Almost Surely Stable Deep Dynamics

Authors: Nathan Lawrence, Philip Loewen, Michael Forbes, Johan Backstrom, Bhushan Gopaluni

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the utility of each approach through numerical examples. 6 Experiments
Researcher Affiliation Collaboration Nathan P. Lawrence Department of Mathematics University of British Columbia Philip D. Loewen Department of Mathematics University of British Columbia Michael G. Forbes Honeywell Process Solutions Johan U. Backström Backstrom Systems Engineering Ltd. R. Bhushan Gopaluni Department of Chemical and Biological Engineering University of British Columbia
Pseudocode No The paper describes algorithmic procedures in text but does not include structured pseudocode or algorithm blocks.
Open Source Code Yes The code for our methods is available here: https://github.com/NPLawrence/stochastic_dynamics.
Open Datasets No The paper describes generating training data from specified systems (e.g., system (16), Eq. (18) with discretization) but does not provide access details or specify a publicly available dataset.
Dataset Splits No The paper mentions 'training data' and 'initial conditions not seen during training' but does not provide specific details on how the dataset was split into training, validation, and test sets (e.g., percentages or counts).
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions using a 'fully connected feedforward neural network' and refers to PyTorch in its references, but it does not specify exact version numbers for PyTorch or other software dependencies.
Experiment Setup Yes For the deterministic case (Example 2), we used a 3-layer neural network with 64 units per layer and ReLU activations for ˆf. The Lyapunov function V was a 2-layer ICNN with 32 units per layer and ReLU activations. We used a learning rate of 1e-3 and trained for 1000 epochs. For the LNN case, we trained for 10000 epochs. We used a batch size of 64 and the Adam optimizer (Kingma and Ba, 2014). For the stochastic case, we used a 3-layer MDN with 64 units per layer and ReLU activations. The Lyapunov function V was the same as the deterministic case. We used a learning rate of 1e-4 and trained for 10000 epochs. We used a batch size of 64 and the Adam optimizer.