4 datasets found

The Long Range Arena dataset consists of 6 tasks with lengths 1K-16K steps encompassing modalities and objectives that require similarity, structural, and visuospatial reasoning.

The proposed method is evaluated on a toy example problem of sequence counting followed by MNIST classification problem.

The MNIST and CIFAR datasets are used to evaluate the proposed Variational State Space Model (VSSM) for autoregressive generation.

The dataset used in this paper is the Long Range Arena (LRA) benchmark for efficient sequence models, which contains 6 tasks with lengths 1K-16K steps, encompassing modalities and...