Setup and train deep convolutional network for speech-to-emotion with input window over the features extracted by +Nicolas and max-pooling on the output sequence. Train it first on the challenge data, then add the extra data to be collected by +Guillaume A. Test on the challenge validation set. Follow the pattern setup by Yann for the deep MLP issue.
Try max-pooling the hidden layer rather than the output layer, i.e., following the sequence-wide pooling by a linear (or non-linear) classifier. There should also be fixed-pool-size pooling layers (with subsampling = strides) inside the convolutional net, as usual for such nets.
Setup and train deep convolutional network for speech-to-emotion with input window over the features extracted by +Nicolas and max-pooling on the output sequence. Train it first on the challenge data, then add the extra data to be collected by +Guillaume A. Test on the challenge validation set. Follow the pattern setup by Yann for the deep MLP issue.
Try max-pooling the hidden layer rather than the output layer, i.e., following the sequence-wide pooling by a linear (or non-linear) classifier. There should also be fixed-pool-size pooling layers (with subsampling = strides) inside the convolutional net, as usual for such nets.