News

🎉 Our paper SEISMOS: A Statistical Signal Detection Framework for Semantic Chunking is accepted to the NeurIPS 2026 main conference (poster)!

Serving as a reviewer for ICLR 2027.

Invited to serve as a reviewer for three NeurIPS 2026 workshops: AI4MetaScience, AIWILD, and SLM-Agents.

Three of our papers on geo-spatial and geo-temporal reasoning, crisis sentiment analysis, and clinical NLP robustness were accepted to AACL-IJCNLP 2026 (2 Main, 1 Findings)—one as first author, one as second author, and one as a co-author.

Invited to serve as a reviewer for WACV 2027.

Two papers on formal and structured computational reasoning are accepted to Findings of EMNLP 2026, one as first author and one as co-author.

Attending ICML 2026 at COEX in Seoul, South Korea, to present our main-track paper, TimeSpot, along with several workshop papers.

Invited to serve as a reviewer for the NeurIPS 2026 Creative AI Track.

Three papers are accepted to ICML 2026 workshops: two at the AI for Math Workshop (AI4Math) and one at Muslims in ML (MusIML). One AI4Math paper also received a reviewer-level Best Paper nomination recommendation!

Serving as a reviewer for the AI for Math Workshop at ICML 2026.

When Machines Decide If a Human Wrote It: Creativity in the Age of AI Detectors is accepted to the ICML 2026 Workshop on Human-AI Co-Creativity.

TimeSpot accepted to ICML 2026 as co-first author

Our paper TimeSpot: Benchmarking Geo-Temporal Understanding in Vision-Language Models in Real-World Settings has been accepted to the ICML 2026 main track. I contributed to this work as a co-first author, in collaboration with QCRI and CIOL.

TimeSpot asks whether a model can infer where and when an image was taken from visual cues alone. It comprises 1,455 ground-level images from 80 countries, with models predicting temporal attributes (season, month, time of day, daylight phase) and geographic attributes (continent, country, climate zone, environment type, latitude-longitude coordinates), plus spatial reasoning tasks that integrate these cues.

State-of-the-art open- and closed-source VLMs perform consistently poorly across tasks, underscoring the need for better methods for reliable geo-temporal reasoning.

Read the paper

BengaliMoralBench, the first large-scale ethics benchmark for the Bengali language and culture, is accepted to ACM FAccT 2026, in collaboration with CIOL and Hanyang University.

SpatiaLab accepted to ICLR 2026

Our paper SpatiaLab: Can Vision-Language Models Perform Spatial Reasoning in the Wild? has been accepted to ICLR 2026, in collaboration with QCRI, Monash University, and CIOL.

SpatiaLab is a diagnostic benchmark of 1,400 visual question-answer pairs spanning six categories of spatial reasoning: relative positioning, depth and occlusion, orientation, size and scale, spatial navigation, and 3D geometry. Unlike synthetic or templated datasets, it is built from realistic, unconstrained scenes that resist shortcut cues.

Our evaluation shows that modern VLMs handle surface-level spatial perception reasonably well but consistently struggle with multi-step and relational spatial reasoning. The best model reaches 54.9% accuracy in the multiple-choice setting against 87.6% for humans, revealing a substantial gap between visual recognition and grounded spatial understanding.

Read the paper

Our position paper on robust clinical NLP datasets, Messy Clinic, is accepted to the EurIPS 2025 Workshop on Multimodal Representation Learning for Healthcare (MMRL4H).

CAPSTONE, our composable scene-to-text framework for zero-shot vision-language reasoning, is accepted to the EMNLP 2025 Industry Track.

Serving as a reviewer for the ACL 2025 Student Research Workshop (SRW).

Our paper on context-aware data cleaning for Bengali text is accepted at SN Computer Science (Springer Nature).

Joined the Department of Electrical and Computer Engineering at North South University as a Lab Instructor for the Data Structures and Algorithms and Programming Language labs.

EnStack, an ensemble-stacking framework of code LLMs for vulnerability detection, is published in the proceedings of IEEE BigData 2024.

Received the Best Paper Award at the IEEE International Conference on Intelligent Systems (IS) 2024, Varna, Bulgaria, for our work on efficient clinical-text preprocessing for transformer encoders.