Assignment submissions of the semester 2020-21-II offering of CS220 at IIT Kanpur

Team Members

Robust image classification under distribution shift using WideResNet-28-10, SWA, Mixup/CutMix, Test-Time Augmentation, and Pseudo-Label Refinement.

MILP-based supply network design: optimal depot selection & flow routing across 2 factories, 6 depots, 6 customers using Gurobi. | IIT Kanpur OR Course Project

Developed a linear model for a XOR-PUF system. Here the XOR-PUF consists of three PUFs, each with eight multiplexers, and takes an 8-bit challenge as input. The goal is to create a new feature vector from the challenge and use a linear model using gradient descent to predict the PUF's response very accurately.

Real-time train weighbridge analysis with OCR resolver and signal debouncing — zero external ML dependencies

VR Memory Palace — A Unity 6 spatial learning app using the Method of Loci to teach Chemical Engineering concepts through a navigable 3D apartment. Built at IIT Kanpur.

Personal homepage: http://pages.cs.wisc.edu/~abhayk/ OR

Developed a computer program to automatically solve traditional CAPTCHA challenges by identifying characters in obfuscated images using image processing and machine learning.

Rank 3 INFORMS RAS 2025 Kaggle | LightGBM multi-class classifier on 20M+ railway sensor readings. Log Loss: 0.0283 | IIT Kanpur

📘 This repository contains assignments (2020, 2025 & 2026), official lecture notes, lecture transcripts and a final exam review & analysis from the Jan-Apr 2026 session for the Fuzzy Sets, Logic and Systems & Applications course offered by IIT Kanpur in collaboration with SWAYAM-NPTEL.

🚀 A 20-day focused DSA prep journal inspired by Goku’s SSJ3 mindset — solving core problems from EPI and LeetCode before IIT Kanpur. Track. Code. Evolve.

Diffusion-based photorealistic face relighting from synthetic data — 62,775 renders, HDRI-conditioned UNet, 60ms inference

Memory-efficient C++17 word-frequency indexer with buffered I/O, versioning, and three query types

MILP-based VRPTW with installation scheduling, solved to exact optimality using IBM CPLEX (Branch-and-Cut + MTZ constraints). 15 benchmark instances solved. | IIT Kanpur OR Course Project

End-to-end avocado pricing pipeline: EDA → demand regression (R²≈0.90) → Gurobi MILP optimization. ~35% revenue uplift vs historical benchmarks. | IIT Kanpur