Reinforcement Learning and Control
Model-based Reinforcement Learning and Planning
Object-centric Self-supervised Reinforcement Learning
Self-exploration of Behavior
Causal Reasoning in RL
Equation Learner for Extrapolation and Control
Intrinsically Motivated Hierarchical Learner
Regularity as Intrinsic Reward for Free Play
Curious Exploration via Structured World Models Yields Zero-Shot Object Manipulation
Natural and Robust Walking from Generic Rewards
Goal-conditioned Offline Planning
Offline Diversity Under Imitation Constraints
Learning Diverse Skills for Local Navigation
Learning Agile Skills via Adversarial Imitation of Rough Partial Demonstrations
Combinatorial Optimization as a Layer / Blackbox Differentiation
Object-centric Self-supervised Reinforcement Learning
Symbolic Regression and Equation Learning
Representation Learning
Stepsize adaptation for stochastic optimization
Probabilistic Neural Networks
Learning with 3D rotations: A hitchhiker’s guide to SO(3)
Stepsize adaptation for stochastic optimization
Stochastic gradient methods are the driving force behind the recent boom of deep learning. As a result, the demand for practical efficiency as well as for theoretical understanding has never been stronger. Naturally, this has inspired a lot of research and has given rise to new and currently very popular optimization methods such as Adam, AdaGrad, or RMSProp, which serve as competitive alternatives to classical stochastic gradient descent (SGD).
However, the current situation still causes huge overhead in implementations. In order to extract the best performance, one is expected to choose the right optimizer, finely tune its hyperparameters (sometimes multiple), often also to handcraft a specific stepsize adaptation scheme, and finally combine this with a suitable regularization strategy. All of this, mostly based on intuition and experience.
We propose [
] a stepsize adaptation scheme for stochastic gradient descent.
It operates directly with the loss function and rescales the gradient in order to make fixed predicted progress on the loss.
We demonstrate its capabilities by conclusively improving the performance of Adam and momentum optimizers.
The enhanced optimizers with default hyperparameters
consistently outperform their constant stepsize counterparts, even the best ones,
without a measurable increase in computational cost.
The performance is validated on multiple architectures including dense nets, CNNs, ResNets, and the recurrent Differential Neural Computer on classical datasets MNIST, fashion MNIST, CIFAR10 and others. The results are presented at the Neural Information Processing Systems (NeurIPS) 2018. The code is available at github/martius-lab repository.
Members
Publications