Unmesh Raskar

Biography

I am a master’s student in the Electrical and Computer Engineering at UW-Madison. My areas of interest are Machine Learning, Computer Vision and Embedded Systems. I completed my bachelor’s in Electrical Engineering from Indian Institute of Technology, Bombay. I worked as a Data Scientist in Siemens. Prior to that, I was engaged in ML research in Zenlabs.

Education

• University of Wisconsin-Madison | August 2023 – Present
  M.S. in Electrical and Computer Engineering
• Indian Institute of Technology, Bombay | July 2016 – August 2020
  B.Tech in Electrical Engineering

Research Interests

• Machine Learning
• Computer Vision
• Embedded Systems
• Generative AI

Awards

• Poster Award and Travel Grant Recipient, Midwest ML Symposium 2024, Minneapolis
• DAC Young Fellow, 61st Design Automation Conference, San Francisco 2024
• RPG Innovation Fest ’21 Zensar First Prize, Smart Utility Servicing
• INSPIRE Fellow ’16

Publications

• U. Raskar et al, “Facial feature extraction and emotion analysis using machine learning”, accepted at
  International Journal of all research education and scientific methods, IJARESM 2022

Work Experience

Cadence Design Systems, San Jose

ML summer intern, Placement and Routing, Virtuoso | May 2024 – August 2024

• Worked on AI-assisted workflows in placement and routing of custom IC/Analog/RF Design inside Virtuoso
• Created a novel, complex ML dataset with limited data and compute and delivered a PoC (Proof-of-Concept)
•  Improved routing efficiency by 12% by training an AI model optimized to perform in minimum compute settings

University of Wisconsin-Madison

Research Assistant, WiSEST Lab | August 2023 – Present

• Developing Computer Vision system for detecting behavior changes in cattle using multimodal data (sensors and video) to efficiently use electricity and water for the cooling systems inside barns

Siemens

Data Scientist | September 2022 – August 2023

• Solved ML engineering problems in NX, a widely used CAD (Computer Aided Design) software
• Reduced the number of click-operations involved in machining feature detection of 3D models using topology graphs of 3D parts
• Created CAM feature detection in manufacturing parts using state-of-the-art DL models trained on 3D point clouds, the
  only AI solution amongst all competitors

Zensar

ML Researcher, Zenlabs | November 2020 – August 2022

• Fields of Research: Computer Vision, Generative AI, Explainable AI, Multi-Armed Bandits
• Worked on a patent, created 4 PoCs (Proof-of-Concept) using state-of-the-art DL techniques
• Automated an expert safety-critical process using AI, signed £0.4 M deal with the client
• Collaborated with academia: Monash University, Jilin University for Automated Android UI Generation
• Published two blogposts (Intro to GANs) and authored a whitepaper (Synthetic-Data-as-a-Service)
• Designed curriculum & executed ML crash-course training for 19 freshers from other IT domains over 6 months

Research Experience

Automated Android UI Design Generation using GANs | Zenlabs | April 2021

• To assist UI/UX designers, developed a GAN to automatically generate UI designs with aesthetic compatibility
• Novel natural-language generation approach reused existing UI components from apps UI designs from RICO dataset
• Modified SeqGAN architecture for the Generator with Monte Carlo Tree Search Policy Gradient
• Devised new Loss functions to capture natural relations in dataset, achieved FID score of 0.075

Synthetic Tabular Data Generation using GANs | Zenlabs | May 2022

• Enabled data sharing for a global financial client facing lengthy legal processes (as per GDPR) using synthetic data
• Used GAN model to learn joint distribution of columns and generate realistic, statistically equivalent tabular data
• Designed Synthetic Data Quality Evaluation framework consisting of likelihood fitness, ML efficacy, KL Divergence, etc
• Synthetic Data unlocks faster testing, analytics, innovation across silos and organizations (3x faster)

Graphical Layout Generation using GANs | Zenlabs | September 2021

• Synthesized realistic 2D layouts of graphic elements for graphic design and scene synthesis
• Used self-attention modules in Generator and novel wireframe rendering Discriminator to produce optimized layouts
• Validated on MNIST digit generation, document layout generation, abstract scene synthesis, tangram graphic design and
  evaluated against state-of-the-art using Inception Score and spatial analysis

Risk Aware Portfolio Optimization using Multi-Armed Bandits | Bachelor’s Thesis, IIT Bombay | November 2019

• Implemented a usecase of the work of the research group that published at NeurIPS 2019, Vancouver
• Modeled Stock Portfolio as a fixed budget, best-arm identification, stochastic multi-armed bandit problem
• Predicted the arm which minimizes a linear combination of the expected loss and a risk-sensitive metric: CVaR
• Verified the proposed novel estimators for the CVaR of unbounded random variables (potentially heavy tailed)

Mentoring

• Teaching Assistant – ECE 210 Introduction to Electrical Engineering  (UW–Madison) –Fall 2023, Fall 2024
• ML Crash Course Training – Moonlight Program (Zenlabs) – May 2021 – December 2021

Extra-Curricular

• Github/EDA coordinator – ML+X, data science @ uw – Spring 2024 – Present
• Public Relations Officer – GSA, ECE – Fall 2024