mattiasc01 Goto Github PK

entropic-optimization-algorithms

Studied a class of entropic optimization algorithms for deep learning, in the unsupervised learning setting. Found evidence that they improve generalization and denoising capabilities of autoencoders.

hierarchy-learning

Learning a hierarchical model of data with neural networks.

jepa

orchestrachess

UCI-compliant chess engine written in Rust. Reached 2300+ Elo rating on Lichess. Features a Pygame GUI.

potts-model-and-phase-transitions

Numerical study of the Potts model using Monte Carlo methods. Found evidence of a phase transition whose nature depends on the number of states, confirming analytical results.

robustness-of-minima-and-finetuning

In deep learning, investigate the impact on finetuning performance of the robustness of a pretrained checkpoint to random and adversarial perturbations of inputs and weight.

single-cell-sequencing-analysis

Analyze single-cell sequencing data and predict hypoxia/normoxia condition of cancerous cells. Presented at the Department of Oncology of Oxford University in July 2022.