Dear all, Please find the minutes of the 1st meeting on integrating DL for WSO2ML
There are 2 potential java libraries that can be used for deep learning (Recommended in DeepLearning.net) - DL4J (http://deeplearning4j.org/) - https://github.com/ivan-vasilev/neuralnetworks However, it was decided to stay with *DL4J* as it has shown to be the most promising in terms of the requirements to be satisfied. Furthermore, the initial focus will be on implementing a *Deep Autoencoder (DA)* (A variation of Deep Networks). Deep Autoencoder is essentially a stack of Multi-Layer Perceptrons with greedy layer-wise pre-training and fine-tuning by backpropagation. Deep Autoencoder was chosen based on following reasons - Simplicity (Compared to other networks such as Convolutional Networks, Stacked Restricted Boltzmann Machines) - High Accuracy (DAs has have shown promising results for various datasets. MNIST - 99.0%+ accuracy) - Easy to generalize. DA can be easily generalized for text, images, etc. After the implementation of DA, the possibility of introducing different techniques *(Denoising, Drop-out, Sparcity) *will be investigated. Thank you -- Regards, Thushan Ganegedara School of IT University of Sydney, Australia
_______________________________________________ Dev mailing list Dev@wso2.org http://wso2.org/cgi-bin/mailman/listinfo/dev