| 研究領域 | 人工知能と脳科学の対照と融合 |
|---|
| 研究課題/領域番号 |
19H04995
|
|---|
| 研究機関 | 国立研究開発法人理化学研究所 |
|---|
研究代表者 |
ベヌッチ アンドレア 国立研究開発法人理化学研究所, 脳神経科学研究センター, チームリーダー (50722352)
|
|---|
| 研究期間 (年度) |
2019-06-28 – 2021-03-31
|
| キーワード | Decision-making / sensory processing / visual cortex / neural computation / optogenetics / cognition |
|---|
| 研究実績の概要 |
The research carried over the fiscal year FY2020 has led to the completion of experimental and theoretical work focused on fundamental questions on sensory processing and sensory-based decision making. Specifically, we have completed two main works that have been submitted for publication. In one work by Dr. Orlandi, we used locaNMF tensor decomposition methods to isolate choice signals in the widefield response dynamics of dorsal-parietal cortical networks during a decision-making task. This work revealed widespread choice signals across these networks, possibly reflecting top-down signals for inference coding, both of sensory signals and of sensory-to-decision computations. Using a recurrent neural network (RNN) we also demonstrated that top-down signals recapitulated the complex decision-making process involved in solving the behavioral task at hand. In a second work by Dr. Abdolrahmani and Lyamzin, we examined how the cognitive state of the animal (sustained attention) affected the overlapping dynamics of sensory and motor-related activations, multiplexed across dorsal-parietal cortical networks during a decision-making task. We found that attention improved the discriminability (demixing) of overlapping visuo-moto signals. Both these works are under review and have been uploaded as pre-prints in bioRxiv public repository.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
We report progress on both experimental and theoretical research.
Experimentally, we have advanced our ability to manipulate the probability of single-cell excitation in large neuronal populations using digital-micromirror technology combined with optogenetic approaches. This technology has played a pivotal role in the completion of a research work already submitted for publication. At the theoretical level, we have implemented machine-learning computational frameworks based on convolutional and recurrent neural networks to capture computational principles underlying neural recordings from the mouse brain. Related work has resulted into two manuscripts that have been submitted for publication.
|
| 今後の研究の推進方策 |
The research plan for next fiscal year focuses on having four more accepted works in peer reviewed journals. They works have already been submitted and they are currently under review. Following reviewers’ comments, we expect additional analysis and experiments will be needed. We also have three more works in preparations. Two of them are experimentally demanding: one project aims at revealing the computational role of “intermediate” visual areas along the ventral visual stream. Specifically, we are training mice in visual-texture discrimination, and we plan to dissect underlying neural circuits in higher visual areas of the mouse cortex. Another project instead relies on holographic technology for single cell-excitation using spatial light modulators in combination with a two-photon microscope. We aim to implement model-based perturbations of large ensembles of cells, at the single cell resolution, as mice perform in a perceptual decision-making task.
|
