In this article, we will see how deep learning is used in Data Science.
Table of contents
- What is deep learning?
- Uses of deep learning
What is deep learning?
Deep learning is a subset of machine learning. It mainly revolves around artificial neural networks which are inspired by the structure and function of the human brain. Unlike machine learning, deep learning eliminates data pre-processing and is capable of working with unstructured data like images and automatically extract necessary features, thus reducing the machine's dependency on humans. The image below shows the difference between machine learning and deep learning.
Uses of deep learning
Since machine learning is an integral part of data science, this essentially makes deep learning too an important part of data science. In the field of big data analytics, deep learning can be used to extract complex patterns from huge amounts of data, data tagging, fast retrieval of information and for various other tasks. Through a hierarchical learning process, deep learning algorithms are capable of extracting complex abstractions as data representation. With the knowledge obtained from these complex data representations, simple linear models work even more efficiently.
In the real world, deep learning is used for image segmentation, analysis and visual recognition. Deep learning algorithms can sort pictures based on locations and faces detected in them, according to dates or events.
It is used to colorize grayscale images and also to add appropriate soundtracks to silent videos.
Wimbledon 2018 even harnessed deep learning to generate highlights. They used IBM Watson to analyze player emotions and then used deep learning to factor in match popularity, player popularity and response of audience to auto-generate highlights.
Long short term memory recurrent neural networks are used in speech-to-text conversions and are an important part of virtual assistants. RNNs are used so that virtual assistants can catch the linguistic nuances and frame correct responses. They evaluate human language and learn from them to execute the command given. Given below is an example of how speech commands can be recognized using recurrent neural networks.
Deep learning can also be used for semantic indexing so that massive amounts of data can be stored efficiently. It makes it possible to get semantic features from high-dimensional data, which reduces dimensions of the complex data representations. It can also be used for document summarization. Image below shows how semantic indexing can be done using deep level-wise learning that is able to classify multiple labels.
One application where data science and deep learning go hand in hand is self-driving cars. Millions of real-time data from cameras, sensors, geo-mapping are fed in continuously and are modeled using deep learning algorithms to make decisions based on the environment of the vehicle such as navigating through traffic, road blocks, identifying routes and signs that are put up (no parking, no horn, school zone).