Disentangling factors of variation in deep representation using adversarial training

Authors: Michael F. Mathieu, Junbo Jake Zhao, Junbo Zhao, Aditya Ramesh, Pablo Sprechmann, Yann LeCun

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on synthetic and real datasets show that the proposed method is capable of generalizing to unseen classes and intra-class variabilities. We evaluate our model on both synthetic and real datasets: Sprites dataset [24], MNIST [15], NORB [16] and the Extended-Yale B dataset [8]. We report both training and testing errors in Table 1.
Researcher Affiliation Academia Michael Mathieu, Junbo Zhao, Pablo Sprechmann, Aditya Ramesh, Yann Le Cun 719 Broadway, 12th Floor, New York, NY 10003 {mathieu, junbo.zhao, pablo, ar2922, yann}@cs.nyu.edu
Pseudocode Yes The diagram of the network is shown in figure 1, and the described training procedure is summarized in on Algorithm 1, in the supplementary material.
Open Source Code No The paper does not provide an unambiguous statement or a direct link to open-source code for the methodology described.
Open Datasets Yes We evaluate our model on both synthetic and real datasets: Sprites dataset [24], MNIST [15], NORB [16] and the Extended-Yale B dataset [8].
Dataset Splits Yes We used early-stopping on a validation set to prevent overfitting. This results in 25 different object identities in the training set and another 25 distinct objects identities in the testing set [for NORB]. The training and testing sets contains roughly 600 and 180 images per individual respectively [for Extended-Yale B].
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts) used for running its experiments.
Software Dependencies No The paper mentions using "Torch7 [3]" but does not specify a version number for Torch7 or any other software dependencies needed to replicate the experiment.
Experiment Setup No The paper states that "The network architectures follow that of DCGAN [21] and are described in detail in the supplementary material," implying that such details are not in the main text. It mentions a "two-layer neural network with 256 hidden units" for a classification task, but not for the main model training setup.