Function-Space Transformer Adapts Representation for Scientific and Visual Tasks

Researchers have introduced the Function-Space Transformer (FST), a novel framework designed to learn from functions represented by discrete samples. Unlike traditional methods that use fixed grids, FST employs a spatially adaptive continuous latent representation with anchors whose locations are predicted and refined from input observations. This adaptive approach allows the representation to better match the input's spatial organization, leading to improved performance. The FST has demonstrated superior results compared to Perceiver IO on PDE prediction tasks and achieved higher accuracy with fewer parameters than Vision Transformer on ImageNet-1K. AI

IMPACT Introduces a novel adaptive representation technique that could improve performance on scientific prediction and visual recognition tasks.

RANK_REASON The cluster contains a research paper detailing a new model architecture. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Burgers
• Darcy flow of polymer from an inclined plane with convective heat transfer analysis: a numerical study
• Fourier Neural Operator
• Function-Space Transformer
• ImageNet-1K
• PDEBench
• Pedram Rooshenas
• Perceiver IO
• vision transformer

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →