Probabilistic Invariant Learning with Randomized Linear Classifiers

Authors: Leonardo Cotta, Gal Yehuda, Assaf Schuster, Chris J. Maddison

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we empirically demonstrate the benefits of this new class of models on invariant tasks where deterministic invariant neural networks are known to struggle.
Researcher Affiliation Academia Leonardo Cotta Vector Institute leonardo.cotta@vectorinstitute.ai Gal Yehuda Technion, Haifa, Israel ygal@cs.technion.ac.il Assaf Schuster Technion, Haifa, Israel assaf@technion.ac.il Chris J. Maddison University of Toronto and Vector Institute cmaddis@cs.toronto.edu
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Source code is available at: https://github.com/cottascience/invariant-rlcs
Open Datasets Yes To address Q1 and Q2, we consider the sorting task proposed in [32]. ... We chose the task of deciding whether a graph sampled from a G(n, p) model with p = 1.1 log(n)/n is connected or not.
Dataset Splits Yes The training sets consisted of 1000 examples, while the the validation and test sets contained 100 examples.
Hardware Specification No The paper does not provide specific details about the hardware used for experiments (e.g., GPU/CPU models, memory, or cloud instance types).
Software Dependencies No All models were trained with Pytorch [25] using Adagrad [10]. The paper mentions software names but does not specify their version numbers.
Experiment Setup Yes We tuned all the hyperparameters on the validation set using a patience of 30. The Deep Sets models found a better learning rate of 0.001 and batch size of 250. The GNN model found a better learning rate of 0.01 and batch size 100. The RSet C model used a batch size of 250 and learning rate 0.5. The RGraph C model used a batch size of 100 and a learning rate of 0.5.