英国伦敦大学学院Laurence Aitchison等研究人员揭示突触可塑性背后的贝叶斯推理。这一研究成果于2021年3月11日在线发表在国际学术期刊《自然—神经科学》上。
研究人员表示，学习，尤其是快速学习，对于生存至关重要。但是，学习是困难的；大量的突触权重必须根据嘈杂的，且常常是模棱两可的感觉信息来设置。在这样的高噪声环境中，跟踪权重上的概率分布是最佳策略。
研究人员假设突触采取了这种策略：本质上，当它们估计权重时，误差值也被考虑其中。然后，突触利用不确定性来调整学习率；权重越不确定，学习率就越高。研究人员还提出了第二个独立的假设：突触通过将其不确定性与突触后电位大小的可变性联系起来，从而传达其不确定性；更多的不确定性导致更大的可变性。
这两个假设将突触可塑性视为贝叶斯推理的一个问题，因此提供了一种学习的规范性观点。突触能够概括已知的学习规则，为突触后电位的大小变化提供了解释，并做出可证伪的实验预测。
附：英文原文
Title: Synaptic plasticity as Bayesian inference
Author: Laurence Aitchison, Jannes Jegminat, Jorge Aurelio Menendez, Jean-Pascal Pfister, Alexandre Pouget, Peter E. Latham
Issue&Volume: 2021-03-11
Abstract: Learning, especially rapid learning, is critical for survival. However, learning is hard; a large number of synaptic weights must be set based on noisy, often ambiguous, sensory information. In such a high-noise regime, keeping track of probability distributions over weights is the optimal strategy. Here we hypothesize that synapses take that strategy; in essence, when they estimate weights, they include error bars. They then use that uncertainty to adjust their learning rates, with more uncertain weights having higher learning rates. We also make a second, independent, hypothesis: synapses communicate their uncertainty by linking it to variability in postsynaptic potential size, with more uncertainty leading to more variability. These two hypotheses cast synaptic plasticity as a problem of Bayesian inference, and thus provide a normative view of learning. They generalize known learning rules, offer an explanation for the large variability in the size of postsynaptic potentials and make falsifiable experimental predictions. We propose that synapses compute probability distributions over weights, not just point estimates. Using probabilistic inference, we derive a new set of synaptic learning rules and show that they speed up learning in neural networks.
DOI: 10.1038/s41593-021-00809-5
Source: https://www.nature.com/articles/s41593-021-00809-5
Nature Neuroscience：《自然—神经科学》，创刊于1998年。隶属于施普林格·自然出版集团，最新IF：28.771
官方网址：https://www.nature.com/neuro/
投稿链接：https://mts-nn.nature.com/cgi-bin/main.plex
本期文章：《自然—神经科学》：Online/在线发表
