$, a purpose-built Chinese benchmark with diverse tasks and settings across three dimensions (Homophony, Rhyme, and Phonetic Similarity), designed to systematically evaluate LLMs' phonological understanding. Our results show that while LLMs excel at recalling correct pronunciations, they generally struggle to leverage phonological knowledge in the flexible and intuitive way that human speakers do. Moreover, through detailed analyses, we propose a hypothesis regarding the underlying mechanism of LLMs' phonological understanding and ``perception'', highlighting an underexplored frontier for future research."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/149813-dashtextttphun-bench-evaluating-llms-on-phonological-understanding-in-chinese","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-07-15T23:40:31Z","registered":"2026-07-15T23:40:31Z","published":null,"updated":"2026-10-17T10:01:45Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/e1vp-8v22","type":"dois","attributes":{"doi":"10.48448/e1vp-8v22","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2026","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Haiwen","familyName":"Hong","name":"Hong, Haiwen","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Longtao","familyName":"Huang","name":"Huang, Longtao","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Hongxing","familyName":"Li","name":"Li, Hongxing","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Weiming","familyName":"Lu","name":"Lu, Weiming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yongliang","familyName":"Shen","name":"Shen, Yongliang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Haolei","familyName":"Xu","name":"Xu, Haolei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Hui","familyName":"Xue'","name":"Xue', Hui","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Yang","familyName":"Zhang","name":"Zhang, Yang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Rui","familyName":"Zhou","name":"Zhou, Rui","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yueting","familyName":"Zhuang","name":"Zhuang, Yueting","nameIdentifiers":[]}],"titles":[{"title":"Seeing but Not Thinking: Routing Distraction in Multimodal Mixture-of-Experts"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2026,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Artificial Intelligence"},{"subject":"Information and Knowledge Engineering"}],"contributors":[],"dates":[{"date":"2026-06-03T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Multimodal Mixture-of-Experts (MoE) models have achieved remarkable performance on vision-language tasks. However, we identify a puzzling phenomenon termed Seeing but Not Thinking: models accurately perceive image content yet fail in subsequent reasoning, while correctly solving identical problems presented as pure text. Through systematic analysis, we first verify that cross-modal semantic sharing exists in MoE architectures, ruling out semantic alignment failure as the sole explanation. We then reveal that visual experts and domain experts exhibit layer-wise separation, with image inputs inducing significant routing divergence from text inputs in middle layers where domain experts concentrate. Based on these findings, we propose the Routing Distraction hypothesis: when processing visual inputs, the routing mechanism fails to adequately activate task-relevant reasoning experts. To validate this hypothesis, we design a routing-guided intervention method that enhances domain expert activation. Experiments on three multimodal MoE models across six benchmarks demonstrate consistent improvements, with gains of up to 3.17% on complex visual reasoning tasks. Our analysis further reveals that domain expert identification locates cognitive functions rather than sample-specific solutions, enabling effective transfer across tasks with different information structures."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/150220-seeing-but-not-thinking-routing-distraction-in-multimodal-mixture-of-experts","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-07-16T00:47:30Z","registered":"2026-07-16T00:47:30Z","published":null,"updated":"2026-10-17T10:01:45Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.25405/ncl.34641534","type":"dois","attributes":{"doi":"10.25405/ncl.34641534","identifiers":[],"creators":[{"givenName":"R","familyName":"Lowry","name":"R Lowry","affiliation":[],"nameIdentifiers":[]},{"givenName":"J","familyName":"Steele","name":"J Steele","affiliation":[],"nameIdentifiers":[]}],"titles":[{"title":"Adult dental treatment costs greatly reduced by water fluoridation"}],"publisher":{"name":"Newcastle University"},"container":{},"publicationYear":2026,"subjects":[],"contributors":[],"dates":[{"date":"2026-10-17","dateType":"Created"},{"date":"2026-10-17","dateType":"Updated"},{"date":"2003-12-31","dateType":"Issued"}],"language":null,"types":{"schemaOrg":"Periodical","resourceTypeGeneral":"ConferenceProceeding","citeproc":"article","bibtex":"misc","ris":"CONF","resourceType":"Conference contribution"},"relatedIdentifiers":[],"relatedItems":[],"sizes":["0 Bytes"],"formats":[],"version":null,"rightsList":[{"rightsUri":"https://rightsstatements.org/page/InC/1.0/?language=en","rights":"Copyrighted"}],"descriptions":[{"descriptionType":"Abstract","description":"In this preliminary study, we examined National Health Service (NHS) treatment spending in primary care using routinely-collected data. The Dental Practice Board, the United Kingdom NHS primary