Malfunctioning neurons in the brain sometimes operate synchronously, reportedly causing many neurological diseases, e.g. Parkinson’s. Suppression and control of this collective synchronous activity are therefore of great importance for neuroscience, and can only rely on limited engineering trials due to the need to experiment with live human brains. We present the first Reinforcement Learning (RL) gym framework that emulates this collective behavior of neurons and allows us to find suppression parameters for the environment of synthetic degenerate models of neurons. We successfully suppress synchrony via RL for three pathological signaling regimes, characterize the framework’s stability to noise, and further remove the unwanted oscillations by engaging multiple PPO agents.

We consider an online, processor sharing system with a single server in a packet-switched network where the processor multiplexes packets from multiple input ports onto the same output link. Under a fluid model, in which incoming data is treated as infinitely divisible, the multiplexer allocates its output capacity to serve fractions of packets from all input ports simultaneously (which we call a fluid policy). However, due to the packet nature of the network traffic, a multiplexer scheduler must approximate this fluid policy by a schedule in which only whole packets are sent (which we call a packetized policy). We assume that at each time instant, the aggregate service rate demanded by a fluid policy is no greater than the output link capacity. Then, for a single server with N input ports, and a scheduler using speedup s of at least 2 and a lookahead window of L/spl ges/0 time steps, we prove bounds on worst-case maximum additional delay D. In particular, we show that ([Ne/sup -s/]-L-2)/sup +//spl les/D min {([Ne/sup 1-s/]-L)/sup +/, [Ne/sup -s/]}. In the case when no lookahead is used (L=0), a tighter bound can be obtained: ([Ne/sup -s/]-2)/sup +//spl les/D/spl les/[Ne/sup -s/}.

We consider a type of on-line, traffic scheduling problem in input queued, crossbar switches. The input to a problem, at each time step, is a set of desired traffic rates. These traffic rates in general cannot be exactly achieved since they assume arbitrarily small fractions of packets can be transmitted at each time step. The goal of the traffic scheduling problem is to closely approximate the given sequence of traffic rates by a sequence of switch uses in which only whole packets are sent. The focus of this paper is bounding the costs incurred in using such an approximation, in terms of the additional buffer size required. We establish universal bounds on the additional buffer size due to sending only whole packets; these bounds do not depend on the particular distribution of the input traffic, require no speedup, and guarantee 100% throughput. Specifically, for an N /spl times/ N input queued, crossbar switch, an on-line, packetizing algorithm is presented that guarantees 100% throughput with a buffer requirement of (N + 1)/sup 2//4 packets per input port with no speedup. The algorithm can he improved to run in O(N log N) time, using a fast algorithm for edge-coloring bipartite multigraphs. In the reverse direction, it is shown for an N /spl times/ N input queued, crossbar switch, that any on-line, packetizing algorithm with no speedup requires a buffer size of N/e - 2 packets per input port. We also extend the main packetizing algorithm in this paper to a general class of switch architectures.