Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Representation Learning via Manifold Flattening and Reconstruction
Authors: Michael Psenka, Druv Pai, Vishal Raman, Shankar Sastry, Yi Ma
JMLR 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present empirical results and comparisons to other models on synthetic high-dimensional manifold data and 2D image data. |
| Researcher Affiliation | Academia | Michael Psenka EMAIL Department of Electrical Engineering and Computer Science University of California, Berkeley Berkeley, CA 94720-1776, USA |
| Pseudocode | Yes | More formally, we propose the following algorithm for training Flattening Networks (Flat Nets): Algorithm 1 Main construction of Flat Net. |
| Open Source Code | Yes | Our code is publicly available. Code for both Flat Net itself and the below experiments is publicly available1. 1. https://github.com/michael-psenka/manifold-linearization |
| Open Datasets | Yes | The mnist database of handwritten digit images for machine learning research. IEEE Signal Processing Magazine, 29(6):141 142, 2012. |
| Dataset Splits | No | The paper describes generating synthetic data with parameters like N = 50, D = 2, d = 1, and M = 5000, and using the MNIST dataset. For MNIST, it mentions using 'a single image' or 'the original MNIST data set' but does not specify how this data was split into training, validation, or test sets. For synthetic data, it generates data points and features, but does not describe conventional dataset splitting methodologies for reproduction. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run its experiments, such as CPU or GPU models, memory, or processing speeds. |
| Software Dependencies | No | The paper mentions comparison models like VAEs and optimizers like Adam, but does not provide specific version numbers for any software dependencies (e.g., Python, PyTorch, TensorFlow, CUDA) that would be needed for reproducibility. |
| Experiment Setup | Yes | We compare the reconstruction and generalization performance of Flat Net against three types of variational autoencoders (VAEs): Each VAE encoder and decoder is a 5 layer multi-layer perceptron, with dimension 100 at each layer, and latent dimension d = 1. Each VAE is trained for 100 epochs with Adam optimizer and 10^-3 learning rate. |