Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Latent Mixture of Symmetries for Sample-Efficient Dynamic Learning
Authors: Haoran Li, CHENHAN XIAO, Muhao Guo, Yang Weng
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Numerical experiments in diverse physical systems demonstrate that Latent Mo S outperforms state-of-the-art baselines in interpolation and extrapolation tasks while offering interpretable latent representations suitable for future geometric and safety-critical analyses. |
| Researcher Affiliation | Academia | Haoran Li Chenhan Xiao Muhao Guo Yang Weng School of Electrical, Computer and Energy Engineering Arizona State University, Tempe, AZ 85281 EMAIL |
| Pseudocode | Yes | Algorithm 1 Multi-level Latent Sequence Generation |
| Open Source Code | Yes | Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: we will submit code and data as supplementary materials. After the review process, we will upload the code to Github. |
| Open Datasets | Yes | We test the following datasets, fully described in Appendix B. (1) Complex Nonlinear ODE Systems... (2) Residential Electricity Consumption... publicly available at [51 53]. (3) Photovoltaic Solar Energy... publicly available Photovoltaic (PV) dataset [54]... (5) Air Quality System. UCI Repository provides measurements... [58]. (6) Electrocardiogram (ECG) signals... UCR Time Series Archive [59]. |
| Dataset Splits | Yes | We adopt a standard formulation consistent with prior work on continuous-time and irregularly sampled time series, such as Latent ODE, Contiformer, RNNt, etc. The setup is as follows: (1) Interpolation. Given a time series with time points (t0, , t N), we condition on the subset of points from (t0, , t N) with a data drop rate (30%, 60%, 90% in Experiments) and reconstruct the full set of points in the same time interval. (2) Extrapolation. We split the time series into two parts (t0, , t N/2) and (t N/2, , t N). We input the first half of the time series and predict the second half. |
| Hardware Specification | Yes | Computing infrastructure. All experiments were run in Python 3.12 on a Mac machine equipped with an Intel Core i5 processor (3.1 GHz) and 8 GB of RAM. |
| Software Dependencies | No | Computing infrastructure. All experiments were run in Python 3.12 on a Mac machine equipped with an Intel Core i5 processor (3.1 GHz) and 8 GB of RAM. |
| Experiment Setup | Yes | The specific architectures and hyper-parameters are described as follows. Encoder. Similar to Latent ODE [18], we employ an ODE-RNN as the encoder... we set the GRU module... two hidden layers with the hidden unit number to be m... ODE solver used to solve the ODE-RNN is the fourth-order Runge Kutta method ("rk4")... Activation function and learning rate. We utilize Tanh as the activation function... We set the learning rate to be 0.001... Hyper-parameters We establish two important hyper-parameters for different datasets: the dimension of the latent space m and the number of experts K0 for non-zero outputs. Specifically, they are shown in the following table. |