Online Control for Meta-optimization

Authors: Xinyi Chen, Elad Hazan

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Illustrative experimental results of our meta-optimization methods are given in Section 4.
Researcher Affiliation Collaboration Xinyi Chen Princeton University Google Deep Mind xinyic@princeton.edu Elad Hazan Princeton University Google Deep Mind ehazan@princeton.edu
Pseudocode Yes Algorithm 1 Meta-optimization; Algorithm 2 Gradient perturbation controller for meta-optimization
Open Source Code No The paper cites external libraries like JAX and DeepMind JAX Ecosystem, but does not provide a link or statement for the source code of its own methodology or implementation.
Open Datasets Yes For a proof-of-concept nonconvex task, we consider MNIST classification with a neural network.
Dataset Splits No The paper mentions running experiments with a certain number of episodes and epochs, and a batch size, but does not specify clear training, validation, or test dataset splits or percentages.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU/CPU models, memory) used for running its experiments, only the number of layers in the neural network.
Software Dependencies No The paper mentions frameworks like JAX in its references but does not provide specific version numbers for JAX or any other software dependencies crucial for replication.
Experiment Setup Yes For meta-optimization, we use L = 3, δ = 0, and a base learning rate of 0.001. We take L = 20, and the network is trained using stochastic gradients from batches of size 256. The hyperparameters of the baselines are tuned.