Random Image

Sourajit Saha

PhD Candidate, Computer Science at University of Maryland, Baltimore County

Visiting Graduate Scholar (Summer 2026), Johns Hopkins University

Email: ssaha2@umbc.edu Location: ITE 338, UMBC, Baltimore, MD 21250

Interactive Visual Search | Multimodal Retrieval | Multimodal Reasoning

Video Understanding | Vision and Language | RL Post-Training for Multimodal Systems

Resume CV

Highlights
ACL 2026 Oral
First-author, ACL 2026 main track; selected for oral (top 15% of papers)
WACV 2025 Oral
First-author, Translation Invariant Polyphase Sampling
US Patent Pending
App. 64/045,450 — Zero-Shot Multimodal Retrieval
197 Citations
h-index 7 · Google Scholar
Great Reviewer Award, ACL 2026 (top 1.4%) Outstanding Reviewer, CVPR 2026 (top 5%) DARPA SciFy Research Assistant JHU HLTCOE SCALE 2024 Intern NSF SRP Fellow 2026 Lambda Research Grant 2025 Session Chair, ACL 2026
Bio

I am a Computer Science PhD candidate (expected 2028), working under the guidance of Tejas Gokhale in the UMBC Cognitive Vision Group at University of Maryland, Baltimore County (UMBC). I am currently a Visiting Graduate Scholar at Johns Hopkins University (Summer 2026), working on multi-set video retrieval-augmented generation, and a Research Assistant on the DARPA SciFy program. I work on interactive multimodal retrieval/search, multimodal reasoning, and video understanding. My research spans the following areas:

♦ Interactive Multimodal Retrieval, Search: Enhancing few-shot and zero-shot video search and retrieval, Developing Scene Graph-based Chain-of-Thought retrieval frameworks, Designing dialogue-driven interactive retrieval systems, Robust Ranking Evaluation.
♦ Video Understanding: Information Theory guided Video Question Answering (VQA), Progressive Video Captioning, Semantic Frame Selection, Video Editing.
♦ Visual Reasoning: Understanding spatial relationships and transformations, Studying counterfactual visual reasoning, Exploring techniques for image and model editing.
♦ Post-Training and RL for Multimodal Systems: Reward-driven post-training for multimodal retrieval, ranking, and reasoning; RL for retrieval policies, adaptive search strategies, and test-time decision-making from user feedback and task-specific retrieval signals.

Experience
Visiting Graduate Scholar Jun 2026 – Present Johns Hopkins University — Department of Computer Science

Multi-set video retrieval-augmented generation (RAG); extending zero-shot multimodal retrieval to long-form, multi-source video corpora.

Research Assistant, DARPA SciFy Program Jan 2025 – Present UMBC — mentor: Tejas Gokhale

First-authored ACL 2026 (Oral) paper on zero-shot multimodal retrieval with multi-scale contextual representations; filed as US Patent App. 64/045,450.

Research Intern, SCALE 2024 Jun – Aug 2024 Johns Hopkins University — HLTCOE, Core Computer Vision Team

Large-scale multilingual, event-based video retrieval.

Graduate Research Assistant Jan – Dec 2022 UMBC — mentors: David Chapman, Milton Halem

Unsupervised domain adaptation for TB screening (IEEE Access 2022); meta-learning for COVID-19 chest X-ray classification (SPIE MI 2022).

Research Assistant Jul 2018 – Jul 2020 Center for Cognitive Skill Enhancement, Independent University Bangladesh

Deep learning for retinal-disease diagnosis from OCT (ICMLA 2019); built a neural-net training-scheduling platform for the CCSE supercomputer.

News scroll for older news ↓
Publications

Most recent publications on Google Scholar — 197 citations, h-index 7.

Zero-Shot Multimodal Retrieval with Multi-Scale Contextual Representations. Sourajit Saha, Tejas Gokhale

ACL 2026 (Oral + Main, top 15%) · US Patent App. 64/045,450 (pending) paper project code

Side Effects of Erasing Concepts from Diffusion Models. Shaswati Saha, Sourajit Saha, Manas Gaur, Tejas Gokhale

EMNLP 2025 (Findings) paper code

Improving Shift Invariance in Convolutional Neural Networks with Translation Invariant Polyphase Sampling. Sourajit Saha, Tejas Gokhale

WACV 2025 (Oral) paper video poster code

RFC-Net: Learning High Resolution Global Features for Medical Image Segmentation on a Computational Budget (Student Abstract). Sourajit Saha, Shaswati Saha, Md Osman Gani, Tim Oates, David Chapman

AAAI 2023 paper code

Mitigating Domain Shift in AI-Based TB Screening With Unsupervised Domain Adaptation. Nishanjan Ravin, Sourajit Saha, Alan Schweitzer, Ameena Elahi, Farouk Dako, Daniel Mollura, David Chapman

IEEE Access paper code

Pairwise Meta Learning Pipeline: Classifying COVID-19 abnormalities on chest radio-graphs. Sourajit Saha, Yaacov Yesha, Yelena Yesha, Aryya Gangopadhyay, David Chapman, Michael Morris, Babak Saboury, Phuong Nguyen

SPIE Medical Imaging Conference 2022 Paper

A comprehensive set of novel residual blocks for deep learning architectures for diagnosis of retinal diseases from optical coherence tomography images. Sharif Amit Kamran, Sourajit Saha, Ali Shihab Sabbir, Alireza Tavakkoli

Springer Book Series, 2020 paper code

Optic-Net: A Novel Convolutional Neural Network for Diagnosis of Retinal Diseases from Optical Tomography Images. Sharif Amit Kamran, Sourajit Saha, Ali Shihab Sabbir, Alireza Tavakkoli

ICMLA 2019 paper code

Awards, Grants & Honors
Academic Service
Collaborators
Acknowledgement

Website theme inspirations: Aniruddha Saha, Martin Saveski, Aditi Partap.

Sourajit Saha
Last updated 07/18/2026