Models for Evaluating Sequentail Memories in a Recurrent Neuronal Network

行为事件序列的存储是记忆组织的基本策略。在神经元水平上存储和保持顺序记忆的计算算法还没有被很好地理解。顺序记忆可以通过一个递归网络实现，因为它位于海马体的CA3区，这是一个关键的大脑结构，参与在线获取情节记忆。CA3中被认为存储顺序记忆的经常性突触连接是高度可塑性的，对神经调节剂特别敏感。因此，CA3区锥体神经元的重新连接，以及行为序列记忆的改变，取决于下游区域的行为评估。首先，我们打算从理论上评估递归神经网络模型中不同记忆轨迹被赋予不同行为重要性(值)的情况下的保留时间。在第二步中，我们开发了一个生物学上可信的理论模型，用于现有顺序记忆的值依赖修改。第三，我们通过研究如何存储序列的时间层次来解决行为序列需要在任意持续时间的时间间隔上进行关联的问题。

The storage of sequences of behavioral events is a fundamental strategy of memory organization. The computational algorithms underlying the storage and retention of sequential memories on the neuronal level are not well understood. Sequential memory can be realized through a recurrent network as it is found in the CA3 region of the hippocampus, a brain structure critically involved in the online acquisition of episodic memories. The recurrent synaptic connections that are thought to store sequential memories in CA3 are highly plastic and particularly sensitive to neuromodulators. Rewiring of CA3 pyramidal neurons, and hence changes of the memory of behavioral sequences, therefore depends on the behavioral evaluation in down-stream regions. At first, we intend to theoretically assess retention times in a recurrent neuronal network model for a situation in which different memory traces are assigned different behavioral importance (value). In a second step, we develop a biologically plausible theoretical model for value-dependent modification of existing sequential memories. Third, we address the problem that behavioral sequences require associations over time intervals of arbitrary duration by investigating how to store temporal hierarchies of sequences.