Understanding Self-Training for Gradual Domain Adaptation

Authors: Ananya Kumar, Tengyu Ma, Percy Liang

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Leveraging the gradual shift structure leads to higher accuracies on a rotating MNIST dataset, a forest Cover Type dataset, and a realistic Portraits dataset. We run experiments on three datasets (see Appendix C for more details): Rotating MNIST:..., Cover Type:..., Portraits:...
Researcher Affiliation Academia Ananya Kumar 1 Tengyu Ma 1 Percy Liang 1 1Stanford University, Stanford, California, USA. Correspondence to: Ananya Kumar <ananya@cs.stanford.edu>.
Pseudocode No The paper describes the algorithm using equations and descriptive text but does not include a formally structured pseudocode or algorithm block.
Open Source Code Yes All code, data, and experiments can be found on Coda Lab at https: //bit.ly/gradual-shift-codalab, code is also on Git Hub at https://github.com/p-lambda/ gradual_domain_adaptation.
Open Datasets Yes Rotating MNIST: Rotating MNIST is a semi-synthetic dataset where we rotate each MNIST image by an angle between 0 and 60 degrees. Cover Type: A dataset from the UCI repository where the goal is to predict the forest cover type at a particular location given 54 features (Blackard and Dean, 1999). Portraits: A real dataset comprising photos of high school seniors across years (Ginosar et al., 2017).
Dataset Splits Yes We split the 50,000 MNIST training set images into a source domain (images rotated between 0 and 5 degrees), intermediate domain (rotations between 5 and 60 degrees), and a target domain (rotations between 55 degrees and 60 degrees).
Hardware Specification No The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running experiments.
Software Dependencies No The paper mentions software components like 'neural networks' and 'logistic regression', but does not provide specific version numbers for any libraries, frameworks, or programming languages used in the experiments.
Experiment Setup Yes For rotating MNIST and Portraits we used a 3-layer convolutional network with dropout(0.5) and batchnorm on the last layer... For the Cover Type dataset we used a 2 hidden layer feedforward neural network with dropout(0.5) and batchnorm on the last layer... For each step of self-training, we filter out the 10% of images where the model s prediction was least confident