Transfer Learning via Minimizing the Performance Gap Between Domains

Authors: Boyu Wang, Jorge Mendez, Mingbo Cai, Eric Eaton

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental evaluation on benchmark data sets shows that gap Boost significantly outperforms previous boosting-based transfer learning algorithms.
Researcher Affiliation Academia Boyu Wang Department of Computer Science University of Western Ontario bwang@csd.uwo.ca Jorge A. Mendez Department of Computer and Information Science University of Pennsylvania mendezme@seas.upenn.edu Ming Bo Cai Princeton Neuroscience Insititute Princeton University mcai@princeton.edu Eric Eaton Department of Computer and Information Science University of Pennsylvania eeaton@seas.upenn.edu
Pseudocode Yes Algorithm 1 gap Boost
Open Source Code Yes Source code for gap Boost is available at https://github.com/bwang-ml/gap Boost.
Open Datasets Yes 20 Newsgroups This data set contains approximately 20,000 documents, grouped by seven top categories and 20 subcategories. The source and target data sets were in the same way as in [10], yielding 6 transfer learning problems. Office-Caltech This data set contains approximately 2,500 images from four distinct domains: Amazon (A), DSLR (D), Webcam (W), and Caltech (C)
Dataset Splits No The paper mentions using a 'training sample' and 'testing' data, but does not explicitly describe a validation set or how data was split for validation.
Hardware Specification No The paper does not provide any specific details about the hardware used for running the experiments (e.g., GPU/CPU models, memory specifications).
Software Dependencies No The paper mentions 'Logistic regression is used as the base learner' but does not specify any software dependencies with version numbers (e.g., specific Python libraries or frameworks with their versions).
Experiment Setup Yes The hyper-parameters of gap Boost were set as γmax = 1 NT as per Remark 5, ρT = 0, which corresponds to no punishment for the target data, and ρS = log 1. Logistic regression is used as the base learner for all methods, and the number of boosting iterations is set to 20. In both data sets we pre-processed the data using principal component analysis (PCA) to reduce the the feature dimension to 100.