Deep Learning Games

Authors: Dale Schuurmans, Martin A. Zinkevich

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To investigate the utility of these methods for supervised learning, we conducted experiments on synthetic data and on the MNIST data set [20].
Researcher Affiliation Collaboration Dale Schuurmans Google daes@ualberta.ca Martin Zinkevich Google martinz@google.com Work performed at Google Brain while on a sabbatical leave from the University of Alberta.
Pseudocode Yes Algorithm 1 Main Loop, Algorithm 2 Regret Matching (RM), Algorithm 3 Exp. Weighted Average (EWA), Algorithm 4 Projected SGD
Open Source Code No The paper does not provide an explicit statement or a link to open-source code for the methodology described.
Open Datasets Yes We conducted experiments on synthetic data and on the MNIST data set [20]. ... The third experiment was conducted on MNIST, which is an n = 10 class problem over m = 784 dimensional inputs with T = 60, 000 training examples, evidently not linearly separable.
Dataset Splits No The paper mentions 'training loss' and 'test loss' and refers to 60,000 training examples for MNIST, but it does not explicitly define a separate 'validation' split or its size, nor does it refer to a standard three-way split that includes validation.
Hardware Specification No The paper mentions a "Tensorflow implementation" but does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions "Tensorflow implementation" but does not specify its version number or any other software dependencies with their versions.
Experiment Setup Yes For this experiment, we used mini-batches of size 100. ... Here we chose the L1 constraint bound to be β = 10 and the initialization scale as σ = 100. For the nonlinear activation functions we used a smooth approximation of the standard Re LU gate fv(x) = τ log(1 + ex/τ) with τ = 0.5. ...RM was run with β = 30 and initialization scales (σ1, σ2, σ3) = (50, 200, 50). ...where RM was run with (β1, β2, β3, β4) = (30, 30, 30, 10) and initialization scales σ = 500.