Supervised Learning with Tensor Networks

Authors: Edwin Stoudenmire, David J. Schwab

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental For the MNIST data set we obtain less than 1% test set classification error.
Researcher Affiliation Academia E. M. Stoudenmire Perimeter Institute for Theoretical Physics Waterloo, Ontario, N2L 2Y5, Canada David J. Schwab Department of Physics Northwestern University, Evanston, IL
Pseudocode No The paper describes algorithms through text and diagrams (e.g., Fig. 6, Fig. 7) but does not include structured pseudocode or algorithm blocks.
Open Source Code Yes For researchers interested in reproducing our results, we have made our codes publicly available at: https://github.com/emstoudenmire/TNML. The codes are based on the ITensor library [19].
Open Datasets Yes To test the tensor network approach on a realistic task, we used the MNIST data set [24].
Dataset Splits No The paper mentions 'training or test sets' and 'training set error' and 'test error' but does not specify a validation set or explicit dataset splits (e.g., percentages, sample counts, or predefined splits for train/validation/test).
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or cloud instances) used for running the experiments.
Software Dependencies No The paper states 'The codes are based on the ITensor library [19]' but does not provide specific version numbers for ITensor or other software dependencies.
Experiment Setup Yes Each image was scaled down from 28x28 to 14x14 by averaging clusters of four pixels... We chose a zig-zag ordering meaning the first row of pixels are mapped to the first 14 external MPS indices... Using the sweeping algorithm in Section 4 to optimize the weights, we found the algorithm quickly converged after a few passes, or sweeps, over the MPS. Typically five or less sweeps were needed... Test error rates also decreased rapidly with the maximum MPS bond dimension m. For m = 10... for m = 20... The largest bond dimension we tried was m = 120...