Removing the Middle-Man
There are several apps out there today that use deep learning models to identify pictures uploaded to its platforms. One example comes from the show “Silicon Valley”, where one of the characters created an app that was advertised to be able to identify food. However, due to some of the limitations of deep learning (and laziness of the creator), the app could only identify whether or not the food item was a hot dog.
Deep learning is a subset of machine learning that can be used for things such as image recognition and classification. Specifically, it is able to remove the manual efforts of identifying image features and allows for the computer to learn these features on its own. Deep learning models are trained by using very large sets of data and neural network architectures that contain many layers and are sometimes considered to be difficult to debug.
Amount of Data:
To properly train a deep learning model, you need a very large and diverse data set. Additionally, this data needs to be labeled data, which is data that has been tagged with the information of which you are trying to predict.
Training these deep learning models requires substantial computing power. This makes specific types of hardware necessary to be able to handle these computations. This article posted by Tim Dettmers is a great hardware guide: http://timdettmers.com/2018/12/16/deep-learning-hardware-guide/
As deep learning models are sometimes defined as ‘blackbox’, it could be very difficult to trace a classification back to which features were important.
While deep learning does remove the middle-man by automating some of the work traditionally performed through feature engineering, it does still require significant effort in data collecting, cleaning, and transformation.
By: Monica Kay Royal