Soujanya Poria
Associate Professor
Works on large language models, reasoning, affective computing, embodied AI and AI safety, leading the DeCLaRe lab.
Soujanya Poria's lab at NTU works on large language models, reasoning & agents and trustworthy & safe ML. 74 papers since 2023, 39 at top venues, mostly ACL and EMNLP. 12 PhD students. 20 alumni, most often to Apple and 1X Technologies.
At a glance
- papers, 2023–2026
- 19/ year
- at T1 venues
- 53%
- PhD students
- 12
- MS students
- 0
- alumni listed
- 20
Papers by yearfrom the lab homepage, 2023–2026
- 13 papers2026 · T1 10 · collecting
- 21 papers2025 · T1 11
- 23 papers2024 · T1 12
- 17 papers2023 · T1 6
T1 venuesOtherVenues: ACL, EMNLP, ICLR, ACM MM, Findings of EMNLP
Research focuspaper titles that use each topic's words
- Large language models33 papers
- Reasoning & agents9 papers
- Trustworthy & safe ML0 papers
- leading0 papers
- affective computing0 papers
Paper list
From the lab homepage.
202613 papers
- AAAI (T1)10 Open Challenges Steering the Future of Vision-Language-Action Models (opens in a new tab)
- ICML (T1)Chain-of-Glimpse: Search-Guided Progressive Object-Grounded Reasoning for Video Understanding (opens in a new tab)
- ICML (T1)Data Agent: Learning to Select Data via End-to-End Dynamic Optimization (opens in a new tab)
- ICLR (T1)Demystifying deep search: a holistic evaluation with hint-free multi-hop questions and factorised metrics (opens in a new tab)
- AAAI (T1)DialogXpert: Driving Intelligent and Emotion-Aware Conversations through Online Value-Based Reinforcement Learning with LLM Priors (opens in a new tab)
- NeurIPS (T1)Exact Flow Linear Attention: Exact Solution from Continuous-Time Dynamics (opens in a new tab)
- Cognitive ComputationLLM Alignment should go beyond Harmlessness–Helpfulness and incorporate Human Agency (opens in a new tab)
- ICLR (T1)Measuring and Mitigating Rapport Bias of Large Language Models under Multi-Agent Social Interactions (opens in a new tab)
- AI for Math Workshop @ ICMLNot All Tokens Are Equal: Intrinsic Loss Reweighting for Verifiable Reinforcement Learning (opens in a new tab)
- ICLR (T1)OffTopicEval: When Large Language Models Enter the Wrong Chat, Almost Always! (opens in a new tab)
- ICLR (T1)Tangoflux: Super fast and faithful text to audio generation with flow matching and clap-ranked preference optimization (opens in a new tab)
- Logical and Symbolic Reasoning in Language Models @ AAAITracking the Evolution of Multimodal Reasoning on Visual Puzzles (opens in a new tab)
202521 papers
- Information FusionAction-guided prompt tuning for video grounding (opens in a new tab)
- NAACL (T1)Are Language Models Puzzle Prodigies? Algorithmic Puzzles Unveil Serious Challenges in Multimodal Reasoning (opens in a new tab)
- ACL (T1)DiffPO: Diffusion-styled Preference Optimization for Inference Time Alignment of Large Language Models (opens in a new tab)
- ACL (T1)Emma-x: An embodied multimodal action model with grounded chain of thought and look-ahead spatial reasoning (opens in a new tab)
- Findings of EMNLPError Typing for Smarter Rewards: Improving Process Reward Models with Error-Aware Hierarchical Supervision (opens in a new tab)
- EMNLP (T1)Evaluating AI for Finance: Is AI Credible at Assessing Investment Risk Appetite? (opens in a new tab)
- Findings of ACLEvaluating LLMs' Mathematical and Coding Competency through Ontology-guided Interventions (opens in a new tab)
- Findings of EMNLPFerret: Faster and effective automated red teaming with reward-based scoring technique (opens in a new tab)
- Findings of EMNLPFrom grounding to manipulation: Case studies of foundation model integration in embodied robotic systems (opens in a new tab)
- NAACL (T1)Libra-leaderboard: Towards responsible AI through a balanced leaderboard of safety and capability (opens in a new tab)
- EMNLP (T1)M-longdoc: A benchmark for multimodal super-long document understanding and a retrieval-aware tuning framework (opens in a new tab)
