Wednesday, July 12, 2017

Deep Learning – Moving Machines Closer to Attaining Artificial Intelligence




introduction
·         In recent years, Deep Learning has become an area of active research, and has also become a buzzword that has been tossed around a lot;
·         Deep Learning is becoming increasingly important these days because it teaches computers the skills our brains do naturally e.g. recognizing handwritten digits;
·         This posting is focused primarily to those with little or no understanding of Deep Learning.  The main goal is to increase understanding of this topic through simple language, and to point out some free online courses and resources that are available for a better understanding of Deep Learning.

DEEP LEARNING & MACHINE LEARNING
·         “Deep Learning” comes under the broader field of “Machine Learning” which was covered in the previous post; 
·         Deep Learning is a specific type of Machine Learning, focusing on high order statistics1;
·         Deep Learning is a new area of Machine Learning research. It was introduced with the objective of moving Machine Learning closer to one of its original goals, Artificial Intelligence2;
·         Please consider having another quick look at the previous post on Machine Learning (July 1, 2017 – see below).

WHAT IS DEEP LEARNING?  We hope to arrive at some understanding of Deep Learning by considering how this activity has been described in various publications:
·         Deep Learning has been described as follows3:
o   Deep Learning is a paradigm (example/ pattern/ model) for performing Machine Learning;
o  Deep Learning is a form of Machine Learning that uses a model that’s very much inspired by the structure of the brain's computing.  (Hence we call this model a neural network); and
o   Deep Learning is an extremely powerful tool for modern Machine Learning;
·         Deep Learning is a branch of Machine Learning that uses algorithms* to do things like recognizing objects and understanding human speech4. [*Algorithms = Processes or set of rules to be followed in solving problems];
·         Deep Learning refers to  learning tasks of artificial neural networks (ANNs) that contain more than one hidden layer. (Deep learning is part of a broader family of Machine Learning methods based on learning data representations, as opposed to task specific algorithms)5;
·        In Deep Learning, a software attempts to mimic the brain activity in layers of neurons in the neocortex, (the wrinkly 80 percent of the brain where thinking occurs).  The software learns in a very real sense, to recognize patterns in digital representations of sounds, images, and other data6. [In the human brain, the neocortex has six layers and contains between 10 and 14 billion neurons];
·         Deep Learning employs numerous, similar, yet distinct, deep neural network architectures to solve various problems in natural language processing, computer vision, bioinformatics, and many other fields7.

PREREQUISITES
·         The prerequisites for understanding and applying Deep Learning are8:
o   Linear algebra;
o   Calculus;
o   Statistics;
o   Programming; and
o   Some Machine Learning.

APPLICATIONS
·         Deep Learning is responsible for recent advances in4:
o   Speech recognition;
o   Computer vision;
o   Natural language processing; and
o   Audio recognition.
·         Apart from speech recognition, Deep Learning methods have dramatically improved the state-of-the-art in9:
o   Visual object recognition;
o   Object detection;
o   Drug discovery;
o   Genomics; and
o   Many other domains.
·         Some inspirational examples of Deep Learning include10:
o   Colorization of Black and White Images;
o   Adding Sounds to Silent Movies;
o   Automatic Machine Translation;
o   Object Classification in Photographs;
o   Automatic Handwriting Generation;
o   Image Caption Generation; and
o   Automatic Game Playing.

FREE ONLINE COURSES & RESOURCES
·         For free online courses and resources relating to Deep Learning, please click here.

REFERENCES
2.       http://deeplearning.net/



Posted by Dr. Nat Tuivavalagi




No comments: