Ideas, made public

Publications

Published and accepted research, from generalization and geometry to language and financial AI.

Conferences & journals

WACV 2027 Accepted

HElIOS: Hyperbolic Generalized Category Discovery for Remote Sensing

Sadia Hussain, Vaibhav Rathore, Divyam Gupta, Saurabh Prasad, Aleksandra Pizurica, Biplab Banerjee

Vaibhav's affiliation: IIT Bombay

Summary & reference for HElIOS: Hyperbolic Generalized Category Discovery for Remote Sensing

Introduces a hyperbolic framework for discovering known and novel categories in remote-sensing imagery.

Sadia Hussain, Vaibhav Rathore, Divyam Gupta, Saurabh Prasad, Aleksandra Pizurica, Biplab Banerjee. “HElIOS: Hyperbolic Generalized Category Discovery for Remote Sensing.” WACV 2027. Accepted.

Interspeech 2026 Accepted

SuTRA: Structurally-Unified Tokenization with Root Awareness

Vaibhav Rathore, Siddhant Gole, Dadhichi Telwadkar, Rooshil Bhatia, Maulik Ruparel, Siddharth Surekha, Neha Bhargava

Summary & reference for SuTRA: Structurally-Unified Tokenization with Root Awareness

A morphology-aware tokenizer that preserves Indic akshara structure and reduces morphological shattering across Hindi, Marathi, and Gujarati.

Vaibhav Rathore, Siddhant Gole, Dadhichi Telwadkar, Rooshil Bhatia, Maulik Ruparel, Siddharth Surekha, Neha Bhargava. “SuTRA: Structurally-Unified Tokenization with Root Awareness.” Interspeech 2026. Accepted.

ACM Transactions on Computing for Healthcare Published

M³ QuestionIng: Multi-modal Multi-span Medical Question Answering

Anisha Saha, Vaibhav Rathore, Abhisek Tiwari, Akash Ghosh, Sai Ruthvik Edara, Sriparna Saha

Summary & reference for M³ QuestionIng: Multi-modal Multi-span Medical Question Answering

Introduces M³QAFrame and a medical question-answering dataset whose answers draw jointly from multiple textual and visual spans.

Anisha Saha, Vaibhav Rathore, Abhisek Tiwari, Akash Ghosh, Sai Ruthvik Edara, Sriparna Saha. “M³ QuestionIng: Multi-modal Multi-span Medical Question Answering.” ACM Transactions on Computing for Healthcare. Published.

NeurIPS 2025 Published

HiDISC: A Hyperbolic Framework for Domain Generalization with Generalized Category Discovery

Vaibhav Rathore, Divyam Gupta, Biplab Banerjee

Summary & reference for HiDISC: A Hyperbolic Framework for Domain Generalization with Generalized Category Discovery

A hyperbolic framework for cross-domain alignment and novel-category discovery without expensive episodic simulation.

Vaibhav Rathore, Divyam Gupta, Biplab Banerjee. “HiDISC: A Hyperbolic Framework for Domain Generalization with Generalized Category Discovery.” NeurIPS 2025. Published.

CVPR 2025 Published

When Domain Generalization Meets Generalized Category Discovery: An Adaptive Task-Arithmetic Driven Approach

Vaibhav Rathore, Shubhranil B, Saikat Dutta, Sarthak Mehrotra, Zsolt Kira, Biplab Banerjee

Summary & reference for When Domain Generalization Meets Generalized Category Discovery: An Adaptive Task-Arithmetic Driven Approach

Introduces DG²CD-Net and the DG-GCD setting, combining episodic adaptation with adaptive task arithmetic.

Vaibhav Rathore, Shubhranil B, Saikat Dutta, Sarthak Mehrotra, Zsolt Kira, Biplab Banerjee. “When Domain Generalization Meets Generalized Category Discovery: An Adaptive Task-Arithmetic Driven Approach.” CVPR 2025. Published.

Workshops & posters

Acceptance is distinct from publication or presentation. AI4F 2026 acceptances are for non-archival posters, not proceedings papers.

