Transport Analysis of Infinitely Deep Neural Network

Authors: Sho Sonoda, Noboru Murata

JMLR 2019 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental According to analytic and numerical experiments, we showed that deep DAEs converge faster and that the extracted features are different from each other, which gives a partial answer to the other question why do DNNs perform better?
Researcher Affiliation Academia Sho Sonoda EMAIL Center for Advanced Intelligence Project RIKEN 1 4 1 Nihonbashi, Chuo-ku, Tokyo 103 0027, Japan Noboru Murata EMAIL School of Advanced Science and Engineering Waseda University 3 4 1 Okubo, Shinjuku-ku, Tokyo 169 8555, Japan
Pseudocode No The paper describes mathematical frameworks, definitions, and theorems but does not include any explicit pseudocode blocks or algorithm sections.
Open Source Code No The paper does not contain any explicit statement about releasing source code or provide links to a code repository.
Open Datasets No The paper uses synthetic data distributions such as "univariate normal distribution", "multivariate normal distribution", "mixture of multivariate normal distributions", and "2-dimensional swissroll data" for its numerical examples. While these are common types of data for theoretical illustration, no specific publicly available dataset is mentioned with a link, citation, or repository access information for direct reproduction.
Dataset Splits No The paper's numerical examples primarily use analytically defined or synthetic data distributions. There is no mention of specific train/test/validation splits in the context of machine learning experiments, as the focus is on theoretical analysis and simulation of flows rather than data-driven model evaluation.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU models, CPU types, memory) used for running the numerical experiments.
Software Dependencies No The paper does not mention any specific software libraries, frameworks, or their version numbers (e.g., Python, TensorFlow, PyTorch, CUDA) that would be needed to replicate the experiments.
Experiment Setup No While the paper describes numerical examples with certain parameters (e.g., noise variance 't', types of DAEs like 'shallow' or 'continuous', initial data distributions), it does not provide specific hyperparameters for training neural networks, such as learning rates, batch sizes, number of epochs, or optimizer settings, especially for the "real NNs" mentioned in Section 6.5.