Conservativeness of Untied Auto-Encoders

Authors: Daniel Im, Mohamed Belghazi, Roland Memisevic

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

Reproducibility Variable Result LLM Response
Research Type Experimental To this end, we train an untied auto-encoder with 500 hidden units with and without weight length constraints4 on the MNIST dataset. We measure symmetricity using sym(A) = (A+AT )/2 2 A 2 which yields values between [0, 1] with 1 representing complete symmetricity. Figure 2a and 2c shows the evolution of the symmetricity of r(x) x during training.
Researcher Affiliation Academia Daniel Jiwoong Im Montreal Institute for Learning Algorithms University of Montreal Montreal, QC, H3C 3J7 imdaniel@iro.umontreal.ca Mohamed Ishmael Belghazi HEC Montreal 3000 Ch de la Cte-Ste-Catherine Montreal, QC, H3T 2A7 mohamed.2.belghazi@hec.ca Roland Memisevic Montreal Institute for Learning Algorithms University of Montreal Montreal, QC, H3C 3J7 roland.memisevic@umontreal.ca
Pseudocode Yes Algorithm 1 Learning to approximate a conservative field with an auto-encoder
Open Source Code No The paper does not provide an explicit statement or link for the open-source code for the described methodology.
Open Datasets Yes To this end, we train an untied auto-encoder with 500 hidden units with and without weight length constraints4 on the MNIST dataset.
Dataset Splits No The paper refers to 'training data' and mentions using the MNIST dataset, but it does not specify explicit training, validation, and test splits (e.g., percentages, sample counts, or predefined standard splits).
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers used to replicate the experiments.
Experiment Setup Yes We train an untied auto-encoder with 500 hidden units... We train an untied auto-encoder with 1000 Re LU units for 500 epochs using BFGS over an equally spaced grid of 100 points in each dimension.