Strength from Weakness: Fast Learning Using Weak Supervision

Authors: Joshua Robinson, Stefanie Jegelka, Suvrit Sra

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our theoretical results are reflected empirically across a range of tasks and illustrate how weak labels speed up learning on the strong task.
Researcher Affiliation Academia 1Massachusetts Institute of Technology, Cambridge, MA 02139. Correspondence to: Joshua Robinson <joshrob@mit.edu>.
Pseudocode Yes Algorithm 1 Pretrain-finetune meta-algorithm
Open Source Code No The paper does not provide an explicit statement or link to open-source code for the methodology described.
Open Datasets Yes For our CIFAR-10 experiments...fine tuning on a small subset CIFAR-100...Databases left to right: MNIST, SVHN, and CIFAR-10. ...TREC fast-based question categorization dataset.
Dataset Splits No The paper describes a 'held out dataset' for training an auxiliary network and discusses 'generalization error', but it does not specify concrete train/validation/test split percentages or sample counts for its experiments.
Hardware Specification No The paper mentions that 'All image-based experiments use either a Res Net-18 or Res Net-34 for the weak feature map g (see Appendix C for full details)', suggesting hardware details might be in the appendix, but it does not provide specific hardware models (e.g., GPU, CPU) in the provided text.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies used in the experiments. While PyTorch is referenced, its version is not stated as a dependency for the authors' work.
Experiment Setup No The paper states 'All image-based experiments use either a Res Net-18 or Res Net-34 for the weak feature map g (see Appendix C for full details)', implying experimental setup details are in the appendix. However, no specific hyperparameters or training configurations are provided in the main text.