Practical  Applications of Deep Learning

Automated systems are being developed to emulate the human learning approach. This feature of Artificial intelligence is known as ‘Deep Learning’. In simple words, deep learning can be considered as a method to automate predictive analytics. It is treated as a sophisticated extension of Machine Learning. If we compare machine learning with deep learning, then we can say that the algorithms used in traditional ML are linear whereas deep learning algorithms are stacked in a hierarchy of increasing complexity and abstraction. Lots of researchers are working on applying deep learning algorithms to automate some manual activities which are unbelievable to be done automatically without human intervention. Some of the interesting applications of Deep Learning are

1.      Automatic Integration of Sounds With Silent Movies

Researchers at MIT have demonstrated an algorithm that has effectively learned how to predict sound. The advance could lead to better sound effects for film and television and robots with improved understanding of objects in their surroundings. For this research,  in training phase of the data, a number of videos(around 1000) representing various sounds (around 46000) were recorded and provided to the system based on deep learning algorithm to deconstruct the sounds to analyze pitch, loudness and other features and then in testing phase to predict the sound of new video the system based on this algorithm, associates the video frames with a database of pre-recorded sound to select the best matching sound.

2.     Automatic Colorization of Black and White Images

In recent past a very old Hindi black and white movie was re-released with colors. Have we ever wondered how it has been possible to color each and every frame of a black and white movie?   It is definitely very time consuming and difficult task to be done by hand with human effort. With the use of Deep learning it can be done to color image in a similar way how a human operator might approach the problem. The deep learning  approach involves the use of very large convolutional neural networks and supervised layers which recreate the image with the addition of color.

3. Automatic Machine Translation

Automatic translation of given words, phrase or sentence in one language, into another language can be done with deep learning. The two specific areas in which deep learning is achieving top results are.

•                 Automatic Translation of Text.
•               Automatic Translation of Images.

With Deep learning algorithms, text translation can be performed without any preprocessing of the sequence. The learning of the dependencies between words and their mapping to a new language can be done through these algorithms.

Identification of letters in images and converting them into text and then recreation of text in images are done with convolutional neural networks which is also called instant visual translation. This task requires the classification of objects within a photograph as one of a set of previously known objects. Amazing results have been achieved using deep learning on benchmark examples of this problem using very large convolutional neural networks.

4. Object Classification and Detection in Photographs

This task requires the classification of objects within a photograph as one of a set of previously known objects. Another variation of this task which is more complex and called object detection involves identifying specifically one or more objects within the scene of the photograph and drawing a box around them.

5. Automatic Handwriting Generation

Generating new handwriting for a given word or phrase from a given corpus of handwriting examples can also be performed using deep learning. First the handwriting is provided as a sequence of coordinates used by a pen during creation of handwriting samples. From this corpus the relationship between the pen movement and the letters is learned and new random examples can be generated. The most remarkable part is the learning of different styles and then their impersonation.

6. Automatic Text Generation

In this very interesting task, a corpus of text is learned and from this model new text is generated, word-by-word or character-by-character. This is a very powerful model and is capable of learning how to spell, punctuate, form sentences and even capture the style of the text in the corpus.

7. Automatic Image Caption Generation

A capable system can be made to generate  a caption describing the contents of a given image. Many researchers came up with various deep learning algorithms in 2014 for solving this problem and had achieved very impressive results.

8. Automatic Game Playing

In this task, a model learns how to play a computer game based on the pixels on the screen only. This is a complex and very difficult task in the domain of deep reinforcement models and is the breakthrough that DeepMind (now part of google) is well-known for achieving.

Summary

In recent years, Deep Learning has become a buzz word. It has been the most researched and talked about topic in data science. It is due to some of the recent breakthroughs in data science which have been provided by the deep learning. According to predictions by researchers, many deep learning applications will affect our life in the near future and we must be prepared to witness this revolutionary technology.

Prof. (Dr.) Meenakshi Narula

Department of Information Technology