Sulagna Saha (Rasha)

Hi, I'm Sulagna — a computer science graduate from Mount Holyoke College, now pursuing my research MSc at Mila and McGill University under the guidance of Prof. David Rolnick. Currently, I am exploring applications of Machine Learning in Forest monitoring and Bioacoustics. Teaching, giving and receiving feedback inspire me and guide my research direction.

In my free time, I play guitar, cook traditional dishes, read about completely unrelated topic to my current work, reflect and organize my life. In my undergrad, I co-founded BONDHU where each year we bring together hundreds of people around pioneer valley to celebrate bengali culture.

Email  /  CV  /  LinkedIn  /  Scholar  /  Twitter  /  Github

profile photo

Research

I am an aspiring researcher and teacher in the intersection between Machine Learning and Conservation, exploring: How can techniques in CS better handle scarce and messy real-world data in a fair way to create impactful ML models? How can I make a pipeline to handle data faster, more accurate and easier for domain experts? How can I deploy a developed algorithm in real world systems and consider situation specific edge cases?

Improving ReforesTree: Correcting Drone Imagery and Satellite-Based AGB Estimation
Sulagna Saha, Autumn Nguyen, David Dao, Gyri Reiersen, Björn Lütjens
Gainforest.Earth, 2024, AAAI 2022
Correction Report / arXiv

Used the correctly cropped drone imageries to figure out the AGB density from popular open-source satellite maps (GFW, Spawn, and Santoro) to re-benchmark with the available field data. This proved that satellite imagery overestimated the AGB density for the tropical forests.

Machine Learning and Multi-source Remote Sensing in Forest Carbon Stock Estimation: A Review
Autumn Nguyen, Sulagna Saha
Independent project
arXiv

Analyzed 80 papers estimating aboveground biomass, figuring out progress on ML use in four major forest types in the last 5 years and shortlisted 25 papers to mention for review.

Iterative Morphological Training Set Decomposition for Endoscopic Tool Segmentation
Yicheng Zhu, Xiaoyi Wu, Sylvia Tan, Cuiling Sun, Sulagna Saha, Yun-Hsuan Su, Kevin Huang
AIM, 2024
Paper

Implemented image migration frameworks (Dynamic Flow and Probability-Based Flow) for segmenting surgical tools in endoscopic images. Efficiently generated data structures and figures, and calculated Dice coefficients to assess segmentation accuracy.

Miscellanea

Posters

 Using Machine Learning To Classify Meteorite Reflectance Spectra
Classifying bird species through audio files using bioacoustic representation learning
Collective Behavior of Robot Swarms

Projects

 wikidata-diff-analyzer(published in rubygems website with 2000+ downloads and merged in WikiEducation Dashboard)
Music Genre Classification
Interactive Leaf Doctors: Segmentation and Health Prediction
Play with Prims and Kruskal

Blogs

 Toward Open-Source thru Outreachy
Behind the Code: Lessons from Building a Ruby Gem Library

Teaching

Teaching Assistant for PHYS-110 Force, Motion, and Energy, COMSC-151 Intro to CS, COMSC-161 OOP, COMSC-205 Data Structures, COMCSC-341 Computer Vision, COMSC-335 Machine Learning

@Mount Holyoke College(2022-2025)

Inspo and adapted from Jon Barron's source code.