Discriminative Unsupervised Feature Learning with Convolutional Neural Networks

Authors: Alexey Dosovitskiy, Jost Tobias Springenberg, Martin Riedmiller, Thomas Brox

NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The feature representation learned by our algorithm achieves classification results matching or outperforming the current state-of-the-art for unsupervised learning on several popular datasets (STL-10, CIFAR-10, Caltech-101).
Researcher Affiliation Academia Alexey Dosovitskiy, Jost Tobias Springenberg, Martin Riedmiller and Thomas Brox Department of Computer Science University of Freiburg 79110, Freiburg im Breisgau, Germany {dosovits,springj,riedmiller,brox}@cs.uni-freiburg.de
Pseudocode No The paper describes the learning algorithm and formulations in text and equations, but does not provide structured pseudocode or algorithm blocks.
Open Source Code Yes Our code and training data are available at http://lmb.informatik.uni-freiburg.de/resources .
Open Datasets Yes To compare our discriminative approach to previous unsupervised feature learning methods, we report classification results on the STL-10 [21], CIFAR-10 [22] and Caltech-101 [23] datasets.
Dataset Splits Yes On STL-10 the SVM was trained on 10 pre-defined folds of the training data. ... For Caltech-101 we followed the usual protocol of selecting 30 random samples per class for training and not more than 50 samples per class for testing. This was repeated 10 times.
Hardware Specification No The paper mentions training networks using Caffe but does not provide any specific details about the hardware used (e.g., CPU/GPU models, memory).
Software Dependencies No The paper mentions using 'Caffe' but does not specify its version or any other software dependencies with version numbers.
Experiment Setup Yes It consists of two convolutional layers with 64 filters each followed by a fully connected layer with 128 neurons. ... A large network, consisting of three convolutional layers with 64, 128 and 256 filters respectively followed by a fully connected layer with 512 neurons... In both models all convolutional filters are connected to a 5 5 region of their input. 2 2 maxpooling was performed after the first and second convolutional layers. Dropout [24] was applied to the fully connected layers.