eXponential FAmily Dynamical Systems (XFADS): Large-scale nonlinear Gaussian state-space modeling
Authors: Matthew Dowling, Yuan Zhao, Memming Park
NeurIPS 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In comparisons with other deep state-space model architectures our approach consistently demonstrates the ability to learn a more predictive generative model. Furthermore, when applied to neural physiological recordings, our approach is able to learn a dynamical system capable of forecasting population spiking and behavioral correlates from a small portion of single trials. |
| Researcher Affiliation | Academia | Matthew Dowling Champalimaud Research, Champalimaud Foundation, Portugal EMAIL Yuan Zhao National Institute of Mental Health, USA EMAIL Il Memming Park Champalimaud Research, Champalimaud Foundation, Portugal EMAIL |
| Pseudocode | Yes | Algorithm 1 End-to-end learning; Algorithm 2 Nonlinear variational filtering |
| Open Source Code | No | The paper does not provide a concrete link to open-source code for the methodology described. |
| Open Datasets | Yes | We considered two popular datasets: i) a pendulum system26 and ii) a bouncing ball27,28. ... We considered recordings from motor cortex of a monkey performing a reaching task30... We analyzed physiological recordings from the DMFC region of a monkey engaged in a timing interval reproduction task32. |
| Dataset Splits | Yes | We generate 500/150/150 trials of length 100 for training/validation/testing. All methods are trained for 5000 epochs for 3 different random seeds. ... For this dataset we take 500/150/150 trials of length 75 for training/validation/testing. ... For this dataset, we partitioned 1800/200/200 training/validation/testing trials sampled at 20ms per bin. ... For this dataset, we partitioned 700/150/150 training/validation/testing trials. |
| Hardware Specification | Yes | The system used for benchmarking wall-clock time was an RTX 4090 with 128GB of RAM with an AMD 5975WX processor. |
| Software Dependencies | No | The paper mentions 'Adam(lr = 0.001)' as the optimizer and neural network architectures like 'MLP' and 'GRU' but does not specify software dependencies with version numbers (e.g., Python, PyTorch/TensorFlow versions). |
| Experiment Setup | Yes | All methods are trained for 5000 epochs for 3 different random seeds. We consider a context window of 50 images and a forecast window of 50 images. ... optimizer: Adam(lr = 0.001) batch size: 128 |