【Day36 文献泛读】Bayesian integration in sensorimotor learning

阅读文献:

Körding, K. P., & Wolpert, D. M. (2004). Bayesian integration in sensorimotor learning. NATURE, 427(6971), 244-247. doi:10.1038/nature02169

文献链接:

Bayesian integration in sensorimotor learning | Nature

Abstract

1) According to bayesian theory, an optimal estimate results from combining information about the distribution of velocities—the prior—with evidence from sensory feedback.

【Day36 文献泛读】Bayesian integration in sensorimotor learning

参见:一个例子搞清楚(先验分布/后验分布/似然估计)_神仙一样的帅哥-CSDN博客_后验分布

2) As uncertainty increases, the system should increasingly rely on prior knowledge.

3) The central nervous system employs probabilistic models during sensorimotor learning.


【Day36 文献泛读】Bayesian integration in sensorimotor learning 

【Day36 文献泛读】Bayesian integration in sensorimotor learning 

【Day36 文献泛读】Bayesian integration in sensorimotor learning 

Result 1:

An examination of the theoretically determined mean squared error for the three models shows that it is minimal for bayesian model (in the middle of Fig.1e), in which subjects could optimally use information about the prior distribution and the uncertainty of the visual feedback to estimate the lateral shift.

【Day36 文献泛读】Bayesian integration in sensorimotor learning

Result 2:

1) Slope that increases with increasing uncertainty (Fig. 2b) is compatible with bayesian model (Fig. 2a).

2) The bias and the slope should have a fixed relationship (Fig. 2c) so that subjects do bayesian estimation: they combine prior knowledge of the distribution with sensory evidence to generate appropriate compensatory movements.

3) The true prior (red line) was reliably learned by each subject (Fig. 2d).

【Day36 文献泛读】Bayesian integration in sensorimotor learning

Result 3:

From Fig 3, subjects can learn complex distributions (bimodal distribution in this experiment) and  represent the bimodal prior, which means they can use bayesian statistics.


Conclusions

1) We show quantitatively that the system performs optimally would require a direct measure of sensory uncertainty before it is integrated with the prior.

2) By imposing experimentally controlled priors we have shown that our results qualitatively match a bayesian integration process, and we expect that such a bayesian process might be fundamental to all aspects of sensorimotor control and learning.

上一篇:[半监督学习] AggMatch: Aggregating Pseudo Labels for Semi-Supervised Learning


下一篇:简述主动学习Active Learning