Subrata Biswas

OWL: Geometry-Aware Spatial Reasoning for Audio Large Language Models

Subrata Biswas, Mohammad Nur Hossain Khan, Bashima Islam

A geometry-aware audio encoder (SAGE) plus a spatially grounded chain-of-thought that reasons over direction-of-arrival and distance. Reduces mean DoA error by 11° and improves spatial QA accuracy by up to 25% over BAT; releases BiDepth, 1M+ QA pairs.

Hair Noise Analysis and Mitigation for Smart Glasses Audio Captures

Subrata Biswas, Daniel Wong, Bashima Islam, Sanjeel Parekh, Vladimir Tourbabin

First systematic study of "hair noise" on AR/smart glasses. Introduces the multi-channel HNM dataset and semi-supervised NMF mitigation baselines, establishing hair noise as a distinct challenge for wearable audio.

RAVEN: Query-Guided Representation Alignment for Question Answering over Audio, Video, Embedded Sensors, and Natural Language

Subrata Biswas, Mohammad Nur Hossain Khan, Bashima Islam

A unified multimodal QA architecture whose core (QuART) assigns query-conditioned relevance scores per token to amplify informative signals and suppress distractors before fusion — robust to modality disagreement.

EgoAdapt: Adaptive Multisensory Distillation and Policy Learning for Efficient Egocentric Perception

Sanjoy Chowdhury, Subrata Biswas, Sayan Nag, Tushar Nagarajan, Calvin Murdock, Ishwarya Ananthabhotla, Yijun Qian, Vamsi Krishna Ithapu, Dinesh Manocha, Ruohan Gao

Adaptive cross-modal distillation and policy learning for efficient egocentric perception — cuts GMACs by up to 89%, parameters by 82%, and energy by 9.6× while matching or beating state-of-the-art.

QUADS: Quantized Distillation Framework for Efficient Speech Language Understanding

Subrata Biswas, Mohammad Nur Hossain Khan, Bashima Islam

A unified framework that jointly optimizes distillation and quantization through multi-stage training, improving adaptability to low-bit regimes while maintaining accuracy.

LOCUS: Localization with Channel Uncertainty and Sporadic Energy

Subrata Biswas, Mohammad Nur Hossain Khan, Alex Colwell, Jack Adiletta, Bashima Islam

A deep learning framework that recovers corrupted multichannel features for sound source localization on batteryless systems, restoring direction-of-arrival accuracy under intermittent energy harvesting.

Missingness-Resilient Video-Enhanced Multimodal Disfluency Detection

Payal Mohapatra, Shamika Likhite, Subrata Biswas, Bashima Islam, Qi Zhu

A practical audio-visual disfluency detection approach with weight-sharing, modality-agnostic encoders that stays robust when video is missing at inference.

Memory-Efficient Energy-Adaptive Inference of Pre-Trained Models on Batteryless Embedded Systems

Subrata Biswas, Pietro Farina, Eren Yildiz, Khakim Akhunov, Saad Ahmed, Bashima Islam, Kasim Sinan Yildirim

FreeML optimizes pre-trained DNNs for memory-efficient, energy-adaptive inference on batteryless systems that face frequent power failures.