নমস্কার — I am Soujanya Poria (সৌজন্য পড়িয়া)

School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore

My group studies how AI can reason, explain itself, and act safely across language, vision, and audio. I'm an Associate Professor at EEE, NTU, leading the DeCLaRe Lab. Before NTU, I was at SUTD.

Research Focus

My research focuses on language and multimodal intelligence, with emphasis on systems that reason, explain, align with human values, and operate robustly across modalities. My recent work spans LLM evaluation, safety and alignment, multimodal reasoning, text-to-audio generation, embodied AI, and affective computing.

Open to Collaboration

Prospective students and collaborators should first review the lab research slides and current publications before reaching out.

Research slides
Recent Updates All publications

Award Highly Cited Researcher 2026

Recognized by Web of Science.

AAAI 2026 Two papers accepted

ICLR 2026 Four papers accepted

ICML 2026 Two papers accepted

2025 Joined EEE, NTU

Moved to the School of Electrical and Electronic Engineering at NTU.

ACL 2025 Four papers accepted

EMNLP 2025 Five papers accepted

Research Group

DeCLaRe Lab

Much of this work grows out of the DeCLaRe Lab, where we study dependable, communicative, and reasoning-capable AI systems across language, vision, audio, and embodied interaction. The group brings together students and collaborators working on multimodal learning, affective computing, trustworthy LLMs, and agents that can act with better grounding.

Selected Publications

A small set of representative papers, grouped by research direction.

Multimodal AI

Representation Learning & Fusion

MISA was one of the early works to use attention for fusing multimodalities.

Affective Computing

Emotion, Sentiment & Dialogue

MELD is used as a benchmark in Qwen-Audio, Qwen2-Audio, Audio-Flamingo, SALMONN, MiniCPM-o-4.5, and Phi-4.

Generative AI

Text-to-Audio Generation

Evaluation

LLM Evaluation & Reasoning

Efficient Foundation Models

Data-Efficient ML, Attention & Memory

Responsible AI

AI Safety, Alignment & Trustworthiness

Knowledge-Intensive AI

Retrieval-Augmented Generation

Embodied AI

Agents, Planning & Robotics