Controlling the Complexity and Lipschitz Constant improves Polynomial Nets
Authors: Zhenyu Zhu, Fabian Latorre, Grigorios Chrysos, Volkan Cevher
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The theoretical results enable us to propose a principled regularization scheme that we also evaluate experimentally in six datasets and show that it improves the accuracy as well as the robustness of the models to adversarial perturbations. |
| Researcher Affiliation | Academia | EPFL, Switzerland {[first name].[surname]}@epfl.ch |
| Pseudocode | Yes | Algorithm 1: Projected SGD; Algorithm 2: Projected SGD + Adversarial Training |
| Open Source Code | No | The paper does not include an explicit statement or link indicating the release of open-source code for the methodology described. |
| Open Datasets | Yes | We conduct experiments on the popular datasets of Fashion MNIST (Xiao et al., 2017), E-MNIST (Cohen et al., 2017) and CIFAR-10 (Krizhevsky et al., 2014). |
| Dataset Splits | No | The paper specifies the sizes for training and test sets for various datasets (e.g., Fashion-MNIST: training set consists of 60,000 examples, and the test set of 10,000 examples), but it does not explicitly mention a separate validation set or its split details. |
| Hardware Specification | No | The paper does not specify the exact hardware (e.g., CPU, GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions the use of "SGD to optimize all the models", but it does not provide specific version numbers for any software dependencies, libraries, or frameworks used in the experiments. |
| Experiment Setup | Yes | Unless mentioned otherwise, all models are trained for 100 epochs with a batch size of 64. The initial value of the learning rate is 0.001. After the first 25 epochs, the learning rate is multiplied by a factor of 0.2 every 50 epochs. The SGD is used to optimize all the models, while the cross-entropy loss is used. The projection is performed every ten iterations. |