Transport of Algebraic Structure to Latent Embeddings

Authors: Samuel Pfrommer, Brendon G. Anderson, Somayeh Sojoudi

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate these structural transport nets for a range of mirrored algebras against baselines that operate directly on the latent space. Our experiments provide strong evidence that respecting the underlying algebraic structure of the input space is key for learning accurate and self-consistent operations.
Researcher Affiliation Academia 1University of California, Berkeley. Correspondence to: Samuel Pfrommer <sam.pfrommer@berkeley.edu>.
Pseudocode Yes Algorithm 1 Transport of algebraic structure from S to L
Open Source Code Yes Source code for our experiments is available on Git Hub.
Open Datasets No The paper mentions generating a synthetic random dataset and training/validation/testing splits from it but does not provide concrete access information (link, DOI, citation) to make this dataset publicly available or reproducible.
Dataset Splits Yes We generate 10^4 such random sets with an 80% training, 10% validation, and 10% testing split.
Hardware Specification Yes Reproducing these experiments takes approximately 5 GPU-days on an RTX A6000 GPU with an i7 core CPU.
Software Dependencies Yes Our codebase was developed against Py Torch 2.1.2.
Experiment Setup Yes Training runs for 10 epochs using Adam with a learning rate of 10^-3. ... Training uses the Adam optimizer with a learning rate of 0.01 and batch size of 10^3 for 10 epochs. ... We use the Adam optimizer for 50 epochs with a learning rate of 10^-4, weight decay of 10^-3, and batch size of 16.