OptEx: Expediting First-Order Optimization with Approximately Parallelized Iterations

Authors: Yao Shu, Jiongfeng Fang, Ying He, Fei Yu

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we provide extensive empirical studies, including synthetic functions, reinforcement learning tasks, and neural network training on various datasets, to underscore the substantial efficiency improvements achieved by Opt Ex in practice.
Researcher Affiliation Academia Yao Shu# , Jiongfeng Fang , Ying Tiffany He , Fei Richard Yu Guangdong Lab of AI and Digital Economy (SZ), China College of Computer Science and Software Engineering, Shenzhen University, China School of Information Technology, Carleton University, Canada
Pseudocode Yes Algorithm 1: Opt Ex
Open Source Code Yes Our implementation is available at https://github.com/youyve/Opt Ex.
Open Datasets Yes Open AI Gym suite [38], CIFAR-10 [41], MNIST [45], Fashion MNIST [46], Shakespeare Corpus, Harry Potter and the Sorcerers Stone
Dataset Splits No The paper mentions using standard datasets like CIFAR-10 and MNIST, but does not explicitly state the training/validation/test splits used for these datasets.
Hardware Specification Yes The wallclock time is evaluated on a single NVIDIA RTX 4090 GPU. The wallclock time is evaluated on an AMD EPYC 7763 CPU.
Software Dependencies No The paper mentions using Deep Q-Network (DQN), Open AI Gym, and jax.vmap, but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes all the baselines...are based on Adam [4] with a learning rate of 0.1, β1 = 0.9, and β2 = 0.999. Hyperparameters, including a learning rate of 0.001, a reward discount factor of 0.95, and a batch size of 256, are applied. based on SGD [1] with a learning rate of 0.001, a batch size of 512.