The Symmetric Generalized Eigenvalue Problem as a Nash Equilibrium

Authors: Ian Gemp, Charlie Chen, Brian McWilliams

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically we demonstrate that this resulting algorithm is able to solve a variety of SGEP problem instances including a large-scale analysis of neural network activations.
Researcher Affiliation Industry imgemp@deepmind.com Charlie Chen ccharlie@deepmind.com Brian Mc Williams Google Research Z urich, Switzerland bmcw@google.com
Pseudocode Yes Algorithm 1 Deterministic / Full-batch γ-Eigen Game; Algorithm 2 Stochastic γ-Eigen Game
Open Source Code Yes A Jax implementation is available at github.com/deepmind/eigengame.
Open Datasets Yes We replicate a synthetic experiment from scikit-learn(Pedregosa et al., 2011) and compare Algorithm 2 to several approaches... loading minibatches of CIFAR-10 images, running them through a deep convolutional network, harvesting the activations, and then passing them to our distributed γ-Eigen Game solver.
Dataset Splits No The paper mentions using minibatches and datasets like CIFAR-10, but it does not specify explicit training, validation, or test dataset splits (e.g., 80/10/10 percentages or sample counts).
Hardware Specification Yes Figure 4 demonstrates our approach (parallelized over 8 TPU chips)
Software Dependencies No The paper mentions using 'Jax', 'Scipy's linalg.eigh(A, B)', and 'scikit-learn', but it does not provide specific version numbers for these software components.
Experiment Setup Yes Hyperparameters are listed in Appx. H.