Learning Logistic Circuits

Authors: Yitao Liang, Guy Van den Broeck4277-4286

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental On MNIST and Fashion datasets, our learning algorithm outperforms neural networks that have an order of magnitude more parameters. We run experiments on standard image classification benchmarks (MNIST and Fashion) and achieve accuracy higher than much larger MLPs and even CNNs with an order of magnitude more parameters.
Researcher Affiliation Academia Yitao Liang, Guy Van den Broeck Computer Science Department University of California, Los Angeles {yliang, guyvdb}@cs.ucla.edu
Pseudocode Yes Algorithm 1: Node probabilities from a real-valued sample x. Algorithm 2: Features x from a real-valued sample x.
Open Source Code Yes Open-source code and experiments are available at https://github.com/UCLA-StarAI/LogisticCircuit.
Open Datasets Yes We choose MNIST and Fashion5 as our testbeds. (Xiao, Rasul, and Vollgraf 2017) (Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms.)
Dataset Splits Yes The learned structure with the highest F1 score on validation after 48 hours of running is used for evaluation.
Hardware Specification No All experiments are run on single CPUs. (This does not provide specific model numbers or detailed specifications.)
Software Dependencies No No specific version numbers for software dependencies (e.g., Python, PyTorch, TensorFlow, etc.) are provided.
Experiment Setup Yes All experiments start with a predefined initial structure; we defer its details to Appendix D. The learned structure with the highest F1 score on validation after 48 hours of running is used for evaluation. Appendix D. Initial Structure. Appendix E Details of Existing Classification Models