Dataset Dynamics via Gradient Flows in Probability Space
Authors: David Alvarez-Melis, Nicolò Fusi
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Through various experiments, we show that this framework can be used to impose constraints on classification datasets, adapt them for transfer learning, or to re-purpose fixed or blackbox models to classify with high accuracy previously unseen datasets. and 7. Experiments We first evaluate our approach for imposing constraints on low-dimensional synthetic datasets (Section 7.1) and then on two settings (Sections 7.2 & 7.3) involving transfer learning with benchmark image classification datasets. |
| Researcher Affiliation | Industry | David Alvarez-Melis 1 Nicol o Fusi 1 1Microsoft Research. Correspondence to: David Alvarez Melis <alvarez.melis@microsoft.com>. |
| Pseudocode | No | The paper describes the numerical solution of gradient flows using mathematical equations (16) and (17) and explanatory text, but does not present a formal pseudocode or algorithm block. |
| Open Source Code | No | The paper does not contain an explicit statement about releasing source code for the described methodology or a link to a code repository. |
| Open Datasets | Yes | We consider four classification datasets: MNIST (Le Cun et al., 2010), USPS, FASHIONMNIST (Xiao et al., 2017) and KMNIST (Clanuwat et al., 2018)... In addition to the *NIST datasets, we use CIFAR10, STL10 and the CAMELYON histopathology dataset (Litjens et al., 2018). |
| Dataset Splits | Yes | We use the standard MNIST, FMNIST, KMNIST, USPS splits. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments (e.g., CPU/GPU models, memory specifications). |
| Software Dependencies | No | The paper mentions using PyTorch and the POT library but does not specify their version numbers, which is required for reproducibility. |
| Experiment Setup | Yes | Training is performed using ADAM with learning rate 1e-3, batch size 64, for 10 epochs. (from Appendix C). |