Permutation Equivariant Neural Networks for Symmetric Tensors

Authors: Edward Pearce-Crump

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 8. Numerical Experiments We demonstrate our characterisation on the following toy experiments. S12-Invariant Task: We evaluate our model on a synthetic S12-invariant task given by the function f(T) := P12 i,j Ti,j,i, where T is a 3-order symmetric tensor. We demonstrate the high data efficiency of our model compared with a standard MLP for this task, as shown in Figure 3. ... S8-Equivariant Task: We evaluate our model on a synthetic S8-equivariant task from (R8) 3 to R8: namely, to extract the diagonal from 8 8 8 symmetric tensors. We evaluate our model against a standard MLP and a standard S8-equivariant model from (R8) 3 to R8. We show the Test Mean Squared Error (MSE) for each of these models in Table 1.
Researcher Affiliation Academia 1Department of Mathematics, Imperial College London, United Kingdom. Correspondence to: Edward Pearce Crump <EMAIL>.
Pseudocode Yes Procedure: Generation of the Transformation Map Labels That Describe the Transformation Corresponding to a (k, l) Bipartition Diagram of a Symmetric Tensor in (Rn) k to a Symmetric Tensor in (Rn) l. Input: A (k, l) bipartition diagram dπ and a value of n. 1. Apply Subprocedure I to obtain all possible grouped outputs for dπ, and for each one, its associated set of partially labelled diagrams. ... Output: A set of transformation map labels describing the transformation Dπ of a symmetric tensor T (Rn) k to a symmetric tensor Dπ(T) (Rn) l for the given value of n, where Dπ corresponds to the (k, l) bipartition diagram dπ.
Open Source Code Yes 1The code is available at https://github.com/epearcecrump/symmetrictensors.
Open Datasets No S12-Invariant Task: We randomly generated a synthetic data set consisting of 5000 symmetric tensors, split into 90% training and 10% test. ... S8-Equivariant Task: We randommly generated a synthetic data set consisting of 10000 symmetric tensors, split into 90% training and 10% test. The paper uses synthetically generated data and does not provide concrete access information for a publicly available or open dataset.
Dataset Splits Yes S12-Invariant Task: We randomly generated a synthetic data set consisting of 5000 symmetric tensors, split into 90% training and 10% test. ... S8-Equivariant Task: We randommly generated a synthetic data set consisting of 10000 symmetric tensors, split into 90% training and 10% test.
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 Yes Both models were optimised with stochastic gradient descent with a learning rate of 0.0001. We trained both models for 50 epochs with a batch size of 50.