Teaching via Best-Case Counterexamples in the Learning-with-Equivalence-Queries Paradigm

Authors: Akash Kumar, Yuxin Chen, Adish Singla

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We characterize the query complexity for the optimal teacher in the Lw EQ paradigm, termed as learning-with-equivalence-queries teaching dimension (Lw EQ-TD). (see Section 3) II. We study the query complexity in the Lw EQ paradigm under different teaching settings: worst-case, random, and best-case, distinguished by the informativeness of counterexamples. We showcase the power of best-case counterexamples picked by the optimal teacher, in contrast to worst-case or random counterexamples, for different hypothesis classes, including Axes-aligned hyperplanes, Monotone monomials, and Orthogonal rectangles. (see Section 4) III. We establish new connections between Lw EQ-TD and Lf S-TD by studying Lw EQ-TD for different learner models based on the richness of their query functions. (see Section 5)
Researcher Affiliation Academia 1Max Planck Institute for Software Systems (MPI-SWS), 2UC San Diego, 3University of Chicago
Pseudocode Yes Algorithm 1: Query protocol between the learner and the teacher
Open Source Code No The paper does not provide any statement or link indicating that source code for the described methodology is publicly available. The ethics checklist indicates '[N/A]' for code-related questions.
Open Datasets No This is a theoretical paper that does not involve training models on datasets. The 'If you ran experiments...' section of the ethics checklist indicates '[N/A]' for data-related questions.
Dataset Splits No This is a theoretical paper that does not involve validation splits or experimental data. The 'If you ran experiments...' section of the ethics checklist indicates '[N/A]' for data-related questions.
Hardware Specification No This is a theoretical paper that does not describe experimental setups requiring hardware specifications. The 'If you ran experiments...' section of the ethics checklist indicates '[N/A]' for hardware details.
Software Dependencies No This is a theoretical paper that does not describe experimental setups requiring specific software dependencies with version numbers. The 'If you ran experiments...' section of the ethics checklist indicates '[N/A]' for software dependencies.
Experiment Setup No This is a theoretical paper that does not describe a practical experimental setup with hyperparameters or training configurations. The 'If you ran experiments...' section of the ethics checklist indicates '[N/A]' for training details.