Bayesian Optimization for Iterative Learning

Authors: Vu Nguyen, Sebastian Schulze, Michael Osborne

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the efficiency of our algorithm by tuning hyperparameters for the training of deep reinforcement learning agents and convolutional neural networks. Our algorithm outperforms all existing baselines in identifying optimal hyperparameters in minimal time.
Researcher Affiliation Academia Vu Nguyen University of Oxford vu@robots.ox.ac.uk Sebastian Schulze University of Oxford sebastian.schulze@eng.ox.ac.uk Michael A. Osborne University of Oxford mosb@robots.ox.ac.uk
Pseudocode Yes Algorithm 1 Bayesian Optimization with Iterative Learning (BOIL)
Open Source Code Yes We release our implementation at https://github.com/ntienvu/BOIL.
Open Datasets Yes We consider three DRL settings including a Dueling DQN (DDQN) [46] agent in the Cart Pole-v0 environment and Advantage Actor Critic (A2C) [25] agents in the Inverted Pendulum-v2 and Reacher-v2 environments. In addition to the DRL applications, we tune 6 hyperparameters for training a convolutional neural network [21] on the SVHN dataset and CIFAR10.
Dataset Splits No The paper mentions using specific datasets like Cart Pole-v0, Inverted Pendulum-v2, Reacher-v2, SVHN, and CIFAR10, but it does not explicitly provide details about training, validation, or test splits (e.g., percentages, sample counts, or specific split methodologies) in the main text.
Hardware Specification Yes All experiments are executed on a NVIDIA 1080 GTX GPU using the tensorflow-gpu Python package.
Software Dependencies No The paper mentions 'tensorflow-gpu Python package' but does not specify a version number for TensorFlow or any other software dependency, which is required for reproducibility.
Experiment Setup Yes We set the maximum number of augmented points to be M = 15 and a threshold for a natural log of GP condition number δ = 20.