sub_rbala

Rishab
Balasubramanian

I am a third year PhD student at Virginia Tech, advised by Tu Vu. My research interests are Language Modeling, Efficiency, Generalization, and Post-Training Alignment. Previously, I worked with Huazheng Wang on adversarial attacks on bandit algorithms and the impact of ordering in recommendation systems. Long long ago, I completed my Bachelor's (B.Tech) from the National Institute of Technology Tiruchirapalli (NIT-Trichy), India, where I was a part of RMI, the robotics club.

Explore my research
Rishab in San Francisco

Research Interests

  • Language Modeling

    How language models learn, reason, and generate text.

  • Efficiency

    Ways to reduce the compute and memory costs of language models.

  • Generalization

    How models perform on tasks and data beyond their training distribution.

  • Post-Training Alignment

    Post-training methods that align language models with human goals and preferences.

02 /

Recent News

  • [10/01/2026]One paper accepted to NeurIPS 2026 as Spotlight!
  • [05/18/2026]Started my internship at Amazon AGI
  • [05/05/2026]New paper on generalization and transferability in language models
  • [02/14/2026]New paper on recursive agents
  • [05/20/2025]One paper accepted to EMNLP 2025.
  • [09/12/2024]One paper accepted to TMLR.
  • [25/08/2024]Started PhD at Virginia Tech.
  • [01/05/2024]One paper accepted to ICML 2024.
  • [09/26/2023]One paper accepted to NeurIPS 2024.
03 /

Research Work

The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment
Rishab Balasubramanian, Pin-Jie Lin, Rituraj Sharma, Anjie Fang, Fardin Abdi, Viktor Rozgic, Zheng Du, Mohit Bansal, Tu Vu
Spotlight (<2%) at NeurIPS 2026.

Efficient Model Development through Fine-tuning Transfer
Pin-Jie Lin, Rishab Balasubramanian, Fengyuan Liu, Nikhil Kandpal, Tu Vu
Accepted to EMNLP 2025.

Adversarial Attacks on Combinatorial Multi-Armed Bandits
Rishab Balasubramanian, Jiawei Li, Prasad Tadepalli, Huazheng Wang, Qingyun Wu, Haoyu Zhao
Accepted to ICML 2024.
[paper]

Adversarial Attacks on Online Learning to Rank with Stochastic Click Models
Zichen Wang, Rishab Balasubramanian, Hui Yuan, Chenyu Song, Mengdi Wang, Huazheng Wang
Accepted to TMLR.
[arXiv] [paper]

Risk Averse Submodular Optimization for the Multi-Objective Travelling Salesman Problem
Rishab Balasubramanian, Lifeng Zhou, Pratap Tokekar, P.B. Sujit
Accepted to IROS 2021.
[paper] [video] [code]

A Cooperative Framework for Autonomous Landings of Quadrotors using Vision on a Moving UGV
Rishab Balasubramanian, P.B. Sujit
Accepted to AIAA SciTech 2021.
[paper] [code] [slides] [video]

04 /

Other Projects

3D reconstruction from endoscopic images
Rishab Balasubramanian, EndovisionAi
[page]

Contrastive Learning for Out-of-Distribution Object Detection
Rishab Balasubramanian, Rupashree Dey, Kunal Rathore
Preprint - arXiv.
[arXiv] [code]

Autonomous Mapping and Safe Footfall Planning for Hexapods
Rishab Balasubramanian, Sun Ge, Guillaume Sartoretti
Autonomous control of a legged robot to map a room using the RTABMap framework.
[page] [code]

Formation Control and Collision Avoidance for Multi-Agent Systems
Rishab Balasubramanian, Prof. M Umapathy
Multi-agent collision avoidance using the ORCA algorithm, compared with other algorithms.
[page] [thesis]

Real-Time American Sign Language to Text Conversion Using Convolutional Neural Networks
Rishab Balasubramanian, Ramasubba Reddy
[page] [abstract] [code]