dental care administrative organisation, analysed treatment data for 1984 and 1985. Treatments for damage arising mainly from dental caries (fillings, crowns and extractions) were selected. Only treatments to permanent teeth counted. The study was limited to 21-29 year olds living in two fluoridated (Birmingham and Newcastle on Tyne) and two non-fluoridated areas (Wolverhampton and Cleveland). The cohorts were chosen deliberately because these age groups were related to the onset of fluoridation in their respective areas: Birmingham started water fluoridation in 1964, Newcastle in 1969, so in 1985 both cohorts were born into non-fluoridated areas, but of the 1995 cohorts, the Birmingham residents had been born in a fluoridated area whereas only half the Newcastle ones were. For both years of the study, all four areas would have benefited from improvements in dental health due to better diet, fluoridated toothpaste etcetera Both non-fluoridated areas showed a similar drop in total caries-related treatments between 1984 and 1994, but Birmingham showed a significantly increased drop, with Newcastle in between. We costed up the treatment items at today’s NHS prices and can calculate the decline in gross fees (£1 sterling)."}],"geoLocations":[],"fundingReferences":[],"url":"https://outputs.ncl.ac.uk/articles/conference_contribution/Adult_dental_treatment_costs_greatly_reduced_by_water_fluoridation/34641534","contentUrl":null,"metadataVersion":0,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"mds","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-10-17T10:01:43Z","registered":"2026-10-17T10:01:44Z","published":null,"updated":"2026-10-17T10:01:44Z"},"relationships":{"client":{"data":{"id":"figshare.ncl","type":"clients"}}}},{"id":"10.48448/tamk-fe86","type":"dois","attributes":{"doi":"10.48448/tamk-fe86","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2026","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Shengpei","familyName":"Jiang","name":"Jiang, Shengpei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Hao","familyName":"Li","name":"Li, Hao","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Hengwei","familyName":"Liu","name":"Liu, Hengwei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Weiming","familyName":"Lu","name":"Lu, Weiming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Qingqing","familyName":"Lyu","name":"Lyu, Qingqing","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yongliang","familyName":"Shen","name":"Shen, Yongliang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Linjuan","familyName":"Wu","name":"Wu, Linjuan","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Yin","familyName":"Zhang","name":"Zhang, Yin","nameIdentifiers":[]}],"titles":[{"title":"AutoTaskEval: Towards Domain-Specific and Fine-Grained Evaluation for LLMs"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2026,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Artificial Intelligence"},{"subject":"Semantic Evaluation"}],"contributors":[],"dates":[{"date":"2026-06-03T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Despite the rapid progress of LLMs, their evaluation remains hindered by static, manually curated benchmarks with limited task coverage and poor adaptability to emerging domains. Existing automated approaches typically operate within fixed task schemas and often fail to autonomously discover new evaluation dimensions, limiting both scalability and effectiveness. To address these gaps, we propose AutoTaskEval, an automated framework that constructs domain-specific benchmarks directly from unstructured corpora. Using a refined Bloom’s Taxonomy, the framework systematically discovers tasks, enriches contextual grounding via iterative Socratic prompting, and generates diverse, progressively challenging evaluation instances. Applied to the complex and knowledge-intensive legal domain, AutoTaskEval uncovers a broader and more fine-grained task space than expert-curated benchmarks while producing high-quality instances that preserve established model-level evaluation trends. We further validate its robustness in a low-structure e-commerce review domain. Together, these results show that AutoTaskEval enables scalable, adaptive, and high-fidelity LLM assessment across domains and model families, advancing autonomous and capability-sensitive evaluation."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/149030-autotaskeval-towards-domain-specific-and-fine-grained-evaluation-for-llms","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-07-15T21:31:47Z","registered":"2026-07-15T21:31:47Z","published":null,"updated":"2026-10-17T10:01:44Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/bhaz-dd06","type":"dois","attributes":{"doi":"10.48448/bhaz-dd06","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2024","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Mingqian","familyName":"He","name":"He, Mingqian","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Weiming","familyName":"Lu","name":"Lu, Weiming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yongliang","familyName":"Shen","name":"Shen, Yongliang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Zeqi","familyName":"Tan","name":"Tan, Zeqi","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Wenqi","familyName":"Zhang","name":"Zhang, Wenqi","nameIdentifiers":[]}],"titles":[{"title":"Advancing Process Verification for Large Language Models via Tree-Based Preference