Madhav Kanda

Hello! I am an MS CS (Thesis Track) student at the University of Illinois Urbana-Champaign. I completed my Bachelor of Technology in Computer Science and Engineering with Honors from IIT Gandhinagar. I am passionate about Machine Learning, Computer Vision, and Probabilistic Programming.

In addition to working as a Computer Vision intern at Spyne.ai, I have gained hands-on experience at leading research labs at the Aalto University in Finland and the Indian Institute of Science in Bengaluru. Additionally, I have actively contributed to open-source projects such as NumPyro, pyprobml, and BDA3. My contributions to these projects have earned me acknowledgment in the preface of Dr. Kevin Murphy’s book “Probabilistic Machine Learning: Advanced Topics.”

Research and Work Experience

  • ASci Summer Intern, Aalto University, Finland
    • Among the 50 students from around the world out of 1600+ applicants who applied for ASci internship program to get selected to work at Aalto University. I worked on Automatic Reparameterization for Bayesian Inference in Prof. Aki Vehtari’s group.
    • Modified Hamiltonian Monte Carlo (HMC) to reduce the divergences during sampling by implementing methods for learning non-linear reparameterization alongside the existing linear ones during the warmup phase.
    • Proposed algorithm performed better then Variationally Inferred Parameterization in terms of effective sample size per gradient evaluations. We are expanding its application to more models and comparing it with alternative methods.
  • Indian Academy of Science Summer Research Fellow, Indian Institute of Science (IISc.), Bengaluru
    • Among the 68 students from the engineering stream out of the 40,000 applicants who applied for the IAS SRFP to get selected to work at IISc. Bangalore. I worked on enhancing text to speech system in Prof. A.G. Ramakrishnan’s MILE lab.
    • Developed an algorithm to enhance the pronunciation of numbers based on their context in a Hindi sentence. Implemented approach using regular expressions to identify patterns in sentence & predict number’s category.
    • Utilized seq2seq models to transliterate English words into Hindi. Subsequently, I enhanced the pronunciation of Indian proper nouns in English by applying phonetic transcription to these transliterated words.
  • Computer Vision Intern, Spyne Ai
    • Performed an in-depth study of existing literature for the integration of Generative Adversarial Networks (GAN), segmentation maps, and masking for commercial implementation of Virtual Try-On HD research paper.

    • Acquired hands-on experience by implementing GAN models from scratch using PyTorch, ensuring a practical understanding of the intricacies in bringing the concepts to commercial viability.