Low-loss connection of weight vectors: distribution-based approaches
Authors: Ivan Anokhin, Dmitry Yarotsky
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we test experimentally the proposed connection methods on the datasets CIFAR10 and MNIST. For each method, we measure the worst accuracy that the method provides along the path.1 For both datasets, we use the standard train test split. |
| Researcher Affiliation | Academia | 1Skolkovo Institute of Science and Technology, Moscow. |
| Pseudocode | No | The paper describes methods and procedures in narrative text but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | We release source code at https://github.com/ avecplezir/distribution-based-connectivity |
| Open Datasets | Yes | In this section, we test experimentally the proposed connection methods on the datasets CIFAR10 and MNIST. |
| Dataset Splits | No | The paper mentions 'standard train test split' but does not explicitly specify a validation split or its details. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, processor types, memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper mentions some software (e.g., POT library, Real NVP model, IAF model) but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | All considered models were trained using the cross-entropy loss with the SGD optimizer, for 400 epochs and 30 epochs on CIFAR10 and MNIST, respectively, with learning rate 0.01 and batch size 128. For CIFAR10 we use the same standard data augmentation as (Huang et al., 2017). For MNIST we do not use any augmentation. The activation function in all the networks is ReLU. |