Soumya Roy

Soumya Roy, Soumya Banerjee, Vinay Verma et al.

Machine unlearning (MUL) refers to the problem of making a pre-trained model selectively forget some training instances or class(es) while retaining performance on the remaining dataset. Existing MUL research involves fine-tuning using a forget and/or retain set, making it expensive and/or impractical, and often causing performance degradation in the unlearned model. We introduce {\pname}, an unlearning-aware vision transformer-based architecture that can directly perform unlearning for future unlearning requests without any fine-tuning over the requested set. The proposed model is trained by simulating unlearning during the training process itself. It involves randomly separating class(es)/sub-class(es) present in each mini-batch into two disjoint sets: a proxy forget-set and a retain-set, and the model is optimized so that it is unable to predict the forget-set. Forgetting is achieved by withdrawing keys, making unlearning on-the-fly and avoiding performance degradation. The model is trained jointly with learnable keys and original weights, ensuring withholding a key irreversibly erases information, validated by membership inference attack scores. Extensive experiments on various datasets, architectures, and resolutions confirm {\pname}'s superiority over both fine-tuning-free and fine-tuning-based methods.

Jiaru Zou, Soumya Roy, Vinay Kumar Verma et al.

Process Reward Models (PRMs) have recently emerged as a powerful framework for enhancing the reasoning capabilities of large reasoning models (LRMs), particularly in the context of test-time scaling (TTS). However, their potential for supervising LRMs on tabular reasoning domains remains underexplored. Through detailed empirical analyses, we identify that existing PRMs, though widely adopted for supervising text-only reasoning steps, struggle with table-specific operations such as sub-table retrieval and schema interaction, leading to critical performance bottlenecks. To address this limitation, we propose TaTToo, a novel table-grounded PRM framework that (i) reasons explicitly over tabular reasoning steps and (ii) integrates tool-based verification to provide precise reward supervision. Concretely, we first design a scalable data curation pipeline that constructs over 60k high-quality step-level annotations by integrating table verification rationales with tool-based executions. Building on the collected data, we train TaTToo with a dual-stage paradigm: cold-start supervised fine-tuning to capture tool-use reasoning patterns, followed by reinforcement learning with tool-grounded reward shaping to align our model with table-based verification. We provide a comprehensive evaluation of the policy improvement induced by our newly designed PRM. Across 5 challenging tabular reasoning benchmarks covering numerical reasoning, fact-checking, and data analysis, TaTToo improves downstream policy LRMs by 30.9% at inference, surpasses strong PRM baselines such as Qwen-2.5-Math-PRM-72B with only 8B parameters, and demonstrates strong generalizability across diverse TTS strategies.

Soumya Roy, Vinay P. Namboodiri, Arijit Biswas

Given an image, we would like to learn to detect objects belonging to particular object categories. Common object detection methods train on large annotated datasets which are annotated in terms of bounding boxes that contain the object of interest. Previous works on object detection model the problem as a structured regression problem which ranks the correct bounding boxes more than the background ones. In this paper we develop algorithms which actively obtain annotations from human annotators for a small set of images, instead of all images, thereby reducing the annotation effort. Towards this goal, we make the following contributions: 1. We develop a principled version space based active learning method that solves for object detection as a structured prediction problem in a weakly supervised setting 2. We also propose two variants of the margin sampling strategy 3. We analyse the results on standard object detection benchmarks that show that with only 20% of the data we can obtain more than 95% of the localization accuracy of full supervision. Our methods outperform random sampling and the classical uncertainty-based active learning algorithms like entropy