1: \begin{abstract}

2: 

3: 

4: A new learning algorithm proposed by Brandt and Lin for neural

5: network \cite{brandt1996supervised, lin2020supervised} has been

6: shown to be mathematically equivalent to the conventional

7: back-propagation learning algorithm, but has several advantages

8: over the back-propagation algorithm, including

9: feedback-network-free implementation and biological plausibility.

10: In this paper, we investigate the convergence of the new

11: algorithm. A necessary and sufficient condition for the algorithm

12: to converge is derived. A convergence measure is proposed to

13: measure the convergence rate of the new algorithm. Simulation

14: studies are conducted to investigate the convergence of the

15: algorithm with respect to the number of neurons, the connection

16: distance, the connection density, the ratio of

17: excitatory/inhibitory synapses, the membrane potentials, and the

18: synapse strengths.

19: 

20: {\em Index Terms} --- Neural Networks, Back-propagation, Deep

21: Learning, Learning Algorithm

22: 

23: \end{abstract}

24: