Multi-step learning and underlying structure in statistical models

Authors: Maia Fraser

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We propose to analyze SSL and other multi-step learning approaches, much in the spirit of Baxter s framework, by defining a learning problem generatively as a joint statistical model on X Y . This determines in a natural way the class of conditional distributions that are possible with each marginal, and amounts to an abstract form of compatibility function. It also allows to analyze both discrete and non-discrete settings. As tool for our analysis, we define a notion of γ-uniform shattering for statistical models. We use this to give conditions on the marginal and conditional models which imply an advantage for multi-step learning approaches. In particular, we recover a more general version of a result of Poggio et al (2012): under mild hypotheses a multi-step approach which learns features invariant under successive factors of a finite group of invariances has sample complexity requirements that are additive rather than multiplicative in the size of the subgroups.
Researcher Affiliation Academia Maia Fraser Dept. of Mathematics and Statistics Brain and Mind Research Institute University of Ottawa Ottawa, ON K1N 6N5, Canada mfrase8@uottawa.ca
Pseudocode No The paper does not include any pseudocode or clearly labeled algorithm blocks. It focuses on theoretical definitions, theorems, and proofs.
Open Source Code No The paper makes no mention of open-sourcing code for the described methodology or providing links to any code repositories.
Open Datasets No The paper is theoretical and does not use specific, publicly available datasets for training. It discusses abstract concepts of data samples (e.g., "sample z = (z1, z2 zm), with zi = (xi, yi) 2 X Y = Z, drawn iid from the distribution p").
Dataset Splits No This is a theoretical paper that does not involve practical experiments with dataset splits. Thus, there is no mention of validation dataset splits or methodology.
Hardware Specification No The paper is theoretical and does not report on experiments requiring specific hardware. Therefore, no hardware specifications are mentioned.
Software Dependencies No The paper is theoretical and does not describe implementation details or dependencies on specific software with version numbers.
Experiment Setup No The paper is purely theoretical and does not describe any experimental setup details such as hyperparameters, training configurations, or system-level settings.