how to train a CNN with non image data which is in a vector or 1-channel matrix form?

[Majid Azimi]

Hi

I have a dataset of mocap (motion caption) positions as vectors which I am going to train a DNN for this dataset.

the sample data would be like a 140-D dimension vector. 

Is it possible to train a CNN for this kind of data?

how to use convolution layers for this kind of data as kernels are e.g.5x5 while the data is  a vector?

If I make the data in a form of matrix, is it possible to train a pretrained CNN e.g. alexnet for this dataset?

Best

Majid

[Jan]

I don't think it makes any sense to use 2D convolutions if your data is actually 1D. You could do it by artificially reshaping your data to 2D (e.g. 140 -> 10x14), but as there is no real relation between the subsequent rows I don't think 2D convolutions are a meaningful thing to do. You could also use 1D convolutions on the 1D data, but I am not sure whether that makes sense either, since neighboring data dimensions do not really relate to each other. Actually I am not sure how well-suited that task of yours is for deep learning. If the time dimension is of interest (which I assume for mocap), maybe you could try a recurrent approach?

Jan