- ICLR (T1)Measuring and Enhancing Trustworthiness of LLMs in RAG through Grounded Attributions and Learning to Refuse (opens in a new tab)
202423 papers
- EACLA Comprehensive Survey of Sentence Representations: From the BERT Epoch to the ChatGPT Era and Beyond (opens in a new tab)
- ICLR (T1)Chain-of-Knowledge: Grounding large language models via dynamic knowledge adapting over heterogeneous sources (opens in a new tab)
- Findings of NAACLCM-TTS: Enhancing Real Time Text-to-Speech Synthesis Efficiency through Weighted Samplers and Consistency Models (opens in a new tab)
- WWW (T1)Consistency Guided Knowledge Retrieval and Denoising in LLMs for Zero-shot Document-level Relation Triplet Extraction (opens in a new tab)
- Workshop on Social Influence in ConversationsDomain-expanded ASTE: Rethinking generalization in aspect sentiment triplet extraction (opens in a new tab)
- Expert SystemsHate speech detection: A comprehensive review of recent works (opens in a new tab)
- COLING (T1)HyperTTS: Parameter Efficient Adaptation in Text to Speech using Hypernetworks (opens in a new tab)
- Workshop on the Scaling Behavior of Large Language ModelsInstructEval: Towards holistic evaluation of instruction-tuned large language models (opens in a new tab)
- ACL (T1)Language Models are Homer Simpson! Safety Re-Alignment of Fine-tuned Language Models through Task Arithmetic (opens in a new tab)
- ACL (T1)Large language models for automated open-domain scientific hypotheses discovery (opens in a new tab)
- EACLLightweight Spatial Modeling for Combinatorial Information Extraction From Documents (opens in a new tab)
- NAACL (T1)Mustango: Toward controllable text-to-music generation (opens in a new tab)
202317 papers
- Information FusionA Review of Deep Learning Techniques for Speech Processing (opens in a new tab)
- Findings of ACLA Robust Information-Masking Approach for Domain Counterfactual Generation (opens in a new tab)
- Findings of EMNLPAdapter Pruning using Tropical Characterization (opens in a new tab)
- Interspeech (T1)ADAPTERMIX: Exploring the Efficacy of Mixture of Adapters for Low-Resource TTS Adaptation (opens in a new tab)
- Cognitive ComputationDialogue relation extraction with document-level heterogeneous graph attention networks (opens in a new tab)
- ICASSP (T1)Evaluating parameter-efficient transfer learning approaches on SURE benchmark for speech understanding (opens in a new tab)
- Findings of ACLFew-shot joint multimodal aspect-sentiment analysis based on generative multimodal prompt (opens in a new tab)
- ACM MMFew-shot multimodal sentiment analysis based on multimodal probabilistic fusion prompts (opens in a new tab)
- Findings of EMNLPkNN-CM: A Non-parametric Inference-Phase Adaptation of Parametric Text Classifiers (opens in a new tab)
- EMNLP (T1)Language guided visual question answering: Elevate your multimodal language model using knowledge-enriched prompts (opens in a new tab)
- EMNLP (T1)LLM-Adapters: An Adapter Family for Parameter-Efficient Fine-Tuning of Large Language Models (opens in a new tab)
- Information FusionMultimodal sentiment analysis: A systematic review of history, datasets, multimodal fusion methods, applications, challenges and future directions (opens in a new tab)
Where alumni went20 listed, 56% to industry
Industry 9Academia 7
Most often to
- Apple
- 1X Technologies
- Alibaba
Also to
- NUS
- Ca’ Foscari University of Venice
- China Academy of Information and Communications Technology
Apply to NTUFall 2027
Not announced yet
Computing and Data Science PhD
- Round
- Main
- Opens
- —
- GRE
- not required
- English
- TOEFL/IELTS
- Fee
- See program page
Not yet confirmed for this intake.
Similar labsnearest by research summary
- USUW
Jeffrey A. Bilmes
- Learning theory & optimization
- USMIT
Paul Liang
21 T1 - CNZhejiang
Zhixuan Chu
- Large language models
- USNYU
Kyunghyun Cho
- Large language models
- USUCLA
Nanyun Peng
- Large language models
- USUSC
Jieyu Zhao
- Interpretability, evaluation & safety of LMs
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.