Abhi Sridhar

Generating adapted neural-net parameters with diffusion models.

A novel application of diffusion models to adapt neural network parameters for handling data distribution shifts. Using a hypernetwork-based approach, diffusion models generate parameters tailored to target domains, demonstrated on image classifiers facing various corruptions. The method adapts to both known and unseen corruptions, offering a flexible alternative to traditional fine-tuning when target-domain data is scarce.

• Python
• PyTorch
• U-Net
• CIFAR-10

Q-learning and DQN agents that solve Bloxorz.

Uses Q-learning and Deep Q-Networks to train a reinforcement learning agent for the first ten levels of Bloxorz. Compares memory use, computational complexity, and convergence time between tabular Q-learning and DQN models. Finds that tabular Q-learning, especially with a by-level approach, is more efficient than DQNs for this discrete, state-space-constrained environment.

• OpenAI Gym
• PyTorch
• Pynput

Blockchain platform for healthcare insurance management.

A decentralized, transparent network to securely share patient data and automate insurance claims. Using smart contracts, ChainLife ensures efficient data handling and rewards users with CLIFE tokens for maintaining healthy lifestyles. The interface enhances healthcare quality and accessibility while protecting patient privacy.

• Solidity
• Web3.js
• Ganache
• Metamask
• Remix

Collision-sensing system for football helmets.

A durable, reliable collision-sensing system for football helmets designed to enhance player safety. Features real-time impact detection and analysis, easy integration into existing helmets, and competitive pricing. By improving safety measures, HitStick aims to support youth participation in contact sports.

• Flutter
• C++
• Firebase
• BluetoothLE