Vector to sequence generative machine learning model?

Hi all,

I'm looking for a vector to sequence machine learning model, which is trained with initial condition vectors and output sequences.

So that in the end it generates a sequence corresponding to a test initial condition vector.

Is there any model capable of doing this?

Answer 1

[sanidhyak]

Hi @Giacomo Notaro,

I understand that you are looking to train a model that takes an initial condition vector as input and outputs a corresponding sequence. This is a classic case of a "sequence generation" problem, which can effectively be modeled using "LSTM-based neural networks" provided by MATLAB’s Deep Learning Toolbox. Do note that there is no "one-line" or "prebuilt function" for "vector-to-sequence" generation.

Kindly refer to the following sample code to build such a model:

% Example input: Initial condition vectors
XTrain = {rand(10,1), rand(10,1), rand(10,1)};  % Cell array of 10x1 vectors

% Example output: Corresponding sequences
YTrain = {rand(1,20), rand(1,20), rand(1,20)};  % Cell array of 1x20 sequences

% Define network layers
layers = [
    sequenceInputLayer(10)                        % Input size = 10
    fullyConnectedLayer(100)
    lstmLayer(100,'OutputMode','sequence')
    fullyConnectedLayer(1)
    regressionLayer];

% Specify training options
options = trainingOptions('adam', ...
    'MaxEpochs', 100, ...
    'MiniBatchSize', 16, ...
    'SequenceLength', 'longest', ...
    'Shuffle', 'every-epoch', ...
    'Verbose', false, ...
    'Plots', 'training-progress');

% Train the model
net = trainNetwork(XTrain, YTrain, layers, options);

% Predict output sequence for a new input vector
newInput = rand(10,1);
YPred = predict(net, newInput);

This setup allows the network to learn the mapping from static condition vectors to dynamic output sequences. The "sequenceInputLayer" and "lstmLayer" components are particularly well-suited for such temporal prediction tasks.

For further reference, please refer to the following official documentation:

• https://www.mathworks.com/help/deeplearning/ref/lstmlayer.html
• https://www.mathworks.com/help/deeplearning/examples/sequence-to-sequence-regression-using-deep-learning.html

Cheers & Happy Coding!