Welcome to my humble abode!
Researcher and Scientist
Hi, welcome! I'm Shashata, a part-time scientist and a full-time dreamer. I often find myself wondering about life's fascinating complexities—like what incredible lottery humanity won to gain "intelligence" unlike any other species on Earth. What intriguing mysteries unfold in our minds when we feel sad, excited, sick, or just not quite ourselves? Why do some memories stick with us forever, while others slip away? I want to find answers to these fascinating questions by zooming in as close as possible—down to single cells and ultramicroscopic details. I'm deeply involved in Connectomics, a field that maps the intricate wiring of the brain. When people say we're "wired" differently, they're actually onto something pretty accurate. I use machine intelligence to explore these puzzles. Right now, I'm in my third year of the EECS PhD program at MIT. I got a strong foundation in computer science and machine learning from MIT and my undergraduate institution at BUET. As you can see, I want to use my limited but growing knowledge to understand the brain better. Even though I truly want to understand the mind, my research is currently focused on solving the technical challenges of mapping it faster and more accurately.
I work with a cool science guy called Nir Shavit. Nir is a sweetheart, and a great computer scientist of his time. Anyway, a lot of my work still revolves around Interpretable Neural Modeling and Language Models, but nothing compares to the thrill I feel when spotting a natural neural connection under the microscope. I'd love to chat more about what I do, and if you're curious enough, maybe we can even collaborate! Just give me a ping.
Research
High-Throughput Connectomics
Connectomics research is notoriously slow, often requiring nearly a decade to reconstruct even a cubic millimeter of neural tissue. My work focuses on using AI to radically accelerate the image acquisition pipeline, aiming to make connectome reconstruction faster and more scalable.
Accurate Neural Reconstruction
Segmentation is crucial for connectome reconstruction but remains highly error-prone. I am developing more precise segmentation algorithms and workflows to enable accurate neural reconstructions and facilitate reliable brain-mapping efforts.
Biological Discoveries through Neural Structures
Even without a full connectome, high-resolution ultrastructural data can reveal fascinating biological insights. A significant part of my research involves crafting advanced Machine Learning methods to uncover neural organization and function from these detailed brain images.
Efficient and Interpretable Neural Modeling
Today's neural networks are often overparameterized, leaving ample room for compression without compromising performance. My research explores pruning, limits of compressibility, superposition, and interpretability in both language and vision models to create more efficient and transparent AI systems.
Computational Biology
Early in my career, I focused on phylogenomics—reconstructing the tree of life from genomic data—and used machine learning to analyze protein and genome sequences. I continue to engage in computational biology research, applying ML techniques to genomics whenever possible.
Publications
NeuroADDA: Active Discriminative Domain Adaptation in Connectomics
In review at ICCV 2025
Wasserstein Distances, Neuronal Entanglement, and Sparsity
ICLR 2025 Spotlight
PRIEST: predicting viral mutations with immune escape capability of SARS-CoV-2 using temporal evolutionary information
Briefings in Bioinformatics 25, no. 3 (2024): bbae218
A connectomics-driven analysis reveals novel characterization of border regions in mouse visual cortex
In review at Neural Networks
Quartet Based Gene Tree Imputation Using Deep Learning Improves Phylogenomic Analyses Despite Missing Data
Journal of Computational Biology, 29.11 (2022), 1156–72 (Also appeared in RECOMB 2022)
Memories
Family Time
My family is visiting me from Bangladesh spring 2025.
Hangout at Nir's Place
The lab generally hangout at Nir's place once or twice a year and talk about random science stuffs. It's fun!
Connectomics Conference at Berlin
My first Europe trip in Summer 2024. I went to Berlin to meet the connectomics world.
NeurIPS 2024 at Vancouver
My first major ML conference appearence. NeurIPS is so massive, ngl!
Funtime with Friends
I play a lot of table tennis (though pretty horrible still) with my friends Shuvom and Ling.
Curriculum Vitae
Education
Ph.D. in Computer Science
Massachusetts Institute of Technology, Cambridge, MA
Affiliations: MIT CSAIL
Research Interests: Connectomics, Efficient and Interpretable Neural Computation
M.Sc. in Computer Science
Massachusetts Institute of Technology, Cambridge, MA
B.Sc. in Computer Science and Engineering
Bangladesh University of Engineering and Technology, Dhaka, Bangladesh
Honors & Awards
HONDA Young Engineer and Scientist (Y-E-S) Award
Dean's Honor List
University Merit Scholarship
Notre Dame College Academic & Extra-curricular Award
Teaching
TA, 6.8610: Quantitative Methods for NLP
Massachusetts Institute of Technology
Lecturer
Bangladesh University of Engineering and Technology
Industry Experience
Machine Learning Research Intern
Neural Magic Inc, Somerville, MA
Machine Learning Engineer
Vertical Innovations Limited, Dhaka, Bangladesh
Contact
shashata@mit.edu
G580, Stata Center
32 Vassar Street
Cambridge, MA-02139