Learning"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2024,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"}],"contributors":[],"dates":[{"date":"2024-10-28T19:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Large Language Models (LLMs) have demonstrated remarkable potential in handling complex reasoning tasks by generating step-by-step rationales. Some methods have proven effective in boosting accuracy by introducing extra verifiers to assess these paths. However, existing verifiers, typically trained on binary-labeled reasoning paths, fail to fully utilize the relative merits of intermediate steps, thereby limiting the effectiveness of the feedback provided. To overcome this limitation, we propose Tree-based Preference Learning Verifier (Tree-PLV), a novel approach that constructs reasoning trees via a best-first search algorithm and collects step-level paired data for preference training. Compared to traditional binary classification, step-level preferences more finely capture the nuances between reasoning steps, allowing for a more precise evaluation of the complete reasoning path. We empirically evaluate Tree-PLV across a range of arithmetic and commonsense reasoning tasks, where it significantly outperforms existing benchmarks. For instance, Tree-PLV achieved substantial performance gains over the Mistral-7B self-consistency baseline on GSM8K (67.55% → 82.79%), MATH (17.00% → 26.80%), CSQA (68.14% → 72.97%), and StrategyQA (82.86% → 83.25%). Additionally, our study explores the appropriate granularity for applying preference learning, revealing that step-level guidance provides feedback that better aligns with the evaluation of the reasoning process."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/108922-advancing-process-verification-for-large-language-models-via-tree-based-preference-learning","contentUrl":[],"metadataVersion":1,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2024-11-29T17:55:24Z","registered":"2024-11-29T17:55:24Z","published":null,"updated":"2026-10-17T10:01:43Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/1dmh-w455","type":"dois","attributes":{"doi":"10.48448/1dmh-w455","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2023","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Microsoft Research Asia"}],"givenName":"Dongsheng","familyName":"Li","name":"Li, Dongsheng","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Weiming","familyName":"Lu","name":"Lu, Weiming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yongliang","familyName":"Shen","name":"Shen, Yongliang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Microsoft Research"}],"givenName":"Kaitao","familyName":"Song","name":"Song, Kaitao","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Xu","familyName":"Tan","name":"Tan, Xu","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yueting","familyName":"Zhuang","name":"Zhuang, Yueting","nameIdentifiers":[]}],"titles":[{"title":"DiffusionNER: Boundary Diffusion for Named Entity Recognition"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2022,"subjects":[{"subject":"Natural Language Processing"},{"subject":"Multilingualism"},{"subject":"Machine Learning"}],"contributors":[],"dates":[{"date":"2022-07-05T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"In this paper, we propose DiffusionNER, which formulates the named entity recognition task as a boundary-denoising diffusion process and thus generates named entities from noisy spans. During training, DiffusionNER gradually adds noises to the golden entity boundaries by a fixed forward diffusion process and learns a reverse diffusion process to recover the entity boundaries. In inference, DiffusionNER first randomly samples some noisy spans from a standard Gaussian distribution and then generates the named entities by denoising them with the learned reverse diffusion process. The proposed boundary-denoising diffusion process allows progressive refinement and dynamic sampling of entities, empowering DiffusionNER with efficient and flexible entity generation capability. Experiments on multiple flat and nested NER datasets demonstrate that DiffusionNER achieves comparable or even better performance than previous state-of-the-art models. \u003ciframe src=\"https://app.sli.do/event/nDU9hzdsVZXUCPdEbR3oSF/embed/polls/90c440ca-29db-4182-8e70-03f53d253409\" width=\"300\" height=\"400\"\u003e\u003c/iframe\u003e"}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/76567-diffusionner-boundary-diffusion-for-named-entity-recognition","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2023-09-20T13:24:07Z","registered":"2023-09-20T13:24:07Z","published":null,"updated":"2026-10-17T10:01:42Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/gkw5-1167","type":"dois","attributes":{"doi":"10.48448/gkw5-1167","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2026","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Xunliang","familyName":"Cai","name":"Cai, Xunliang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Maoxin","familyName":"He","name":"He, Maoxin","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Feiye","familyName":"Huo","name":"Huo, Feiye","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Weile","familyName":"Jia","name":"Jia, Weile","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Meituan"}],"givenName":"Jiacheng","familyName":"Li","name":"Li, Jiacheng","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Jiayu","familyName":"Qin","name":"Qin, Jiayu","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Pingwei","familyName":"Sun","name":"Sun, Pingwei","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Guangming","familyName":"Tan","name":"Tan, Guangming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Jianchao","familyName":"Tan","name":"Tan, Jianchao","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Zhidong","familyName":"Yang","name":"Yang, Zhidong","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Xiangyu","familyName":"Zhang","name":"Zhang, Xiangyu","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Tong","familyName":"Zhao","name":"Zhao, Tong","nameIdentifiers":[]}],"titles":[{"title":"WISCA: