Works

Fairness is Not Flat: Geometric Phase Transitions Against Shortcut Learning

[RESEARCH][AI][PAPER]

A geometric method to fight shortcut learning in neural networks: a tiny single-neuron 'auditor' exposes the spurious features a model would otherwise exploit (like Capital-Gain in the Adult Census dataset), and pruning them forces the network into a higher-capacity regime where it learns fairer, merit-based decisions. The approach cuts counterfactual gender bias from 21.18% to 7.66%, outperforms L1 regularization, and runs at a fraction of the cost of methods like Just Train Twice.

> Read Paper

Performance and Storage Analysis of CRYSTALS Kyber as a Post Quantum Replacement for RSA and ECC

[RESEARCH][CRYPTO][PAPER]

An analysis of the post-quantum algorithm CRYSTALS-Kyber, testing its speed and efficiency against RSA and ECC. The results show Kyber is significantly faster at key generation and shared secret derivation on standard CPUs (Intel and ARM), confirming it is a practical, immediate solution for the post-quantum era.

> Read Paper