Neural Piecewise-Constant Delay Differential Equations
Authors: Qunxi Zhu, Yifei Shen, Dongsheng Li, Wei Lin9242-9250
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | With such a promotion, we show that the Neural PCDDEs do outperform the several existing continuous-depth neural frameworks on the one-dimensional piecewise-constant delay population dynamics and real-world datasets, including MNIST, CIFAR10, and SVHN. |
| Researcher Affiliation | Collaboration | Qunxi Zhu,1* Yifei Shen,2* Dongsheng Li,3 Wei Lin1,4 1 Research Institute of Intelligent Complex Systems, State Key Laboratory of Medical Neurobiology and MOE Frontiers Center for Brain Science, Institutes of Brain Science, Fudan University 2 Department of Electronic and Computer Engineering, The Hong Kong University of Science and Technology, Hong Kong 3 Microsoft Research Asia 4 School of Mathematical Sciences, Shanghai Center for Mathematical Sciences, Center for Computational Systems Biology, and LCNBI, Fudan University |
| Pseudocode | No | The paper describes mathematical formulations and computational steps but does not include any clearly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | No | The paper does not provide any statement about making the source code available or include any links to a code repository. |
| Open Datasets | Yes | We conduct experiments on several image datasets, including MNIST, CIFAR10, SVHN, by using the (unshared) NPCDDEs and the other baselines. |
| Dataset Splits | No | The paper mentions training data and test data but does not explicitly provide details about a validation dataset split or its size. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies or version numbers for libraries or tools used in the experiments. |
| Experiment Setup | Yes | For a fair comparison, we construct all models without augmenting the input space, and for the NDDEs, we assume that the initial function keeps constant (i.e., the initial function ϕ(t) = input for t 0), which is different from the initial function used for the NDDEs in (Zhu, Guo, and Lin 2021). We note that our models are orthogonal to these models, since one can also augment the input space and model the initial state as the feature of an NODE in the framework of NPCDDEs. Additionally, the number of the parameters for all models are almost the same (84k params for MNIST, 107k params for CIFAR10 and SVHN). Notably, the vector fields of all the models are parameterized with the convolutional architectures (Dupont, Doucet, and Teh 2019; Zhu, Guo, and Lin 2021), where the arguments that appeared in the vector fields are concatenated and then fed into the convolutional neural networks (CNNs). |