When compared to other architectures, where the bottleneck of the system is memory usage, the CRNN outperforms the others. CRNN consistently outperforming the k2c2 model - which shares the same 2D structure - may indicate that learning a global structure is more important than focusing on local structures for summarization.

Memory Usage in CNN/CRNN Model Comparison

In the figure shown, where higher AUC is better, CRNN outperforms k2c1 in 44/50 tags, and k2c1 outperforms k1c2 in 48/50 tags.

AUC in CNN/CRNN Model Comparison

When compared to other architectures in terms of computation time, k2c2 and CRNN have similar results and are better than k1c1 and k2c1. Training speed, however, is ranked as such: k2c1 > k2c2 > k1c2 > CRNN. There are two factors that likely lead to this (child nodes).

Computation Time in CNN/CRNN Model Comparison

The following children nodes delve into the experimental results of the "CRNN for Music Classification" paper. The paper compares the CRNN architecture in this context to the CNN architectures k1c2, k2c2, and k2c1. For more details on these models, see the models child node under "CRNNs for Music Classification" parent node.

University of Connecticut

The children of this node are results of the paper linked to by reference "CONVOLUTIONAL RECURRENT NEURAL NETWORKS FOR MUSIC CLASSIFICATION." It compares common CNN structures to a CRNN structure in the context of music classification.

CRNNs for Music Classification

Choi, K., Fazekas, G., Sandler, M., & Cho, K. (2017, March). Convolutional recurrent neural networks for music classification. In 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 2392-2396). IEEE.

CONVOLUTIONAL RECURRENT NEURAL NETWORKS FOR MUSIC CLASSIFICATION

Experimental Results of "CRNN for Music Classification" Paper

Four different models were compared: One CRNN and 3 CNNs.  All convolutional and fully connected layers have identical optimization techniques and activation functions - batch normalization and ELU activation function. The CRNN architecture can be compared to the CNN architectures (k1c2, k2c1, k2c2) where 'k' specifies the kernel shape and 'c' specifies the convolutional dimension.

Learn Before

Related

Learn After