Modularized Morphing of Neural Networks

Authors: Tao Wei, Changhu Wang, Chang Wen Chen

ICLR 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments have been conducted based on the state-of-the-art Res Net on benchmark datasets, and the effectiveness of the proposed solution has been verified.
Researcher Affiliation Collaboration Tao Wei University at Buffalo Buffalo, NY 14260 taowei@buffalo.edu Changhu Wang Microsoft Research Beijing, China, 100080 chw@microsoft.com Chang Wen Chen University at Buffalo Buffalo, NY 14260 chencw@buffalo.edu
Pseudocode Yes Algorithm 1 Algorithm for Simple Morphable Modules
Open Source Code No The paper does not contain any statement about releasing code or a link to a code repository.
Open Datasets Yes CIFAR10 (Krizhevsky & Hinton, 2009) is a benchmark dataset on image classification and neural network investigation. It consists of 32 32 color images in 10 categories, with 50,000 training images and 10,000 testing images.
Dataset Splits Yes CIFAR10 (Krizhevsky & Hinton, 2009) ... It consists of 32 32 color images in 10 categories, with 50,000 training images and 10,000 testing images." and "This dataset consists of 1,000 object categories, with 1.28 million training images and 50K validation images.
Hardware Specification No The paper does not mention any specific hardware (GPU, CPU model, memory, etc.) used for running the experiments.
Software Dependencies No The paper mentions training parameters like decay, momentum, optimizer (SGD), batch size, and learning rate schedule, but does not provide specific software dependencies with version numbers (e.g., libraries, frameworks).
Experiment Setup Yes In the training process, we follow the same setup as in (He et al., 2015). We use a decay of 0.0001 and a momentum of 0.9. We adopt the simple data augmentation with a pad of 4 pixels on each side of the original image. A 32 32 view is randomly cropped from the padded image and a random horizontal flip is optionally applied.