Peshal Agarwal

Peshal Agarwal

Machine Learning Engineer

Biography

I am a Machine Learning engineer at Pallon (a ETH AI startup) responsible for building computer vision models to identify defects inside sewers. I earned my masters in Statistics with a focus on AI from ETH Zurich. I completed my masters thesis in Unsupervised Domain Adaptation under the supervision of Prof. Luc Van Gool at the Computer Vision Lab at ETH.

Before moving to Switzerland, I did my bachelors and masters in Mathamatics and Scientific Computing from IIT Kanpur in India where I was fortunate to work with Prof. Debasis Kundu on Bayesian Analysis.

I had the privilege to intern at IBM Research and Goldman Sachs during my studies. You can find some of work at Google Scholar.

Alt Text 1 Alt Text 2 Alt Text 3
Interests
  • Artificial Intelligence
  • Computer Vision
  • Adversarial Machine Learning
Education

  • MS in Statistics, 2021

    ETH Zurich

  • MS in Mathematics and Computing, 2018

    Indian Institute of Technology (IIT) Kanpur

  • BS in Mathematics and Computing, 2018

    Indian Institute of Technology (IIT) Kanpur

Experience

 
 
 
 
 
Pallon
Software Engineer (ML)
Pallon
August 2021 – Present Zurich, Switzerland
  • Develop deep learning solutions for automating sewer inspection by identifying defects for 150+ clients.
  • Trained transformers (Swin, DETR) for object detection and Segment Anything Model (SAM) for semantic segmentation on 1TB+ data.
  • Led the team for transitioning the code base from TensorFlow 1.x to PyTorch 2.0, resulting in a notable 28% reduction in training time and 50% decrease in development time.
  • Build data pipelines for training and CI/CD framework for smooth deployment. Integrated tools and services like Docker, MLFlow, GCP, and Poetry for optimized workflow.

Skills

Technical
Python
PyTorch / Tensorflow
GCP / AWS
Domain
Machine Learning
Statistics
Linear Algebra

Projects

*
All Deep Learning Other
Neural Network Verifier
Build a precise and scalable automated verifier for proving the robustness of fully connected and convolutional neural networks against adversarial attacks
Code
Neural Network Verifier
Adversarial ML
Implemented Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) attacks on networks trained over CIFAR10 and MNIST.
Code
Adversarial ML
Bayesian Analysis of Skew Normal Distribution
Formulated suitable parameters on all the three parameters of Geometric Skew Normal distribution to perform Bayesian analysis and evaluated the fit on multiple datasets using Kolmogorov-Smirnov test statistic.
PDF
Bayesian Analysis of Skew Normal Distribution
Topic Modeling with Metadata
Analysed Dirichlet-Multinomial Regression (DMR) model for topic modeling with metadata. Derived and implemented Stochastic Gradient Riemann Langevin Dynamics on DMR model.
PDF Slides
Topic Modeling with Metadata
Taste and Tell
Build an automatic review generator and restaurant recommender system
PDF Slides
Taste and Tell

Recent Publications

Quickly discover relevant content by filtering publications.
Thomas Parnell, Andreea Anghel, Małgorzata Łazuka, Nikolas Ioannou, Sebastian Kurella, Peshal Agarwal, Nikolaos Papandreou, Haralampos Pozidis (2020). SnapBoost: A Heterogeneous Boosting Machine. NeurIPS 2020.

PDF Cite Code Source Document

Accomplish­ments

Coursera
Introduction to Docker
Coursera Oct 2020
See certificate

Contact

Feel free to connect with me.