The influence of attractor topology on seizure initiation in the hippocampal region

本研究项目的目的是阐明不同海马区回路的计算目的与其癫痫易感性之间的关系。特别是，这笔赠款旨在部分资助一名研究人员，该研究人员将专注于使用机器学习方法，将各种神经网络中的计算目的与癫痫敏感性联系起来。事实上，在这笔助学金的帮助下，我聘请了一位机器学习专家，但加入我的团队的不是研究生，而是一位名叫伊斯梅尔·拉姆赞博士的博士后。他在实现这笔赠款的目标方面取得了长足的进步。Ramzan设计了两个机器学习任务，模拟两个不同的海马区CA3和内嗅皮层(EC)的计算角色。然后，他对训练过的递归神经网络施加了诱发癫痫的扰动，并评估了它们对癫痫的敏感性。初步结果表明，类EC网络发生癫痫的风险更大。此外，Ramzan博士已经开始探索将任务结构与癫痫敏感性联系起来的潜在机制。他通过推断训练有素的网络的能量格局，并比较两种情况下低能量流形的拓扑结构来做到这一点。

The goal of this research project is to elucidate how the computational purpose of different hippocampal circuits is related to their susceptibility for seizures. In particular, this grant is intended to partially fund a researcher who would focus on using machine learning approaches to connect computational purpose with seizure susceptibility in various neural networks. Indeed, I have been able to hire a machine learning expert with the help of this grant, but instead of a graduate student, a postdoc named Dr. Ismaeel Ramzan has joined my group. He has made considerable progress towards the aims of this grant.Dr. Ramzan has devised two machine learning tasks that mimic the computational role of two different hippocampal regions called CA3 and entorhinal cortex (EC). He has then applied a seizure-inducing perturbation upon the trained recurrent neural networks and has assessed their susceptibility to seizures. Preliminary results indicate that EC-like networks are at greater risk for developing seizures. Moreover, Dr. Ramzan has begun to probe the underlying mechanisms that connect task structure with seizure susceptibility. He does so by inferring the energy landscape of trained networks and comparing the topologies of the low-energy manifolds between the two cases.