okitouni Goto Github PK

algorithmic-sticking

Grokking (delayed generalization) on algorithmic datasets is ripe with interesting phenomena. For instance, algorithmic sticking – or the observation that some models prefer approximation of the data generating process over statistical parroting even in the presence of noise.

autogradle

Just a little sandbox implementation of automatic differentiation.

disco

dynamic-hinge

energyflow

energymover-dual

Energy Mover's Distance using KR duality

geometric-hep

General framework based on Graph Neural Networks for tracking, event classification, segmentation, clustering, and more.

grokkingeffectivetheory

hep-bnn

Bayesian neural network implementations for various High Energy Physics problems.

ids.131-dr

Demand Response project.

kbformer-tab-experiments

kbgen-weights

learning-symmetries

learning_without_training

lipschitz-network-bench

lipschitz-sgd

mitjuniorlab

Some Jupyter notebook examples for data analysis

mode

Moment Decorrelation for constrained NN outputs with respect to protected attributes. (Decorrelated (flat), linear, quadratic, etc.)

monotonenorm

Lipschitz normed layers for robust and monotonic architectures

my-website

newgan

nflows

Normalizing flows in PyTorch

nn-emd

nnn

n³ The unorthodox terminal file manager

nuclr

A foundation model for nuclear physics

nuclr-mechinterp

Mechanistic Interpretability of the Neural Co-learned Representations model.

okitouni.github.io

papermplstyle

problemcompilation

Collection of various problems.

pyjet

The interface between FastJet and NumPy