From Neurons to Neutrons: A Case Study in Interpretability

Authors: Ouail Kitouni, Niklas Nolte, Vı́ctor Samuel Pérez-Dı́az, Sokratis Trifinopoulos, Mike Williams

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

Reproducibility Variable Result LLM Response
Research Type Experimental In all our experiments, we will consider one or several observables to predict with various models. The performance of the models will generally be measured by a Root-Mean Square error (RMS) on a holdout set.
Researcher Affiliation Collaboration 1NSF Institute for Artificial Intelligence and Fundamental Interactions (IAIFI) 2Massachusetts Institute of Technology 3FAIR at Meta 4Harvard John A. Paulson School of Engineering and Applied Sciences 5Center for Astrophysics | Harvard & Smithsonian 6School of Engineering, Science and Technology, Universidad del Rosario.
Pseudocode No The paper describes procedures and algorithms conceptually but does not include any formal pseudocode or algorithm blocks.
Open Source Code Yes Example code is available here: https://github.com/samuelperezdi/nuclr-icml
Open Datasets Yes The data sources are: for the various energies the Atomic Mass Evaluation (AME) (Wang et al., 2021) and for the charge radii the Atomic Data and Nuclear Data Tables 99 (2013) (Angeli & Marinova, 2013).
Dataset Splits Yes We train models with different train/validation splits (10% to 90% in 10% increments, 3 random seeds each), varying batch size for consistent total optimization steps, and keeping other hyperparameters constant.with 50% of the data held out as a validation set in each setting to gauge the generalization performance.
Hardware Specification Yes Most training runs were on Nvidia V100 GPUs with some done on Nvidia A6000 GPUs.
Software Dependencies No The paper mentions using SiLU activations and AdamW optimizer, but it does not provide specific version numbers for any software libraries, programming languages, or other dependencies.
Experiment Setup Yes The runs used to generate the embeddings and visualizations have the following parameters: EPOCHS = 200,000 HIDDEN DIM = 2048 LR = 0.0001 WD = 0.01 DEPTH = 2 Seed = 0