AI4F Workshop at ACM ICAIF 2026 Accepted poster Non-archival workshop

Align Before You Quantize: Making Leech Lattice Tokenization Work for Time Series

Akanksh Adugani Vanjari, Vaibhav Rathore, Siddhant Gole, Siddharth Sureka, Neha Bhargava

Vaibhav's affiliation: Motilal Oswal Financial Services

Summary & reference for Align Before You Quantize: Making Leech Lattice Tokenization Work for Time Series

Uses hyperspherical alignment and phased training to improve lattice-codebook utilization and reconstruction for time-series tokenization.

Akanksh Adugani Vanjari, Vaibhav Rathore, Siddhant Gole, Siddharth Sureka, Neha Bhargava. “Align Before You Quantize: Making Leech Lattice Tokenization Work for Time Series.” AI4F Workshop at ACM ICAIF 2026. Accepted poster; non-archival workshop.

AI4F Workshop at ACM ICAIF 2026 Accepted poster Non-archival workshop

CandleDiff: Contextual Diffusion for Structurally Valid OHLC Generation

Siddhant Gole, Nachiketa Ratnakar Patil, Vaibhav Rathore, Siddharth Sureka, Neha Bhargava

Vaibhav's affiliation: Motilal Oswal Financial Services

Summary & reference for CandleDiff: Contextual Diffusion for Structurally Valid OHLC Generation

Introduces contextual diffusion for controllable firm-level OHLC generation with guaranteed candle geometry and market-aware conditioning.

Siddhant Gole, Nachiketa Ratnakar Patil, Vaibhav Rathore, Siddharth Sureka, Neha Bhargava. “CandleDiff: Contextual Diffusion for Structurally Valid OHLC Generation.” AI4F Workshop at ACM ICAIF 2026. Accepted poster; non-archival workshop.

AI4F Workshop at ACM ICAIF 2026 Accepted poster Non-archival workshop

Position: Many Agents, Few Minds — Model Concentration as a Governance Metric for Agent-Native Financial Markets

Vaibhav Rathore, Siddhant Gole, Siddharth Sureka, Neha Bhargava

Vaibhav's affiliation: Motilal Oswal Financial Services

Summary & reference for Position: Many Agents, Few Minds — Model Concentration as a Governance Metric for Agent-Native Financial Markets

Argues that agent-based finance can decentralize intermediation while concentrating decisions in shared foundation models, and proposes a metric for that risk.

Vaibhav Rathore, Siddhant Gole, Siddharth Sureka, Neha Bhargava. “Position: Many Agents, Few Minds — Model Concentration as a Governance Metric for Agent-Native Financial Markets.” AI4F Workshop at ACM ICAIF 2026. Accepted poster; non-archival workshop.

Multi-LLL Workshop at AACL-IJCNLP 2026 Published

Beyond Accuracy: What Language-Subspace Interventions Capture, Remove, and Improve

Shaanil Punglia, Dadhichi Telwadkar, Vaibhav Rathore, Siddharth Sureka, Neha Bhargava

Summary & reference for Beyond Accuracy: What Language-Subspace Interventions Capture, Remove, and Improve

Evaluates whether language-subspace interventions truly remove language information and improve multilingual mathematical reasoning without sacrificing response-language fidelity.

Shaanil Punglia, Dadhichi Telwadkar, Vaibhav Rathore, Siddharth Sureka, Neha Bhargava. “Beyond Accuracy: What Language-Subspace Interventions Capture, Remove, and Improve.” Multi-LLL Workshop at AACL-IJCNLP 2026. Published.

NewInML Workshop at NeurIPS 2026 Poster

Towards Unconstrained Pedestrian Attribute Recognition: Evaluating Domain Shift and Open-World Constraints

Dattaraj Salunkhe, Vaibhav Rathore, Biplab Banerjee

Summary & reference for Towards Unconstrained Pedestrian Attribute Recognition: Evaluating Domain Shift and Open-World Constraints

Studies pedestrian attribute recognition under domain shift and open-world constraints.

Dattaraj Salunkhe, Vaibhav Rathore, Biplab Banerjee. “Towards Unconstrained Pedestrian Attribute Recognition: Evaluating Domain Shift and Open-World Constraints.” NewInML Workshop at NeurIPS 2026. Poster.