About
Hello, I'm Sanjan.#
I am a final-year B.Tech student in Computer Science & Engineering at the Institute of Engineering & Management, Kolkata (CGPA 9.33/10, top 15%). My research sits at the intersection of representation and transfer learning, trustworthy ML (XAI + robustness), online learning under distribution shift, and computer vision — with applications in precision agriculture and remote sensing.
I currently work as a Research Intern at the University of Nebraska–Lincoln (supervised by Dr. Sruti Das Choudhury) and as a Research Scholar at the University of Calcutta. I have been featured in UNL's news coverage for my research contributions.
Representation / Transfer Learning
Trustworthy ML (XAI + Robustness)
Online Learning under Distribution Shift
Computer Vision (Hyperspectral / Multimodal)
9.33
CGPA
8
Publications
3+
Years Research
4
Research Roles
Education
Academic Background#
Institute of Engineering & Management (IEM), Kolkata
B.Tech in Computer Science & Engineering · CGPA: 9.33/10
Aug 2022 – Jul 2026 (Expected)
Ranked in the top 15% of class by CGPA.
Research Experience
Where I've Worked#
Research Intern — University of Nebraska, Lincoln, USA
Supervisor: Dr.
Sruti Das Choudhury
Jun 2025 – Present
- Co-authored a human-centered XAI study combining clustering, SHAP-driven interpretability, and narrative visualization on agronomic and healthcare datasets.
- Built an interactive visual-analytics pipeline on UNL greenhouse phenotyping data (42 plants, 9 genotypes, 25 days) coupling temporal embeddings, DTW-based clustering, and SHAP/LIME-linked causal views.
- Engineered HyperProbe, a lightweight Streamlit-based human-in-the-loop hyperspectral analysis tool (517–1700 nm, 243 bands).
- Featured in the university's news story (August 2025).
Research Scholar — University of Calcutta
Jan 2025 – Dec 2025
- Developed FH-FAM, a fuzzy-hypergraph feature-selection algorithm achieving 81.43% accuracy with 89.28% feature reduction across 15 datasets.
- Proposed SIF-HFAM, a strong intuitionistic fuzzy hypergraph framework with greedy (1−1/e)-approximation guarantee, achieving ~78% accuracy while removing ~98% of features.
Student Research Lead — Generative AI CoE, IEM
Nov 2024 – Present
- Led end-to-end research execution and operations — mentored projects, drove 10+ journal teams, managed CoE website, and spearheaded ReelBook (Pearson collaboration).
Project Intern — bair.ai (IEM Research Foundation)
Aug 2024 – Mar 2025
- Built MemeMetric, an end-to-end cluster-based cryptocurrency forecasting system with automated reporting and real-time social-media sentiment signals via NLP.
Undergraduate Research Assistant — IEDC (CSE)
Mar 2024 – Aug 2024
- Co-authored an IEM-HEALS 2024 paper on pharma stock analysis and built TraderBot, a Flask + MongoDB real-time trading simulator.
Study Abroad — National University of Singapore (NUS)
Jul 2023
- Studied AI, IoT, Machine Learning & Data Analytics — lectured by Dr. Peter Leong, Dr. Eric Cambria, and others.
Publications
Selected Papers#
Published / Accepted
1
ReproPheno and ReproPhenoNet: A Large-Scale Multimodal Benchmark Dataset
and Deep Learning Framework for Reproductive-Stage Plant Phenotyping
AAAI AgriAI Workshop, 2026
2
PlantPhenoLM: Phenotype-Genotype Mapping Inference with Multi-Turn LLM
Reasoning and Selective Prediction
3
Conversation as Belief Revision: GreedySAT Revision for Global Logical
Consistency in Multi-Turn LLM Dialogues
4
Fuzzy Hypergraph Feature Association Map for High-Dimensional Feature
Selection in Agriculture and Remote Sensing
5
Enhancing Interpretability Through Clustering, Explainable AI, and
Narrative Visualization
Information, 2025
6
Machine Learning-Driven Insights For Stock Market Analysis And Trading
7
The COVID-19 Shock: An Analysis Of Impacts And Responses Of Indian Stock
Market
8
Submitted / Under Review
9
Greedy Optimization with Provable Guarantees for Non-Uniform
Intuitionistic Hypergraph-Based Feature Selection
Pattern Recognition, 2026 (under review)
10
MiQ-MCP: Valid and Conditionally Robust Uncertainty Quantification for
High-Frequency Financial Time Series via Mondrian Conformalized Quantile Regression
Computational Economics, 2025 (under review)
Projects
Research Projects#
Geodesic Optimal Transport (Transfer Geometry)
- Implemented sliced-Wasserstein OT diagnostics on frozen ResNet-18 features; benchmarked across 48 transfer settings (CIFAR-10/STL-10/SVHN).
- Strong correlations with zero-shot transfer (Pearson r ≈ −0.71) and low-data adaptation (Spearman ρ ≈ 0.60 at 200-shot).
Grokking + LoRA (Low-Rank Tax)
- Controlled modular-addition experiments (p=97) comparing full-parameter vs. LoRA-on-frozen-base training for 15k epochs.
- Reproduced classic grokking (99% train → 99% val) and quantified rank/LR thresholds.
Volatility-Scaled AdaGrad (Online Learning)
- Implemented VS-AdaGrad, a CPU-efficient drift-aware online optimizer scaling AdaGrad using discounted residual volatility.
- Reduced regret proxy vs. AdaGrad by 18.4% (small drift) and 19.8% (medium drift); outperformed tuned OGD by 23.7–63.8%.
Skills
Technical Toolkit#
Programming
ML / AI
XAI
Data
Tools
Cloud
Contact
Get in Touch#
I'm always open to research collaborations and discussions. Feel free to reach out!