Stateful ODE-Nets using Basis Function Expansions
Authors: Alejandro Queiruga, N. Benjamin Erichson, Liam Hodgkinson, Michael W. Mahoney
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present empirical results to demonstrate the predictive accuracy and compression performance of Stateful ODE-Nets for both image classification and NLP tasks. Each experiment was repeated with eight different seeds, and the figures report mean and standard deviations. |
| Researcher Affiliation | Collaboration | Alejandro Queiruga Google Research afq@google.com N. Benjamin Erichson University of Pittsburgh erichson@pitt.edu Liam Hodgkinson ICSI and UC Berkeley liam.hodgkinson@berkeley.edu Michael W. Mahoney ICSI and UC Berkeley mmahoney@stat.berkeley.edu |
| Pseudocode | Yes | Algorithm 1 in the appendix describes the calculation of Eq. (10), fused into one loop. |
| Open Source Code | Yes | Research code is provided as part of the following Git Hub repository: https://github.com/afqueiruga/Stateful Ode Nets. Our models are implemented in Jax [3], using Flax [20]. |
| Open Datasets | Yes | We evaluate the performance on both MNIST [27] and CIFAR-10 [24]. ... The English-GUM treebank [47] from the Universal Dependencies treebanks [36]. |
| Dataset Splits | No | The paper does not explicitly provide specific validation dataset split information (e.g., percentages, sample counts, or explicit mention of a validation set). |
| Hardware Specification | No | The paper mentions support from |
| Software Dependencies | No | Our models are implemented in Jax [3], using Flax [20]. The paper mentions software but does not specify version numbers for Jax or Flax. |
| Experiment Setup | Yes | Details about the training process, and different model configurations are provided in Appendix E. Each experiment was repeated with eight different seeds, and the figures report mean and standard deviations. ... Our model uses an embedding size of 128, with Key/Query/Value and MLP dimensions of 128 with a single attention head. The final Ode Block has K = 64 piecewise constant basis functions and takes NT = 64 steps. |