AIConfPaper
← Search

Soumya Suvra Ghosal

12 accepted papers

2026

Safety Recovery in Reasoning Models Is Only a Few Early Steering Steps Away

Soumya Suvra Ghosal, Souradip Chakraborty, Vaibhav Singh, Furong Huang, Dinesh Manocha, Amrit Singh Bedi
ICML 2026poster

Reinforcement learning (RL) based post-training for explicit chain-of-thought (e.g., GRPO) improves the reasoning ability of multimodal large-scale reasoning models (MLRMs). But recent evidence shows that it can simultaneously degrade safety alignment and increase jailbreak success rates. We propose…

Cited by 0Source ↗Scholar ↗
2026

VisRef: Visual Refocusing while Thinking Improves Test-Time Scaling in Multi-Modal Large Reasoning Models

Soumya Suvra Ghosal, Youngeun Kim, Zhuowei Li, Ritwick Chaudhry, Linghan Xu, Hongjing Zhang +3
CVPR 2026

Advances in large reasoning models have shown strong performance on complex reasoning tasks by scaling test-time compute through extended inference-time thinking. However, recent studies observe that in vision-dependent tasks, extended textual reasoning at inference time can often degrade performanc

Cited by 0Source ↗Scholar ↗
2025

Bounded Rationality for LLMs: Satisficing Alignment at Inference-Time

Mohamad Fares El Hajj Chehade, Soumya Suvra Ghosal, Souradip Chakraborty, Avinash Reddy, Dinesh Manocha, Hao Zhu +1
ICML 2025poster

Aligning large language models with humans is challenging due to the inherently multifaceted nature of preference feedback. While existing approaches typically frame this as a multi-objective optimization problem, they often overlook how humans actually make decisions. Research on bounded rationalit…

Cited by 0Source ↗PDF ↗Scholar ↗
2025

Collab: Controlled Decoding using Mixture of Agents for LLM Alignment

Souradip Chakraborty, Sujay Bhatt, Udari Madhushani Sehwag, Soumya Suvra Ghosal, Jiahao Qiu, Mengdi Wang +4
ICLR 2025poster

Alignment of Large Language models (LLMs) is crucial for safe and trustworthy deployment in applications. Reinforcement learning from human feedback (RLHF) has emerged as an effective technique to align LLMs to human preferences, and broader utilities, but it requires updating billions of model para…

Cited by 1Source ↗PDF ↗Scholar ↗
2025

Does Thinking More Always Help? Mirage of Test-Time Scaling in Reasoning Models

Soumya Suvra Ghosal, Souradip Chakraborty, Avinash Reddy, Yifu Lu, Mengdi Wang, Dinesh Manocha +3
NeurIPS 2025poster

Recent trends in test-time scaling for reasoning models (e.g., OpenAI o1, DeepSeek R1) have led to a popular belief that extending thinking traces using prompts like “Wait” or “Let me rethink” can improve performance. This raises a natural question: Does thinking more at test-time truly lead to bet…

Cited by 0Source ↗Scholar ↗
2025

Immune: Improving Safety Against Jailbreaks in Multi-modal LLMs via Inference-Time Alignment

Soumya Suvra Ghosal, Souradip Chakraborty, Vaibhav Singh, Tianrui Guan, Mengdi Wang, Ahmad Beirami +4
CVPR 2025poster

With the widespread deployment of Multimodal Large Language Models (MLLMs) for visual-reasoning tasks, improving their safety has become crucial. Recent research indicates that despite training-time safety alignment, these models remain vulnerable to jailbreak attacks--carefully crafted image-prompt…

Cited by 3Source ↗PDF ↗Scholar ↗
2025

PromptRefine: Enhancing Few-Shot Performance on Low-Resource Indic Languages with Example Selection from related Example Banks

Soumya Suvra Ghosal, Soumyabrata Pal, Koyel Mukherjee, Dinesh Manocha
NAACL 2025long

Large Language Models (LLMs) have recently demonstrated impressive few-shot learning capabilities through in-context learning (ICL). However, ICL performance is highly dependent on the choice of few-shot demonstrations, making the selection of the most optimal examples a persistent research challeng…

Cited by 0Source ↗PDF ↗Scholar ↗
2025

RELIC: Enhancing Reward Model Generalization for Low-Resource Indic Languages with Few-Shot Examples

Soumya Suvra Ghosal, Vaibhav Singh, Akash Ghosh, Soumyabrata Pal, Subhadip Baidya, Sriparna Saha +1
EMNLP 2025

Reward models are essential for aligning large language models (LLMs) with human preferences. However, most open-source multilingual reward models are primarily trained on preference datasets in high-resource languages, resulting in unreliable reward signals for low-resource Indic languages. Collect

Cited by 0Source ↗PDF ↗Scholar ↗
2024

How to Overcome Curse-of-Dimensionality for Out-of-Distribution Detection?

Soumya Suvra Ghosal, Yiyou Sun, Yixuan Li
AAAI 2024technical

Machine learning models deployed in the wild can be challenged by out-of-distribution (OOD) data from unknown classes. Recent advances in OOD detection rely on distance measures to distinguish samples that are relatively far away from the in-distribution (ID) data. Despite the promise, distance-base…

Cited by 14Source ↗PDF ↗code ↗Scholar ↗
2024

IntCoOp: Interpretability-Aware Vision-Language Prompt Tuning

Soumya Suvra Ghosal, Samyadeep Basu, Soheil Feizi, Dinesh Manocha
EMNLP 2024main

Image-text contrastive models such as CLIP learn transferable and robust representations for zero-shot transfer to a variety of downstream tasks. However, to obtain strong downstream performances, prompts need to be carefully curated, which can be a tedious engineering task. To address the issue of…

Cited by 2Source ↗PDF ↗Scholar ↗
2024

Transfer Q-star : Principled Decoding for LLM Alignment

Souradip Chakraborty, Soumya Suvra Ghosal, Ming Yin, Dinesh Manocha, Mengdi Wang, Amrit Bedi +1
NeurIPS 2024poster

Aligning foundation models is essential for their safe and trustworthy deployment. However, traditional fine-tuning methods are computationally intensive and require updating billions of model parameters. A promising alternative, alignment via decoding, adjusts the response distribution directly wit…

Cited by 20Source ↗PDF ↗Scholar ↗