Bayesian Representation Learning with Oracle Constraints

Authors: Theofanis Karaletsos, Serge Belongie, Gunnar Rätsch

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

Reproducibility Variable Result LLM Response
Research Type Experimental In Sections 4 and 5, we present experimental results for benchmarking the proposed approaches, illustrating their properties and discussing the benefits they confer over competing approaches.
Researcher Affiliation Academia Theofanis Karaletsos Computational Biology, Sloan Kettering Institute 1275 York Avenue, New York, USA Theofanis.Karaletsos@ratschlab.org Serge Belongie Cornell Tech 111 8th Avenue #302, New York, USA sjb344@cornell.edu Gunnar R atsch Computational Biology, Sloan Kettering Institute 1275 York Avenue, New York, USA Gunnar.Ratsch@ratschlab.org
Pseudocode No The paper describes the generative model and learning process in textual steps and mathematical formulations, but it does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper mentions using open-source libraries like Theano, but it does not provide an explicit statement or link indicating that the source code for the methodology developed in this paper is publicly available.
Open Datasets Yes We use a relatively small dataset that is, however, well-suited to illustrate the features of the algorithm and facilitates the interpretation of the factorized latent spaces: the Yale Faces dataset (Lee et al., 2005).
Dataset Splits No We split it into 300 test images and 2, 114 training images. (The paper explicitly mentions training and testing splits for the Yale Faces dataset, but it does not specify a separate validation split or how it was used for hyperparameter tuning or model selection.)
Hardware Specification No All experiments were run on Graphics Processing Units (GPUs) using a theano (Bergstra et al., 2010) implementation and did not take more than a few hours each. (The paper mentions 'GPUs' but does not specify any particular GPU models, CPU details, or other specific hardware components used for the experiments.)
Software Dependencies No In all experiments we used diagonal Normal distributions as priors for the latent space and rms Prop with momentum (Graves, 2013) or ADAM (Kingma & Ba, 2014) as an optimizer. All experiments were run on Graphics Processing Units (GPUs) using a theano (Bergstra et al., 2010) implementation and did not take more than a few hours each. (The paper lists several software components and optimizers, but it does not provide specific version numbers for them, e.g., 'theano 0.7' or 'ADAM 1.0'.)
Experiment Setup Yes We proceed to learn fully unsupervised models of these images using an architecture with 200 hidden deterministic units and 50 latent variables. The deterministic layers use tanh nonlinearities.