Portfolio
⚠️ Some links may be missing. I’m in the process of migrating certain projects from (internal) github.iu.edu to (external) github.com ⚠️
Published Research Projects
- What Do Hebbian Learners Learn? Reduction Axioms for Iterated Hebbian Learning. With Saul Blanco and Lawrence Moss. AAAI, Feb 2024.
Summary: TODO
pdf | talk | poster | github
Technologies used: Lean 4, formal verification, neural networks, mathematical logic - The Logic of Hebbian Learning. With Saul Blanco and Lawrence Moss. FLAIRS (Florida AI Research Society), May 2022.
Summary: TODO
pdf | talk | github
Technologies used: Python, pyparsing, networkx, numpy, scipy, matplotlib, neural networks, mathematical logic - Logics for Sizes with Union or Intersection. With Saul Blanco, Alex Kruckman, and Lawrence Moss. AAAI, Feb 2020.
Summary: TODO
pdf | talk | github
Technologies used: Python, prover9 - CBR Confidence as a Basis for Confidence in Black Box Systems. With Larry Gates (first author) and David Leake. ICCBR, Sep 2019.
Summary: TODO
pdf | github
Technologies used: TODO
Other Software Projects
These are various projects I did throughout college and my PhD. For one reason or another, I didn’t try to make them into publishable papers. However, they were significant projects that showcase work that I am proud of!
- Verified Neural Network Model Builder (In Progress)
An implementation of my PhD dissertation project. A program that, when finished, will build a neural network that is formally verified to obey a set of user-specified constraints on its learning and inference.
github
Technologies used: Lean 4, mathlib, formal verification, neural networks - Neural Networks Augmented with CBR-Style Adaptation
TODO
report | github
Technologies used: Python, sklearn, numpy - Neural Network Adaptation Audit
A carefully designed experiment to test whether trained neural networks learn case-based adaptation (which is a kind of careful reasoning).
report | jupyter notebook
Technologies used: Python, sklearn, numpy, neural networks - Syllogistic Entailment for Vector Semantics
TODO
report | github
Technologies used: Python, word2vec, NLTK, sklearn, numpy, scipy - Notakto Player
A program that uses reinforcement learning to learn winning strategies for the game of Notakto (a particularly hard variant of Tic-Tac-Toe).
report | github
Technologies used: Python, tensorflow, numpy, pandas, matplotlib, convolutional neural networks - Sense-Able
A proof-of-concept LIDAR obstacle sensor for visually impaired people. This was my senior team project at USC, in collaboration with a team of other students and our client (P. B. Mumola, LLC). I personally wrote extra documentation for the SDK as part of my Technical Writing class project.
github | docs | demo
Technologies used: C++, Qt, OpenCV, LIDAR hardware and SDK