Deep Learning For Opaque Data: A user-friendly tool to build a Machine Learning classifier
From object recognition to machine translation, deep learning is being used to automate many complex tasks due to its impressive performance.
However, training a deep learning model for a new dataset is challenging. The number of potential deep learning models to fit a dataset is large due to the required complexity to learn the direct mapping from raw input to output, hence expertise is needed to find a suitable deep learning model among many candidates.
We created a demonstrator to allow industry partners to apply deep learning on new labeled datasets in an intuitive manner; it has a simple user interface to facilitate non-expert use of deep learning. The demonstrator includes a default deep learning model which performs well on a diverse set of datasets that we tried. For better performances, it allows the users to specify custom deep learning models for their datasets .
Our demonstrator makes it simple and straightforward for non-expert users to evaluate whether deep learning can be effective on their own labeled datasets. The software allows the user to build a machine learning classifier without going through the feature extraction process which is time consuming and requires deep domain knowledge. For experienced users, the demonstrator enables the user to quickly evaluate different deep learning architectures.