I am working on neural networks and having matlab 2021a and 2023a, but i face invalid solver name option while using optimization techniques.

I am working on neural networks and having matlab 2021a and 2023a versions, but i face invalid solver name option while using optimization techniques like adadelta, adagrad, trainbr,trainlm,traingdx, trainscg, trainbfg with LSTM. how i solve this problem?

options = trainingOptions('trainbr', ...
    'MaxEpochs', 10, ...
    'MiniBatchSize', 10, ...
    'GradientThreshold', 1, ...
    'Verbose', 1, ...
    'Plots', 'training-progress');

it gives invalid solver name.

Answer 1

[Ayush]

Hi @Muhammad Shoaib

I understand that you are getting invalid solver name option while using optimization techniques like adadelta, adagrad, trainbr, trainlm, traingdx, trainscg, trainbfg. Till R2023b release solver name supported for training Deep learning Neural Networks are:

1. sdgm
2. rmsprop
3. adam
4. lbfgs

You may refer to documenation for more information: https://www.mathworks.com/help/deeplearning/ref/trainingoptions.html#mw_4f311de8-ad53-4201-ac1a-47f66e0de52d

Training function supported for shallow neural networks are given here: https://www.mathworks.com/help/deeplearning/ref/fitnet.html#bu2w2vc-1-trainFcn

Also, you may refer to file exchange for other optimization techniques from here: https://www.mathworks.com/matlabcentral/fileexchange/71069-gradient-descent-optimization

Thanks,

Ayush

[Muhammad Shoaib]

@Ayush thanks

but can you tell me how to add these given function in matlab library or how to file exchange for the given optimization techniques?.

Also, you may refer to file exchange for other optimization technniques from here: https://www.mathworks.com/matlabcentral/fileexchange/71069-gradient-descent-optimization

[Ayush]

Hi @Muhammad Shoaib MATLAB built-in function trainingOptions will only expect the type of inputs specified in the documentation. However, if you want to define the custom network you may refer to the link: https://www.mathworks.com/help/deeplearning/ug/train-network-using-custom-training-loop.html