A Lightweight Model Transition Method to Improve LLM Training via Weight Scaling"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2026,"subjects":[{"subject":"Computational Linguistics"},{"subject":"Natural Language Processing"},{"subject":"Artificial Intelligence"},{"subject":"Machine Learning"}],"contributors":[],"dates":[{"date":"2026-06-03T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Transformer architecture gradually dominates the LLM field. Recent advances in training optimization for Transformer-based large language models (LLMs) primarily focus on architectural modifications or optimizer adjustments. However, these approaches lack systematic optimization of weight patterns during training. Weight pattern refers to the distribution and relative magnitudes of weight parameters in a neural network. To address this issue, we propose a Weight Scaling method called WISCA to enhance training efficiency and model quality by strategically improving neural network weight patterns—without changing network structures. By rescaling weights while preserving model outputs, WISCA indirectly optimizes the model’s training trajectory. Experiments demonstrate that WISCA significantly improves convergence quality (measured by generalization capability and loss reduction), particularly in LLMs with Grouped Query Attention (GQA) architectures and LoRA fine-tuning tasks. Empirical results show 5.6% average improvement on zero-shot validation tasks and 2.12% average reduction in training perplexity across multiple architectures."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/151486-wisca-a-lightweight-model-transition-method-to-improve-llm-training-via-weight-scaling","contentUrl":[],"metadataVersion":2,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2026-07-16T04:08:02Z","registered":"2026-07-16T04:08:03Z","published":null,"updated":"2026-10-17T10:01:41Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/4rym-d993","type":"dois","attributes":{"doi":"10.48448/4rym-d993","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2022","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Weiming","familyName":"Lu","name":"Lu, Weiming","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yongliang","familyName":"Shen","name":"Shen, Yongliang","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Zeqi","familyName":"Tan","name":"Tan, Zeqi","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Yiquan","familyName":"Wu","name":"Wu, Yiquan","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Zhejiang University"}],"givenName":"Shuai","familyName":"Zhang","name":"Zhang, Shuai","nameIdentifiers":[]}],"titles":[{"title":"De-Bias for Generative Extraction in Unified NER Task"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2022,"subjects":[{"subject":"Language Models"},{"subject":"Natural Language Processing"}],"contributors":[],"dates":[{"date":"2022-05-11T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"Named entity recognition (NER) is a fundamental task to recognize specific types of entities from a given sentence. Depending on how the entities appear in the sentence, it can be divided into three subtasks, namely, Flat NER, Nested NER, and Discontinuous NER. Among the existing approaches, only the generative model can be uniformly adapted to these three subtasks. However, when the generative model is applied to NER, its optimization objective is not consistent with the task, which makes the model vulnerable to the incorrect biases. In this paper, we analyze the incorrect biases in the generation process from a causality perspective and attribute them to two confounders: pre-context confounder and entity-order confounder. Furthermore, we design Intra- and Inter-entity Deconfounding Data Augmentation methods to eliminate the above confounders according to the theory of backdoor adjustment. Experiments show that our method can improve the performance of the generative NER model in various datasets."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/50344-de-bias-for-generative-extraction-in-unified-ner-task","contentUrl":[],"metadataVersion":1,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2022-08-16T07:27:55Z","registered":"2022-08-16T07:27:55Z","published":null,"updated":"2026-10-17T10:01:41Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/429h-4h02","type":"dois","attributes":{"doi":"10.48448/429h-4h02","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Artificial Intelligence 2021","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"Neal","familyName":"Ravindra","name":"Ravindra, Neal","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Kensho Technologies"}],"givenName":"Arijit","familyName":"Sehanobish","name":"Sehanobish, Arijit","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[],"givenName":"David","familyName":"van Dijk","name":"van Dijk, David","nameIdentifiers":[]}],"titles":[{"title":"Gaining