Positive Concave Deep Equilibrium Models

Authors: Mateusz Gabor, Tomasz Piotrowski, Renato L. G. Cavalcante

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments demonstrate the competitiveness of our pc DEQ models against other implicit models. The experiments were carried out on three commonly known computer vision datasets: MNIST, SVHN, and CIFAR-10.
Researcher Affiliation Academia 1Faculty of Electronics, Photonics, and Microsystems,Wrocław University of Science and Technology, Wrocław, Poland 2Faculty of Physics, Astronomy and Informatics, Nicolaus Copernicus University, Toru n, Poland 3Fraunhofer Heinrich Hertz Institute, Berlin, Germany.
Pseudocode No The paper provides architectural diagrams but does not include explicit pseudocode or algorithm blocks.
Open Source Code Yes The Pytorch source code of pc DEQs along with examples of use is available at the following link: https://github.com/ mateuszgabor/pcdeq
Open Datasets Yes The experiments were carried out on three commonly known computer vision datasets: MNIST, SVHN, and CIFAR-10. MNIST dataset consists 70,000 grayscale handwritten digit images. SVHN dataset consists of 99,289 RGB digit images from house numbers. CIFAR-10 consists of 60,000 RGB images of 10 classes.
Dataset Splits No The paper provides total training and testing examples in Table 4, but does not explicitly specify the size or method for a validation split.
Hardware Specification Yes The experiments were performed using the Google Colab platform with a NVIDIA Tesla T4 16GB GPU.
Software Dependencies No The paper mentions using PyTorch and the AdamW optimizer but does not specify their version numbers or other software dependencies with versions.
Experiment Setup Yes All other hyperparameters for each dataset and architecture are shown in Tables 5, 6 and 7.