PACF Learning for Mortgage Approval Data in Vermont
- - Audited historical mortgage approval data in Vermont using PAC and metric-fair learning, demonstrating a possibly fair mortgage approval process.
- - Developed a novel similarity metric for individual fairness in Python.
- - Trained a linear classifier under a set of fairness constraints using PyTorch, showing that the constraints only cause a 2 percentage point decrease in accuracy.
https://github.com/albertqi/pacf-hmda2022
Verification of the X3DH Protocol in Tamarin
- - Formally proved the security guarantees of the X3DH key agreement protocol in Tamarin, verifying both forward secrecy and cryptographic deniability.
https://github.com/albertqi/x3dh-tamarin
Robust ResNet-18
- - Trained a ResNet-18 model on the CIFAR-10 dataset using PyTorch that is robust to various transformations.
- - Showed that robustness generalizes well to photometric transformations.
https://github.com/albertqi/robust-resnet18
Optimizing Distributed Training on Intermittently Connected Networks
- - Evaluated several strategies for optimizing distributing training on intermittently connected networks in Java and Python, including gossip learning and sparse weight aggregation.
- - Demonstrated that performing all-reduce every four iterations results in a 53.8% faster time-to-accuracy than standard all-reduce.
https://github.com/albertqi/odticn
DSML
- - Developed a distributed state management library using C++ to provide a seamless way to share state across multiple computers in a system while keeping network, memory, and computational overhead extremely low.
https://github.com/BeyondPerception/dsml
MiniML
- - Engineered a meta-circular interpreter in OCaml for a Turing-complete, ML-based language supporting atomic data types, lazy expressions, and lexically scoped environment semantics.
https://github.com/albertqi/miniml
Nobles
- - Developed a mobile app using Ionic and Angular to allow 500+ students at the Noble and Greenough school to view their schedule, directories, lunch menus, athletic results, personal information, and more.
https://github.com/albertqi/nobles-app