Deep Stochastic Mechanics

Authors: Elena Orlova, Aleksei Ustimenko, Ruoxi Jiang, Peter Y. Lu, Rebecca Willett

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

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical simulations verify our theoretical findings and show a significant advantage of our method compared to other deep-learning-based approaches used for quantum mechanics.
Researcher Affiliation Collaboration 1Department of Computer Science, University of Chicago, Chicago, USA 2Share Chat, London, UK 3Department of Physics, University of Chicago, Chicago, USA 4Department of Statistics, University of Chicago, Chicago, USA.
Pseudocode Yes Algorithm 1 Training algorithm pseudocode
Open Source Code Yes The code of our experiments can be found on Git Hub 4. 4https://github.com/elena-orlova/deep-stochastic-mechanics
Open Datasets No The paper describes generating initial conditions and using numerical solutions as ground truth but does not provide concrete access information (link, DOI, citation with author/year) for a publicly available or open dataset used for training.
Dataset Splits No The paper mentions training procedures and evaluation against ground truth, but it does not specify explicit training/validation/test dataset splits or cross-validation setup details.
Hardware Specification Yes In our experiments, we usually use a single NVIDIA A40 GPU. For the results reported in Section 5.4, we use an NVIDIA A100 GPU.
Software Dependencies No The paper mentions using 'SCIPY library (Virtanen et al., 2020)', 'QMSOLVE library', and 'NETKET library (Vicentini et al., 2022)'. While these software tools are named, specific version numbers for their usage are not provided for reproducibility.
Experiment Setup Yes Further details on numerical solvers, architecture, training procedures, hyperparameters of our approach, PINNs, and t-VMC can be found in Appendix C. ... A basic NN architecture for our approach and the PINN is a feed-forward NN with one hidden layer with tanh activation functions. ... We represent the velocities u and v using this NN architecture with 200 neurons... We use the Adam optimizer (Kingma & Ba, 2014) with a learning rate 10 4. In our experiments, we set wu = 1, wv = 1, w0 = 1. ... The number of time steps is N = 1000, and the batch size B = 100.