Table 1.

Comparison between deep multimedia and traditional learning methods [43]

Deep multimedia learning methods	Traditional learning methods
Features are learned from the fused data	Prior knowledge is required to extract features manually
Little preprocessing of the input data is required	Preprocessing is needed for early fusion
Dimensionality reduction is done by the architecture	It requires perform dimensionality reduction
Performs early, intermediate, or late fusion	Performs early or late fusion
Fusion architecture is learned during training	Fusion architecture is designed manually
Requires many data for training	Not a lot of training data is required
GPUs are need for training time	Use of GPUs are not critical
Hyper parameter tuning is needed	Hyper parameter tuning is not needed