Learning Sparse Latent Representations with the Deep Copula Information Bottleneck

Authors: Aleksander Wieczorek*, Mario Wieser*, Damian Murezzan, Volker Roth

ICLR 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our method on artificial and real data.
Researcher Affiliation Academia Aleksander Wieczorek , Mario Wieser , Damian Murezzan, Volker Roth University of Basel, Switzerland {firstname.lastname}@unibas.ch
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not include an explicit statement about releasing source code for the methodology or provide a link to a code repository.
Open Datasets Yes We consider the unnormalised Communities and Crime dataset Lyons et al. (1998) from the UCI repository1. The dataset consisted of 125 predictive, 4 non-predictive and 18 target variables with 2215 samples in total. In a preprocessing step, we removed all missing values from the dataset. In the end, we used 1901 observations with 102 predictive and 18 target variables in our analysis. 1http://archive.ics.uci.edu/ml/datasets/communities+and+crime+unnormalized
Dataset Splits No We split the samples into test (20k samples) and training (180k samples) sets.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions the use of the Adam optimizer but does not specify the versions of programming languages, libraries, or other software dependencies used for the experiments.
Experiment Setup Yes We use a latent layer with ten nodes that model the means of the ten-dimensional latent space t. The variance of the latent space is set to 1 for simplicity. The encoder as well as the decoder consist of a neural network with two fully-connected hidden layers with 50 nodes each. We use the softplus function as the activation function. Our model is trained using mini batches (size = 500) with the Adam optimiser (Kingma & Ba, 2014) for 70000 iterations using a learning rate of 0.0006.