Bharat Yalavarthi

Bharat Yalavarthi

Interpretability · Alignment

PhD Student · University at Buffalo, SUNY

I am interested in intepreting deep learning models, and in using interpretability techniques to improve them. This includes making models more robust and better aligned, and building evaluation benchmarks that reveal where they fail. I am advised by Dr. Nalini Ratha and Dr. Venu Govindaraju.

Looking for research internships — Fall 2026 onward.

Publications

Full list on Google Scholar. (* equal contribution)

Interpretability

  • Label-Free Mitigation of Spurious Correlations in VLMs Using Sparse Autoencoders B. C. Yalavarthi, N. Ratha, V. Govindaraju ICLR 2026 Finds spurious feature directions in SAE basis and projects them out of VLM representations — fully zero-shot, no labels or retraining.
    Paper
  • Aligning Characteristic Descriptors with Images for Human-Expert-like Explainability B. C. Yalavarthi, N. Ratha NeurIPS 2024 Workshop on Interpretable AI Uses Mistral-7B and CLIP to generate expert-style justifications for face and medical predictions, grounded in characteristic descriptors.
    Paper

Privacy & Secure ML

  • Shielding Latent Face Representations From Privacy Attacks A. R. Kaushik, B. C. Yalavarthi, A. Ross, V. Boddeti, N. Ratha IEEE FG 2025 Defends face-template representations against attribute-inference and reconstruction attacks.
    Paper
  • Enhancing Privacy in Face Analytics Using Fully Homomorphic Encryption B. Yalavarthi *, A. R. Kaushik *, A. Ross, V. Boddeti, N. Ratha IEEE FG 2024 An FHE-based scheme that cuts attribute leakage from face templates by 35% over prior work with no accuracy loss.
    Paper
  • Efficient Convolution Operator in FHE Using Summed Area Table B. Yalavarthi, C. Jutla, N. Ratha ICPR 2024 A summed-area-table formulation that speeds up convolutions on encrypted data.
    Paper

Education

  • Ph.D. in Computer Science 2025 – Present
    University at Buffalo, SUNY — UB Presidential Fellowship

    Advisors: Dr. Nalini Ratha & Dr. Venu Govindaraju

  • M.S. in Computer Science 2022 – 2024
    University at Buffalo, SUNY
  • B.Tech. in Computer Science 2016 – 2020
    Vellore Institute of Technology (VIT)

Experience

  • Researcher Aug 2024 – Jan 2025
    SUNY Research Foundation, Buffalo, NY

    Benchmarking and Evaluating MLLMs on Visual Tasks.

  • Research Assistant Jan 2023 – May 2024
    SUNY Research Foundation, Buffalo, NY

    Human Expert-like Explainability, Privacy Preserving ML, Fully Homomorphic Encryption

  • Software Engineer — ADAS & Navigation Oct 2020 – Jul 2022
    Harman International (Samsung)

    Delivered 20+ ADAS and route guidance features for VW Trucks/Coaches. Modular redesign cut software maintenance costs by 25%.