Exploring the Latent Space of Autoencoders with Interventional Assays

Authors: Felix Leeb, Stefan Bauer, Michel Besserve, Bernhard Schölkopf

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments & Results. Experimental Setup We apply our new tools on a small selection of common benchmark datasets, including 3D-Shapes [66], MNIST [67], and Fashion-MNIST [68] 3. Our methods are directly applicable to any VAE-based model, and can readily be extended to any autoencoders. Nevertheless, we focus our empirical evaluation on vanilla VAEs and some β-VAEs (denoted by substituting the β, so 4-VAE refers to a β-VAE where β = 4). Specifically, here we mostly analyze a 4-VAE model with a d = 24 latent space trained on 3D-Shapes (except for table 1) referred to as Model A, and include results on a range of other models in the appendix. All our models use four convolution and two fully-connected layers in the encoder and decoder. The models are trained using Adam [69] with a learning rate of 0.001 for 100k steps (see appendix for details).
Researcher Affiliation Academia Email: fleeb@tuebingen.mpg.de Max Planck Institute for Intelligent Systems, Tübingen, Germany
Pseudocode No The paper describes processes and derivations mathematically and conceptually, but it does not include any formally structured pseudocode or algorithm blocks.
Open Source Code Yes We release our code at https://github.com/felixludos/latent-responses.
Open Datasets Yes We apply our new tools on a small selection of common benchmark datasets, including 3D-Shapes [66], MNIST [67], and Fashion-MNIST [68] 3.
Dataset Splits No The paper mentions using benchmark datasets for empirical evaluation and discusses 'training samples' in a toy example, but it does not specify explicit train/validation/test split percentages or sample counts for the main experiments, nor does it refer to predefined splits with citations.
Hardware Specification No The paper details the model architecture and training procedure ('All our models use four convolution and two fully-connected layers in the encoder and decoder. The models are trained using Adam [69] with a learning rate of 0.001 for 100k steps'), but it does not specify any hardware components such as GPU models, CPU types, or memory used for running the experiments.
Software Dependencies No The paper mentions 'Adam [69]' as the optimizer and that models are built with 'four convolution and two fully-connected layers', implying the use of deep learning frameworks, but it does not provide specific version numbers for any software dependencies (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes All our models use four convolution and two fully-connected layers in the encoder and decoder. The models are trained using Adam [69] with a learning rate of 0.001 for 100k steps (see appendix for details).