Sample-Then-Optimize Batch Neural Thompson Sampling

Authors: Zhongxiang Dai, YAO SHU, Bryan Kian Hsiang Low, Patrick Jaillet

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we verify their empirical effectiveness using practical Auto ML and reinforcement learning experiments. We demonstrate our empirical effectiveness in real-world experiments including automated machine learning (Auto ML) and reinforcement learning (RL) tasks, as well as a task on optimization over images.
Researcher Affiliation Academia Dept. of Computer Science, National University of Singapore, Republic of Singapore Dept. of Electrical Engineering and Computer Science, MIT, USA
Pseudocode Yes Algorithm 1 STO-BNTS Algorithm 2 STO-BNTS-Linear
Open Source Code Yes Our code is available at https://github.com/daizhongxiang/sto-bnts.
Open Datasets Yes For the RF and XGBoost hyperparameter tuning tasks, we use the datasets for diabetes diagnosis and MNIST, respectively, from OpenML [63, 64]. For the CNN hyperparameter tuning tasks, we use the MNIST dataset. All these datasets are publicly available from OpenML.
Dataset Splits Yes We use a random 80/10/10 train/validation/test split for the diabetes diagnosis and MNIST datasets.
Hardware Specification Yes All experiments are run on a cluster with NVIDIA A100 GPUs.
Software Dependencies No All algorithms are implemented in Python and PyTorch. (No version numbers provided)
Experiment Setup Yes For all methods, we use an NN architecture with a depth of L = 8 and a width of m = 64 unless specified otherwise. We use L = 2, m = 256 for all methods.