Jun, 2020
一种实用的稀疏逼近算法用于实时循环学习
A Practical Sparse Approximation for Real Time Recurrent Learning
TL;DR介绍了Sparse n-step Approximation(SnAp)来优化Real Time Recurrent Learning(RTRL)的影响矩阵,在保持计算成本低的同时提高了网络的学习速度,特别是在稀疏网络时,n=2时的SnAp能够优于backpropagation in terms of learning speed when updates are done online。