/differential-ml

Tensorflow and Pytorch implementation of differential machine learning (https://arxiv.org/abs/2005.02347, by Brian Huge and Antoine Savine).

Primary LanguagePythonMIT LicenseMIT

Differential Machine Learning

Tensorflow and Pytorch implementation of differential machine learning (https://arxiv.org/abs/2005.02347, by Brian Huge and Antoine Savine). Differential Machine Learning (DML) is a regularization technique for neural networks that leverages the availability of derivatives of the training labels with respect to the training inputs. Those derivatives are called differential labels. DML forces the derivative of the neural network to be close to the differential labels.

Installing this repo

Pulling the latest version of DML

git clone git@github.com:tum-ai/differential-ml.git
cd differential-ml

Setting up the virtual environment with Python 3.11.0

Assumes a working installation of pyenv and poetry

pyenv install 3.11.0
pyenv virtualenv 3.11.0 differential-ml
pyenv local differential-ml

Installing dependencies

poetry install