發(fā)布時(shí)間：2018-08-08 14:51

【摘要】：突觸可塑性為神經(jīng)網(wǎng)絡(luò)的學(xué)習(xí)機(jī)制提供了基礎(chǔ)�；趩蝹�(gè)半導(dǎo)體光放大器(SOA)的非線(xiàn)性偏振旋轉(zhuǎn)(NPR)和交叉增益調(diào)制(XGM)效應(yīng)實(shí)現(xiàn)了反脈沖時(shí)間依賴(lài)可塑性(anti-STDP)學(xué)習(xí)機(jī)制。通過(guò)調(diào)整SOA驅(qū)動(dòng)電流,可以實(shí)現(xiàn)長(zhǎng)時(shí)程增強(qiáng)窗口(LTP)和長(zhǎng)時(shí)程抑制窗口(LTD)的高度和寬度調(diào)整,能更好地模擬神經(jīng)網(wǎng)絡(luò)。實(shí)驗(yàn)測(cè)量得到的anti-STDP曲線(xiàn)與生物系統(tǒng)中測(cè)量得到的學(xué)習(xí)曲線(xiàn)相吻合。使用該anti-STDP光路得到的學(xué)習(xí)曲線(xiàn)的時(shí)間窗口約為幾百皮秒,其速度是人類(lèi)大腦STDP學(xué)習(xí)機(jī)制的108倍。由于該anti-STDP光路系統(tǒng)簡(jiǎn)單,且SOA易于與其他器件集成,該anti-STDP光路可以用于實(shí)現(xiàn)大規(guī)模超快神經(jīng)擬態(tài)計(jì)算系統(tǒng)。
[Abstract]:Synaptic plasticity provides the basis for the learning mechanism of neural networks. Based on the nonlinear polarization rotated (NPR) and cross-gain modulated (XGM) effect of a single semiconductor optical amplifier (SOA), the anti-pulse time-dependent plasticity (anti-STDP) learning mechanism is realized. By adjusting the driving current of SOA, the height and width of (LTP) and (LTD) can be adjusted, and the neural network can be simulated better. The anti-STDP curve obtained from the experiment is in agreement with the learning curve obtained in the biological system. The time window of the learning curve obtained by using the anti-STDP optical path is about a few hundred picoseconds, which is 108 times faster than the STDP learning mechanism of the human brain. Because the anti-STDP optical circuit system is simple and the SOA is easy to integrate with other devices, the anti-STDP optical path can be used to realize the large-scale ultra-fast neural pseudo computing system.
【作者單位】： 北京交通大學(xué)理學(xué)院光信息科學(xué)與技術(shù)研究所發(fā)光與光信息技術(shù)教育部重點(diǎn)實(shí)驗(yàn)室;北京交通大學(xué)電氣工程學(xué)院;
【基金】：國(guó)家自然科學(xué)基金(61571035,61401017,61378061)
【分類(lèi)號(hào)】：TN248.4

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 劉大偉;韓玲;韓曉勇;;基于深度學(xué)習(xí)的高分辨率遙感影像分類(lèi)研究[J];光學(xué)學(xué)報(bào);2016年04期

【共引文獻(xiàn)】

相關(guān)期刊論文 前4條

1 劉李;高海亮;潘志強(qiáng);傅俏燕;顧行發(fā);;基于深度學(xué)習(xí)的在軌輻射定標(biāo)方法研究[J];航天返回與遙感;2017年02期

2 李強(qiáng);王智;崔粲;樂(lè)燕思;宋曉佳;孫，

本文編號(hào)：2172133