Task structure and nonlinearity jointly determine learned representational geometry

Authors: Matteo Alleman, Jack Lindsey, Stefano Fusi

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this study, we conduct an in-depth investigation of the impact of input geometry, label geometry, and nonlinearity on learned representations. We employ a parameterized family of classification tasks that allows us to probe the impact of each of these factors independently and focus on single-hidden-layer networks in which we can precisely describe representation learning dynamics over the course of training.
Researcher Affiliation Academia Matteo Alleman , Jack Lindsey & Stefano Fusi Department of Neuroscience, Columbia University ma3811@columbia.edu, jackwlindsey@gmail.com, sf2237@columbia.edu
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks (clearly labeled algorithm sections or code-like formatted procedures).
Open Source Code No The paper does not provide concrete access to source code (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described in this paper.
Open Datasets Yes To assess the applicability of our findings to more realistic tasks, we trained convolutional networks image classification task, experimenting with two architectures a small network with two convolutional and two fully connected layers, and the Res Net-18 architecture and two datasets, CIFAR-10 and STL-10.
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers like Python 3.8, CPLEX 12.4) needed to replicate the experiment.
Experiment Setup No The paper lacks specific experimental setup details such as concrete hyperparameter values (learning rate, batch size, number of epochs), optimizer settings, or other system-level training configurations in the main text.