On the consistency of top-k surrogate losses

Authors: Forest Yang, Sanmi Koyejo

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our contributions are primarily theoretical, and are outlined as follows: ... We employ these losses in synthetic experiments, observing aspects of their behavior which reflect our theoretical analysis.
Researcher Affiliation Collaboration Forest Yang 1 2 Sanmi Koyejo 1 3 1Google Research Accra 2University of California, Berkeley 3University of Illinois at Urbana-Champaign.
Pseudocode No The paper describes its methods mathematically and textually, but it does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets No The paper describes the generation of synthetic data for experiments ('We construct training data which matches the above setting', 'randomly sample from a d dimensional Gaussian'), but does not provide access information (link, DOI, citation) to a publicly available dataset.
Dataset Splits No The paper describes generating separate training and test sets but does not explicitly mention a validation set or specific proportions/counts for a train/validation/test split.
Hardware Specification Yes A machine with an Intel Core i7 8th-gen CPU with 16GB of RAM was used.
Software Dependencies No The paper mentions using 'Pytorch' but does not specify a version number or list other software dependencies with their versions.
Experiment Setup Yes We train our neural architecture on the data using batch gradient descent, setting the loss of the last layer to be each of {ψ1, . . . , ψ5} with k = 2. ... We optimize with Adam for 500 epochs, using a learning rate of 0.1 and full batch.