Sayantan Das

Independent Researcher

ETH Zurich (Dr. Bastian Rieck)

Biography

Thanks for visiting. I am a researcher largely interested in the unison of Differential Geometry and Generative Modeling. I love to ask questions like:

Geometry has direct implications in interpretability for Machine Learning, but how far can we use it to reason how Deep Networks behave?
How can we make topological tools usable/differentiable for Machine Learning?

Previsously, I have worked at Indian Space Research Organisation and Indian Statistical Institute Kolkata where I worked on Microwave Image Processing and Hyperspectral Deep Learning respectively. In my free time , I read about Biblical mythology and Philosophy in an attempt to deconstruct reasons for human existence.

Interests

Generative Models
Topological Data Analysis
Differential Geometry

Education

BTech in Computer Science and Engineering, 2020
Institute of Engineering and Management Kolkata

Skills

Python

100%

Machine Learning

100%

Meshes and Point Cloud

100%

Linux

100%

Experience

Independent Researcher with Dr. Bastian Rieck

ETH Zurich

Aug 2020 – Present Remote

Working on Topological Data Analysis and Manfiold Learning for Deep Generative Models

Research Associate

Indian Statistical Institute

Mar 2020 – Aug 2020 Kolkata

Working with Bhabatosh Chanda on Hyperspectral Deep Learning

Google Code-In 2019 Mentor

Tensorflow

Dec 2019 – Feb 2020 Remote

Tasks I mentored for :

[Screencast] Install TF2.0 GPU on Windows
Build a model using tf.keras Functional API.
Complete Adding Mozilla Common Voice Dataset
Deploy your model on Firebase for iOS /Android
Create a Pull Request adding a Dataset to Tensorflow Datasets Github Repository
Make an object detection App using TensorFlow Lite

Research Intern

Indian Space Research Organization

Jun 2019 – Aug 2019 Ahmedabad

Worked with the Microwave Data Processing Department on Synthetic Aperture Radar Applications

Accomplishments

Mentor

OpenMined Feb 2020 – May 2020

Open Source Contributions and Mentoring on Secure and Private AI

Project Administrator

GirlScript Summer of Code Feb 2020 – Jun 2020

For my project - https://ucalyptus.github.io/Spotify-Recommendation-Engine/

Facilitator for IEM Kolkata

Explore ML by Google AI Jul 2019 – Apr 2020

Flipped Classrooms in AI -> https://github.com/explore-ml-iemk/ML-Summit-Workshops

Projects

Cloud Movers Distance

Employing Optimal Transport metrics for Point Cloud Registration

GaborConv2D

Gabor Convolutional Layer in Pytorch. Created using subclassing the ConvND class. Similar implementation can be made in tf2 by subclassing tf.python.keras.convolutional’s Conv class which describes the Conv op.

Pneumothorax Segmentation Using Hypercolumns

Collapsed Lung Segmentation using Hypercolumns . Open to Contributions

EarthEngine Deep Learning

Land Cover Classification based on Landsat-8 imagery from Google Earth Engine

Scikit on GRPC

Using Protocol Buffers and gRPC client-server communication to deploy a scikit-learn joblib exported model.

Spotify Recommendation Engine

This is a rapid prototyped presentation of how a Spotify Recommendation Engine should work . A system that recommends songs from your existing playlists using Spotify API and a bit of classical machine learning techniques. Vision

Recent & Upcoming Talks

IEM Computer Vision Club Meeting Amidst COVID 19 Lockdown

Dec 9, 2020 3:12 AM — 3:12 AM

Computer Vision Meets Remote Sensing

Dec 9, 2020 3:11 AM — 3:11 AM

Point Target Analysis and Detection of Moving Targets in SAR Images

Dec 9, 2020 3:11 AM — 3:11 AM

Rethinking Earth Observation Using Computer Vision Appraoaches

Dec 9, 2020 3:10 AM — 3:10 AM

Featured Publications

Adrish Dey, Sayantan Das

2020-02-12 NeurIPS 2020

Topo Sampler: A Topology Constrained Noise Sampling for GANs

This work studies disconnected manifold learning in generative models in the light of point-set topology and persistent homology. Under this formalism, the topological similarity of latent space in generative models with the underlying manifold of data distribution facilitates better generalization. To achieve this, we introduce a topology-constrained noise sampler, responsible for mapping the samples from Gaussian spheres to a latent embedding space, which in turn is constrained to be topologically similar to the manifold underlying the data distribution. We study the effectiveness of this method in GANs for learning disconnected manifolds. This is ongoing research, with the current report containing preliminary empirical experiments.

PDF Code Poster Video NeurIPS 2020 Workshop TDA and Beyond