Till now what I have experienced that second order method such as Scaled conjugate gradient methods provide better convergence and it is faster too. How ever if you are thinking for a evolutionary method you can try with differential evolution and bio geography based optimization. Hope it will work better in smaller size ANN.
Gradient Evolution, Triangular DE are some evolutionary algorithms, and used for optimize parameters of NN like initial weights and bias. Some application of artificial intelligence used this algorithm like Uncertain system control, Nonlinear system approximation, Time series analysis, Prediction and Forecasting etc. Please see the following link:
the phrase " evolutionary algorithms " is limited.It is better to utilize "Meta-heuristic algorithms". By this consideration, I think PSO (especially APSO) is one of the high speed algorithms for this category.
you should remember, the convergence speed is not the only problem. for instance, trapping in the local minimum is a usual disadvantages of this algorithm.
in addition, if you want newer algorithms, I suggest the links below:
From my experiences, I think using hybrid methods made good results.
For using hybrid method, try to hybrid two methods which one of them has a high performance in Exploration and the other in Exploitation. If you select two proper algorithms, it will be have good results.
For example, PSO has high speed and made good results for Exploitation, while:
GA has a high accuracy in Exploration; Since good hybridizing these two algorithms can made good results.
Of course, I suggest you to utilize new algorithms. GA-PSO is almost an old method.
Sure, cuckoo search algorithm has many parameters for convergence control, but too many parameters does not mean that complexity of the algorithm is getting high ?