Pratyaksh Gautam

prof_pic.jpg

I’m a student researcher working as a research assistant at the University of Jyväskylä (JYU), under the supervision of Petri Toivainen. My work at JYU is on concept-based interpretability of deep neural networks for music, specifically looking at the causal relationships between lower-level concepts like instruments and higher-level concepts like genre.

I’m working my Master’s thesis at IIIT Hyderabad, advised by Vinoo Alluri and Makarand Tapaswi.

My research is primarily concerned with how deep neural networks process and organize information for auditory tasks. I’m also interested in interpretability and auditory processing at large, multi-modal deep learning in vision-language and audio-language models.

selected publications

  1. Auditory CNN Analysis: What Do Layers Encode?
    Pratyaksh Gautam, Makarand Tapaswi*, and Vinoo Alluri*
    Under Review at ICASSP, 2025
  2. Localizing Auditory Concepts in CNNs
    Pratyaksh Gautam, Makarand Tapaswi*, and Vinoo Alluri*
    Accepted at the Mechanistic Interpretability Workshop, ICML, 2024