Far-HO: A Bilevel Programming Package for Hyperparameter Optimization and Meta-Learning
This provides a unified tool for researchers and practitioners in machine learning to efficiently handle gradient-based hyperparameter tuning and meta-learning tasks.
The authors introduced Far-HO, a TensorFlow-based software package that tackles hyperparameter optimization and meta-learning using a bilevel programming framework, enabling optimization of learning rates, loss weighting, and hyper-representations.
In (Franceschi et al., 2018) we proposed a unified mathematical framework, grounded on bilevel programming, that encompasses gradient-based hyperparameter optimization and meta-learning. We formulated an approximate version of the problem where the inner objective is solved iteratively, and gave sufficient conditions ensuring convergence to the exact problem. In this work we show how to optimize learning rates, automatically weight the loss of single examples and learn hyper-representations with Far-HO, a software package based on the popular deep learning framework TensorFlow that allows to seamlessly tackle both HO and ML problems.