software

FlowBench

PyTorch benchmark for reproducible evaluation of diffusion and flow-matching models under Gaussian and heavy-tailed generative settings.

Artifacts: benchmark description available on request

Keywords: diffusion models, flow matching, heavy-tailed distributions, reproducible benchmarks

Metis

AutoML toolbox for model selection, ensembling, and predictive-performance optimization.

Artifacts: description available on request

Keywords: AutoML, model selection, ensembling, predictive optimization

optimal_bootstrap

Code for high-dimensional analysis of bootstrap ensemble classifiers. This project accompanies the AISTATS work on bootstrap methods for LSSVM ensembles, random matrix theory, and practical rules for selecting ensemble hyperparameters.

Artifacts: code, paper

Keywords: bootstrap ensembles, random matrix theory, high-dimensional statistics

Bandpy

Gym-compatible Python package for benchmarking single- and multi-agent bandit algorithms across synthetic and real-data environments.

Artifacts: code

Keywords: multi-agent bandits, benchmarking, Gym environments, reproducible experiments

HemoLearn

Python package for estimating the haemodynamic response function from resting-state or task fMRI BOLD signals using blind deconvolution.

Artifacts: code

Keywords: fMRI, haemodynamic response function, blind deconvolution, inverse problems

Carpet

Research code for solving 1D total-variation regularized optimization problems, including classical solvers and learnable unrolled optimization algorithms.

Artifacts: code, paper

Keywords: unrolled optimization, total variation, inverse problems, sparse regularization