First, you need to have the objective function.

Dear Saurabh Kumar ,

You can use PSO as the optimizer for training Autoencoder only if your Autoencoder is feedforward (non-recurrent neural network). The way you do it:

+ Determine input, output and number of hidden layers, number of nodes in each hidden layer.

+ Determine the objective function, possibly MSE, MAPE, MAE, ...

+ Encoding input, hidden, and output weights into 1 solution (1 particle in swarm set)

+ Then implement PSO.

You can read more articles below for more details:

https://ieeexplore.ieee.org/abstract/document/8599578

Python code is available here:

https://github.com/chasebk/code_FLNN

Thieu Nguyen sir I am working in python i am unable to understand how to use PSO during during training of Autoencode. I have MxN data set i.e. M particles and N dimensions. I want use fitness function as MSE I am unable to use PSO as optimizer. please help me out. There is also a library of PSO but i am not understanding how to use my fitness function mse in this. Thanks

I am also providing this. please tell me how to use mse as fitness function. the way i have used is not correct.

options = {'c1': 2, 'c2': 2, 'w':1.4}

costFunc = 1/1+mse

optimizer = ps.single.GlobalBestPSO(n_particles=562, dimensions=24, options=options)

best_cost, best_pos = optimizer.optimize(fx.costFunc, iters=100).minimize(costFunc)