Related Experiment Video
Updated: Jun 21, 2026
09:44
Recording and Analyzing Multimodal Large-Scale Neuronal Ensemble Dynamics on CMOS-Integrated High-Density Microelectrode Array
Published on: March 8, 2024
Graded information extraction by neural-network dynamics with multihysteretic neurons
Yukihiro Tsuboshita1, Hiroshi Okamoto
1Corporate Research & Technology Development Group, Fuji Xerox Co. Ltd., Japan.
Summary
This study introduces a novel neural network algorithm inspired by brain function for retrieving graded information. The multihysteretic neuron-network dynamics significantly improved keyword extraction performance compared to existing methods.
Area of Science:
- Computational neuroscience
- Artificial intelligence
- Information retrieval
Background:
- The brain's graded persistent activity suggests mechanisms for retrieving nuanced information.
- Multihysteretic neuron models can generate continuous attractors, unlike bistable neurons.
Purpose of the Study:
- To develop a novel neural network algorithm for graded information retrieval.
- To evaluate the algorithm's performance in keyword extraction from documents.
Main Methods:
- Proposed a multihysteretic neuron-network dynamics algorithm.
- Applied the algorithm to document keyword extraction.
- Compared performance against standard retrieval methods and bistable neural networks.
Main Results:
- The proposed algorithm achieved significantly higher performance in keyword extraction (measured by average precision).
- Outperformed standard document retrieval methods.
- Demonstrated superior performance compared to neural networks using bistable neurons.
Conclusions:
- Multihysteretic neuron-network dynamics are effective for graded information retrieval.
- Handling graded information at the single-cell level is crucial for high-performing algorithms.
- The algorithm shows promise for advanced information processing tasks.
