Related Experiment Video
Updated: Jan 29, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
10.0K
Characterization of deep neural network features by decodability from human brain activity
Tomoyasu Horikawa1, Shuntaro C Aoki1, Mitsuaki Tsukamoto1
1ATR Computational Neuroscience Laboratories, 2-2-2 Hikaridai, Seika, Soraku, Kyoto 619-0288, Japan.
Scientific Data
|February 13, 2019
Summary
This study compares human brain activity and deep neural network (DNN) vision. We found systematic differences between DNN features and human brain signals, highlighting a gap in current AI models.
Area of Science:
- Neuroscience
- Computer Science
- Artificial Intelligence
Background:
- Deep neural networks (DNNs) achieve near human-level object recognition.
- Comparative studies between DNNs and human vision are increasing.
- DNNs are used as models for hierarchical visual processing.
Purpose of the Study:
- To investigate the relationship between human brain activity and DNN feature representations.
- To identify and quantify the differences between DNNs and human vision.
- To provide a dataset for further research into brain- DNN discrepancies.
Main Methods:
- Functional magnetic resonance imaging (fMRI) was used to measure human brain activity.
- Deep neural network (AlexNet, VGG19) feature values were decoded from fMRI signals.
- Decoding accuracies of individual DNN features were analyzed and ranked.
- fMRI data from five human subjects viewing images were collected.
Main Results:
- Human brain activity patterns could be decoded into DNN feature values.
- Not all DNN features were equally decodable, indicating a gap.
- Decoding accuracies of individual features showed high inter-subject correlation.
- Systematic differences between DNNs and human visual processing were identified.
Conclusions:
- There are significant, systematic differences between DNNs and human vision.
- The presented dataset can help elucidate the gap between artificial and human visual systems.
- Further research using this dataset may lead to improved AI models and applications.
More Related Videos
11:14A Novel Experimental and Analytical Approach to the Multimodal Neural Decoding of Intent During Social Interaction in Freely-behaving Human Infants
Published on: October 4, 2015
11.5K
03:31Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
1.1K
Related Concept Videos
02:26Protein Networks
4.5K
An organism can have thousands of different proteins, and these proteins must cooperate to ensure the health of an organism. Proteins bind to other proteins and form complexes to carry out their functions. Many proteins interact with multiple other proteins creating a complex network of protein interactions.
These interactions can be represented through maps depicting protein-protein interaction networks, represented as nodes and edges. Nodes are circles that are representative of a protein,...
These interactions can be represented through maps depicting protein-protein interaction networks, represented as nodes and edges. Nodes are circles that are representative of a protein,...
4.5K
02:26Protein Networks
2.8K
2.8K
01:37Neural Regulation
43.3K
Digestion begins with a cephalic phase that prepares the digestive system to receive food. When our brain processes visual or olfactory information about food, it triggers impulses in the cranial nerves innervating the salivary glands and stomach to prepare for food.
43.3K
02:26tRNA Activation
22.9K
Aminoacyl-tRNA synthetases are present in both eukaryotes and bacteria. Though eukaryotes have 20 different aminoacyl-tRNA synthetases to couple to 20 amino acids, many bacteria do not have genes for all of these aminoacyl-tRNA synthetases. Despite this, they still use all 20 amino acids to synthesize their proteins. For instance, some bacteria do not have the gene encoding the enzyme that couples glutamine with its partner tRNA. In these organisms, one enzyme adds glutamic acid to all of the...
22.9K
02:18Network Covalent Solids
16.2K
Network covalent solids contain a three-dimensional network of covalently bonded atoms as found in the crystal structures of nonmetals like diamond, graphite, silicon, and some covalent compounds, such as silicon dioxide (sand) and silicon carbide (carborundum, the abrasive on sandpaper). Many minerals have networks of covalent bonds.
To break or to melt a covalent network solid, covalent bonds must be broken. Because covalent bonds are relatively strong, covalent network solids are typically...
To break or to melt a covalent network solid, covalent bonds must be broken. Because covalent bonds are relatively strong, covalent network solids are typically...
16.2K
01:20COPD: Pathogenesis and Clinical Features
1.8K
Chronic obstructive pulmonary disease (COPD) is a group of lung conditions that progressively worsen over time, including chronic bronchitis and emphysema. This cluster of diseases collectively leads to a gradual and irreversible decline in lung function over time.
The primary cause for the onset of COPD is cigarette smoking and exposure to air pollution. These hazardous factors initiate a chain reaction within the lungs, resulting in chronic inflammation, damage to the airways, and a...
The primary cause for the onset of COPD is cigarette smoking and exposure to air pollution. These hazardous factors initiate a chain reaction within the lungs, resulting in chronic inflammation, damage to the airways, and a...
1.8K