Pramod Kaushik Mudrakarta

Machine Learning @ Google

Mountain View, California, United States

About

7+ years experience in Machine Learning research, and passionate about making real-world impact. Published at top conferences (NeurIPS, ACL, ICLR). Familiar with best software engineering practices.

Experience

  • Software Engineer at Google
    Nov 2019 - Present · 6 yrs 9 mos

  • The University of Chicago ()
    • Teaching Assistant
      Sep 2014 - Aug 2019 · 5 yrs

      Held lab sessions, recitations, office hours and one-on-one support sessions every academic quarter for a wide variety of computer science courses - from introductory computer programming (C++, Python, DrRacket) to machine learning, large-scale data analysis and natural language processing.

    • Ph.D. Student
      Sep 2014 - Aug 2019 · 5 yrs

      Wrote a thesis proposing techniques that address three prominent challenges in the modern Machine Learning pipeline: 1) exploiting multiresolution/multiscale structure in large datasets 2) improving the interpretability of Deep Neural Network models, and 3) efficient deploying a large number of Deep Neural Network models on embedded devices

  • Software Engineering Intern in Research at Google
    Jun 2018 - Sep 2018 · 4 mos

    - Intern in the Mobile Vision team - Analyzed memorization in deep neural networks - Developed parameter-efficient methods for repurposing pretrained deep networks to new tasks - Published a paper at ICLR 2019

  • Software Engineering Intern at Google
    Jun 2017 - Sep 2017 · 4 mos

    - Intern in Google Brain (TensorFlow team) - Developed methods for understanding what deep neural networks learn, and uncovering their weaknesses/failure modes. - Crafted adversarial attacks against state-of-the-art question-answering models by exploiting failure modes - Published two papers, one at ACL 2018

  • Applied Scientist Intern at Amazon
    Jun 2016 - Sep 2016 · 4 mos

    - Intern in the Personalization team - Developed Recurrent Neural Network models for product recommendations based on user purchase history