ORA (Round 5): Hidden brain states underlying efficient representations in working memory

人类的工作记忆不仅用于跟踪我们周围的变化，而且对我们的学习能力也至关重要。说到工作记忆的本质，有两个事实是毋庸置疑的：首先，它是出了名的有限，在任何给定的时间里只能记住少量的东西。其次，需要通过将备忘录组织成对象和事件的结构化表示来优化这种宝贵的有限资源。然而，除了这些基本事实之外，工作记忆中存储空间的性质并没有得到很好的定义，目前还存在激烈的争论。到目前为止，即使是对大脑活动的先进测量也无法提供决定性的证据。测量工作记忆的神经关联的基本挑战阻碍了对记忆限制如何产生以及如何缓解的更好理解。过去的研究假设工作记忆表现在持续的神经活动中；然而，我们认为工作记忆的神经编码实际上比这种静态活动状态要深入得多。直到最近，还没有明显的方法来探索持续延迟活动之外的工作记忆表征，但我们开发的一种新技术（“神经声纳”）允许对神经网络的潜在隐藏状态进行新的观察。我们将运用我们的新方法来确定工作记忆容量限制的基本性质，以及如何优化编码、维护和工作记忆引导的决策，以获得卓越的记忆性能。

Human working memory not only serves to keep track of changes around us, but is also vital to our ability to learn. When it comes to the nature of working memory, two facts are beyond doubt: First, it is notoriously limited and can hold only a handful of items at any given time. Second, this precious limited resource needs to be optimised, by organising memoranda into structured representations of objects and events.Beyond these basic facts, however, the nature of storage space in working memory is not well defined and is currently heavily debated. Even advanced measures of brain activity have so far not been able to provide decisive evidence. Fundamental challenges in measuring the neural correlates of working memory impede a better understanding of how memory limits arise and how they might be alleviated.Past research has assumed working memory is represented in persistent neural activity; however, we argue that neural coding for working memory actually goes far deeper than such static activity states. Until recently, there was no obvious method to explore working memory representations beyond persistent delay activity, but a new technique ('neural sonar') we developed allows a fresh perspective on the underlying, hidden states of the neural network. We will apply our new approach to determine the fundamental nature of capacity limits in working memory, and how encoding, maintenance and working memory-guided decision making can be optimised for superior memory performance.