Insight into SARS-CoV-2 Infection and COVID-19 Severity Using Self-Supervised Edge Features and Graph Neural Networks"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2021,"subjects":[{"subject":"Artificial Intelligence"}],"contributors":[],"dates":[{"date":"2021-02-02T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":"A molecular and cellular understanding of how SARS-CoV-2 variably infects and causes severe COVID-19 remains a bottleneck in developing interventions to end the pandemic. We sought to use deep learning (DL) to study the biology of SARS-CoV-2 infection and COVID-19 severity by identifying transcriptomic patterns and cell types associated with SARS-CoV-2 infection and COVID-19 severity. To do this, we developed a new approach to generating self-supervised edge features. We propose a model that builds on Graph Attention Networks (GAT), creates edge features using self-supervised learning, and ingests these edge features via a Set Transformer. This model achieves significant improvements in predicting the disease state of individual cells, given their transcriptome. We apply our model to single-cell RNA sequencing datasets of SARS-CoV-2 infected lung organoids and bronchoalveolar lavage fluid samples of patients with COVID-19, achieving state-of-the-art performance on both datasets with our model. We then borrow from the field of explainable AI (XAI) to identify the features (genes) and cell types that discriminate bystander vs. infected cells across time and moderate vs. severe COVID-19 disease. To the best of our knowledge, this represents the first application of DL to identifying the molecular and cellular determinants of SARS-CoV-2 infection and COVID-19 severity using single-cell omics data."}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/73490-gaining-insight-into-sars-cov-2-infection-and-covid-19-severity-using-self-supervised-edge-features-and-graph-neural-networks","contentUrl":[],"metadataVersion":3,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2025-02-10T13:49:02Z","registered":"2025-02-10T13:49:02Z","published":null,"updated":"2026-10-17T10:01:40Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}},{"id":"10.48448/4wb2-2j30","type":"dois","attributes":{"doi":"10.48448/4wb2-2j30","identifiers":[],"creators":[{"nameType":"Organizational","name":"Association for Computational Linguistics 2022","affiliation":[],"nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Covera Health"}],"givenName":"Nabila","familyName":"Abraham","name":"Abraham, Nabila","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Covera Health"}],"givenName":"Murray","familyName":"Becker","name":"Becker, Murray","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Covera Health"}],"givenName":"Anasuya","familyName":"Das","name":"Das, Anasuya","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Covera Health"}],"givenName":"Richard","familyName":"Herzog","name":"Herzog, Richard","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Covera Health"}],"givenName":"Benjamin","familyName":"Odry","name":"Odry, Benjamin","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Covera Health"}],"givenName":"Jayashri","familyName":"Pawar","name":"Pawar, Jayashri","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Covera Health"}],"givenName":"McCullen","familyName":"Sandora","name":"Sandora, McCullen","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Kensho Technologies"}],"givenName":"Arijit","familyName":"Sehanobish","name":"Sehanobish, Arijit","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Covera Health"}],"givenName":"Danielle","familyName":"Torres","name":"Torres, Danielle","nameIdentifiers":[]},{"nameType":"Personal","affiliation":[{"name":"Covera Health"}],"givenName":"Ron","familyName":"Vianu","name":"Vianu, Ron","nameIdentifiers":[]}],"titles":[{"title":"Explaining the Effectiveness of Multi-Task Learning for Efficient Knowledge Extraction from Spine MRI Reports"}],"publisher":{"name":"Underline Science Inc."},"container":{},"publicationYear":2022,"subjects":[{"subject":"Machine Learning"},{"subject":"Natural Language Processing"}],"contributors":[],"dates":[{"date":"2022-06-27T14:00:00Z","dateType":"Available"}],"language":null,"types":{"schemaOrg":"MediaObject","resourceTypeGeneral":"Audiovisual","citeproc":"article","bibtex":"misc","ris":"MPCT","resourceType":"Conference talk"},"relatedIdentifiers":[],"relatedItems":[],"sizes":[],"formats":[],"version":null,"rightsList":[],"descriptions":[{"descriptionType":"Abstract","description":""}],"geoLocations":[],"fundingReferences":[],"url":"https://underline.io/lecture/54353-explaining-the-effectiveness-of-multi-task-learning-for-efficient-knowledge-extraction-from-spine-mri-reports","contentUrl":[],"metadataVersion":1,"schemaVersion":"http://datacite.org/schema/kernel-4","source":"api","isActive":true,"state":"findable","reason":null,"viewCount":0,"downloadCount":0,"referenceCount":0,"citationCount":0,"partCount":0,"partOfCount":0,"versionCount":0,"versionOfCount":0,"created":"2022-08-26T08:49:07Z","registered":"2022-08-26T08:49:07Z","published":null,"updated":"2026-10-17T10:01:40Z"},"relationships":{"client":{"data":{"id":"undsci.oopmau","type":"clients"}}}}],"meta":{"total":139482316,"totalPages":400,"page":1},"links":{"self":"https://api.datacite.org/dois?affiliation=true\u0026publisher=true\u0026state=findable","next":"https://api.datacite.org/dois?affiliation=true\u0026page%5Bnumber%5D=2\u0026page%5Bsize%5D=25\u0026publisher=true"}}