| 研究実績の概要 |
I propose a theoretical model where the prediction at each hierarchical level is established to minimize the mean-squared prediction errors received at the same level, and use this model to extract prediction and prediction-error signals from EEG data obtained during an auditory "local-global" paradigm where the temporal regularities of the stimuli were manipulated at two hierarchical levels. Prediction-error and prediction signals were found in the gamma (>30Hz) and beta (13~30Hz) frequency bands, respectively, and their hierarchical structures were represented by distinctive spatiotemporal dynamics. Spatially, first-level and second-level prediction-error signals (PE1 and PE2) were found in the centrocephalic (C3, C4) and central midline (Cz) areas, respectively, and first-level and second-level prediction signals (P1 and P2) were found in the central midline (Cz) and frontocentral (FC) areas, respectively. Temporally, both P1 and P2 were activated before the sensory input (S), with the order of P2->P1->S->PE1->PE2. To further examine how predictions were learned during stimulus exposure, we combined our model with Bayesian updating to reveal how prediction errors evolved during this learning process, which was further confirmed in our EEG data. Our findings provide a comprehensive view of how hierarchical predictive coding is dynamically implemented using distinct cortical areas and oscillatory frequencies, and a platform to probe the functional integrity of multi-level prediction in psychiatric disorders. I aim to submit the paper in mid 2021.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
3: やや遅れている
理由
The data collection in 30 healthy subjects were done in 2019, but the data collection in schizophrenic patients are delayed due to (1) my job transition from Kyoto University to the University of Tokyo in 2019, and (2) the COVID-19 pandemic afterward.
|
| 今後の研究の推進方策 |
I will pursue the following projects based on the findings of the current project:
1. Temporal prediction. This project aims to extract prediction signals that encode complex timings within a temporal sequence and examine how they dynamically adjust during the sequence. This project will use human EEG, human ECoG, and marmoset ECoG.
2. Prediction signals in spontaneous activity. This project aims to extract prediction signals during spontaneous activity between epochs of sensory stimulation. This project will use human MEG.
3. Macrocircuit motifs for multimodal prediction. This project will identify macrocircuits in the human brain that learn and encode multilevel and multimodal predictions, and their operation in schizophrenia.
|
