Related Experiment Video
Updated: Sep 4, 2025

Assessment and Communication for People with Disorders of Consciousness
Published on: August 1, 2017
Are we ready for a new paradigm shift? A survey on visual deep MLP
Ruiyang Liu1, Yinghui Li2, Linmi Tao1
1Department of Computer Science and Technology, BNRist, Tsinghua University & Key Lab of Pervasive Computing, Ministry of Education of China, Beijing 100084, China.
Deep multilayer perceptron (MLP) models show promise as a new computer vision paradigm. This review analyzes MLPs against convolution and self-attention, exploring their potential for future computing devices.
Area of Science:
- Computer Vision
- Deep Learning
- Artificial Intelligence
Background:
- Deep multilayer perceptron (MLP) models are gaining traction in the vision community.
- Historical paradigm shifts in computer vision were driven by large datasets and increased computing power.
Purpose of the Study:
- To evaluate if MLPs can establish a new paradigm in computer vision.
- To compare MLPs with existing dominant methods like convolution and self-attention.
Main Methods:
- Detailed comparison of intrinsic connections and differences between convolution, self-attention, and token-mixing MLPs.
- Analysis of token-mixing MLP advantages and limitations.
- Review of recent MLP-like variants, module designs, network architectures, and applications.
Main Results:
- MLPs, convolution, and self-attention represent current mainstream approaches for weighted summations in the GPU era.
- Token-mixing MLPs offer specific advantages and limitations that warrant detailed examination.
Conclusions:
- MLPs present a potential new paradigm for computer vision.
- Future development of MLPs should consider integration with next-generation computing devices.
More Related Videos
05:36Central and Divided Visual Field Presentation of Emotional Images to Measure Hemispheric Differences in Motivated Attention
Published on: November 16, 2017
07:34Perceptual and Category Processing of the Uncanny Valley Hypothesis' Dimension of Human Likeness: Some Methodological Issues
Published on: June 3, 2013
Related Concept Videos
Surveys
Auditory Perception
Types of Surveys
Survey Safety
Perception
Bottom-up processing begins at the sensory level, where receptors detect external environmental stimuli. These could include the tactile sensation of...
Perceptual Constancy
Size constancy is the recognition that an object remains the same size, even when its image on the retina changes. For instance, a bus is perceived to be large enough to carry people, even if it looks tiny from...