ABSTRACT
For the benefit of designing scalable, fault resistant optical neural networks (ONNs), we investigate the effects architectural designs have on the ONNs' robustness to imprecise components. We train two ONNs - one with a more tunable design (GridNet) and one with better fault tolerance (FFTNet) - to classify handwritten digits. When simulated without any imperfections, GridNet yields a better accuracy (â¼98%) than FFTNet (â¼95%). However, under a small amount of error in their photonic components, the more fault tolerant FFTNet overtakes GridNet. We further provide thorough quantitative and qualitative analyses of ONNs' sensitivity to varying levels and types of imprecisions. Our results offer guidelines for the principled design of fault-tolerant ONNs as well as a foundation for further research.
ABSTRACT
Cortico-hippocampal interactions during sleep are believed to reorganize neural circuits in support of memory consolidation. However, spike-timing relationships across cortico-hippocampal networks-key determinants of synaptic changes-are poorly understood. Here we show that cells in prefrontal cortex fire consistently within 100 ms after hippocampal cells in naturally sleeping animals. This provides evidence at the single cell-pair level for highly consistent directional interactions between these areas within the window of plasticity. Moreover, these interactions are state dependent: they are driven by hippocampal sharp-wave/ripple (SWR) bursts in slow-wave sleep (SWS) and are sharply reduced during REM sleep. Finally, prefrontal responses are nonlinear: as the strength of hippocampal bursts rises, short-latency prefrontal responses are augmented by increased spindle band activity and a secondary peak approximately 100 ms later. These findings suggest that SWR events are atomic units of hippocampal-prefrontal communication during SWS and that the coupling between these areas is highly attenuated during REM sleep.
