2005 Fiscal Year Final Research Report Summary
Emergent functions of networks of neurons with complex information processing abilities
| Project/Area Number |
16300096
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Bioinformatics/Life informatics
|
|---|
| Research Institution | RIKEN (2005)
Tamagawa University (2004) |
|---|
Principal Investigator |
FUKAI Tomoki RIKEN, Lab.for Neural Circuit Theory, Team Leader, 脳回路機能理論研究チーム, チームリーダー (40218871)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
TERAMAE Jun-nosuke RIKEN, Lab.for Neural Circuit Theory, Research scientist, 脳回路機能理論研究チーム, 研究員 (50384722)
|
|---|
| Project Period (FY) |
2004 – 2005
|
| Keywords | Brain, Neural network systems / Cognitive model / Biological information processing / Soft computing / Mathematical engineering / Statistical physics |
|---|
| Research Abstract |
Neurons have long been treated as processing units that can perform only simple computations, such as integrate-and-fire, and complex functions of the brain is considered to emerge from computations in neural networks. However, results of recent experiments have revealed that some single neurons perform rather complicated information processing like working memory, by storing and representing analog-type information with their firing rates. In our preceding study, we attempted to construct a model neuron that achieves such a single-cell memory operation with its multiple stable firing rates. Here, we consider neural networks of complex neuron models with multiple stable states. A simplest example includes temporal integrator neural network with noise-induced bi-stable neurons. Temporal integration of externally or internally driven information is a fundamental brain function required for a variety of cognitive behaviors. This process is generally linked with graded rate changes in cort
… More
ical neurons, which typically appear during a delay period of cognitive task in the prefrontal and other cortical areas. We have proposed a neural network model that produces graded (climbing or descending) neuronal activity with modifiable slopes. The model comprises stochastic bistable neurons that are innervated by a balanced background input and are interconnected randomly via recurrent synapses at an equal magnitude of the maximum conductance. Driven by an external input, individual model neurons exhibit bimodal rate changes between a baseline and an elevated firing state. These bimodal changes are temporally organized by reverberating synaptic input to generate graded activity with a nearly constant slope in the neuronal population. Numerical and analytical methods have revealed that the network model displays such temporal integrator-like activity with moderate tuning of the background input intensity and uniform synaptic weight. To test the validity of the proposed mechanism, we have analyzed the graded activity of anterior cingulate cortex neurons in monkeys performing delayed conditional Go/No-go discrimination tasks. We show that the graded delay-period activity of cingulate neurons exhibits bimodal activity patterns and trial-to-trial variability that are similar to those predicted by the proposed model. These results were reported in various national and international meetings, and have been submitted for publication in physics and neuroscience journals. Part of the results will appear as a review article in computational neuroscience journal. Less
|
