Teaching a black-box learner

Authors: Sanjoy Dasgupta, Daniel Hsu, Stefanos Poulis, Xiaojin Zhu

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we use Algorithm 1 to shrink several synthetic and real datasets, that is, to find subsets (teaching sets) of the data that yield the same final classifier. We illustrate its behavior in experiments with kernel machines and neural nets.
Researcher Affiliation Collaboration 1University of California, San Diego 2Columbia University 3NTENT 4University of Wisconsin Madison. Correspondence to: Sanjoy Dasgupta <dasgupta@eng.ucsd.edu>.
Pseudocode Yes Figure 1. The teacher’s algorithm.
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes Real datasets We also looked at the MNIST and fashion MNIST (Xiao et al., 2017) datasets, both with 60K points.
Dataset Splits No The paper uses terms like "full data set", "training set", and evaluates "accuracy" but does not specify explicit training, validation, or test dataset splits with percentages or sample counts.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory specifications) used for running the experiments.
Software Dependencies No The paper mentions the use of "SVM learners" and "convolutional network" but does not provide specific version numbers for software dependencies or libraries used.
Experiment Setup Yes For all experiments we used the same termination criterion: the algorithm terminated when it got within .01 of the accuracy of the learner that was trained using the full data. Also, to initialize the weight Tx of each data point we set the confidence parameter δ of Algorithm 1 to .1.