About me

Graduating in Summer 2026 and seeking internships for Quant Researcher/Software Developer/ML Engineer!



**Update: Our research on Latency-Constrained Input-Aware Quantization of Time Series Inference Workflows at the Edge has been accepted at IEEE International Conference on Computer Communications (INFOCOM). As an A* conference with a 19% acceptance rate this year, we're honored to be presenting this work in London, UK.

I am Ashitabh Misra, a 4th-year PhD student at the University of Illinois Urbana-Champaign in the Department of Computer Science.

I am fortunate to be advised by Dr. Tarek Abdelzaher.

My research is focused on on-device, low-latency machine learning technologies. I specialize in developing adaptive compression techniques for neural networks and probabilistic models, with the goal of making AI more efficient and accessible on resource-constrained devices. Specifically, I am exploring how to leverage physics-based insights and time-series characteristics to estimate optimal compression implementations that achieve maximum accuracy while incurring minimal latency and memory overhead.

Background

Education

  1. Ph.D. in Computer Science

    University of Illinois Urbana-Champaign

    2021 — Present

  2. Bachelor of Technology in Computer Science and Engineering

    Indian Institute of Technology, Bhilai

    2016 — 2020

Experience

  1. Egnyte, Machine Learning Intern

    Poznan, Poland May 2018 ‑ July 2018

    • Developed and deployed a hybrid probabilistic-ML event detection pipeline for smart data caching in Egnyte’s macOS app
    • The solution replaced existing solutions by capturing 12× more target events

    Eduvisor, Software Developer Intern

    Haryana, India December 2017

    • Developed and deployed video processing tools to streamline watermarking and editing instructional videos that were used by multiple schools during COVID-19

    ShopClues, Data Science Intern

    Haryana, India May 2017- July 2017

    • Developed ML-based product recommendation systems that were deployed as part of ShopClues Advertisement Engin
    • The end-to-end pipeline consisted of pre-processing metadata of 1 million products,and novel dimensionality reduction techniques to efficiently train Random Forrest models using XGBoost

News

Latest News

  • [Paper Accepted at IEEE International Conference on Computer Communications (INFOCOM), Acceptance rate: 19%] Latency-Constrained Input-Aware Quantization of Time Series Inference Workflows at the Edge

    15th December 2024

    We developed a novel quantization methodology for neural networks that classify real-time signals (audio, seismic, etc.). Our approach leverages the physical properties of signals to create an input-aware quantization scheme that outperforms traditional methods. The system algorithmically adapts to operating conditions and input characteristics, ensuring an optimal balance between performance and resource consumption in real time.

  • [Data Collection at Army Research Lab, Maryland, USA] Data collection activity of Audio and Seismic signals of moving military vehicles under various operating conditions

    06 August 2024

    I had a great experience working with the team at the Army Research Lab and folks at UIUC as part of the Internet-of-Battlefield-Things (IoBT). We collected audio and seismic data of multiple target vehicles under various operating conditions like extensive background noise, regular and irregular moving patterns, along with various background activities at different durations of the day.

  • [Passed my Qualification Exam!]

    University of Illinois at Urbana Champaign

    20 November 2023

  • [Paper Presentation] ViX: Analysis-driven Compiler for Efficient Low-Precision Variational Inference

    Design, Automation, and Test in Europe Conference (DATE)

    20 April 2023

  • [Paper Accepted!] ViX: Analysis-driven Compiler for Efficient Low-Precision Variational Inference

    Design, Automation, and Test Test in Europe Conference (DATE)

    18 November 2022

Contact

Contact Form