“Deep learning” a relatively new field of artificial intelligence research promises general, powerful, and fast machine learning, moving us one step closer to AI. It tries to mimic the activity of the brain by using so-called neural networks.
But how did it turn from an uncertain academic topic into one of tech’s most exciting fields in under a decade? What is its business impact? How different is it from Machine learning? Which top organisations are working on it?
What is Machine Learning?
In a nutshell, ML is field of Computer science that uses statistical (or mathematical) techniques to construct a model (or system) from observed data rather than have user enter specific set of instructions that define the model for that data.
Some common examples where ML is used in businesses is trying to find out if your customer is loyal or not, whether a customer will make a purchase or not, predict the sentiment of a document, etc. A more complex example could be predicting fluctuating stock market prices.
Most often, these algorithms work on precise set of features extracted from your raw data. Features could range from something as basic as age of a customer, number purchases a customer made, etc. to something as complex as pixel values for images, bag of words representation for text data, etc. So better the “Features” that represent the data, the more accurate your algorithm will be. The feature extractors are used to extract correct data features for a given sample, and pass this information to a classifier/ predictor. But machine learning algorithms are considered very “shallow”. Now why is that? This is for the simple reason that for ML algorithms, we need to provide lots and lots of training examples and the most accurate feature set that will train your algorithm well, thus manually correcting its mistakes. So, to avoid this human intervention, we have a new generation of ML algorithms called “Deep Learning”, being developed.
The need for Deep Learning
“Deep” learning algorithms attempt to model high-level abstractions (features) in data by using model architectures composed of multiple non-linear transformations.
Unlike machine learning, deep learning is mostly unsupervised, i.e. for example, creating large-scale neural nets (neural networks technique) that allow the computer to learn and “think” by itself without the need for direct human intervention.
It is a type of approach which involves building and training neural networks .You can think of a neural network as a black box decision making technique. They take an array of numbers (that represent words, image pixels, etc.), run a set of functions on that array, and give outputs. This is the basic principle of a neural network.
To understand more about neural networks take a look at the article Understanding Neural Networks by Jigsaw Faulty Neha Shitut
While applying this to real world problems, for example, face detection, neural networks takes in very complex functions, which means these arrays are very large, having around millions of numbers. Thus, the problem it solves is reducing task of making new feature extractor for each and every type of data (pixels of images, words of text documents, etc.). Also, the beauty of these algorithms is that they learn very well, use optimal set of “features” and are very fast.
To summarize, shallow machine learning algorithms involve a lot of duplication of effort to express things that a deep algorithm could more compactly. Hence, a deep algorithm can more gracefully reuse previous computations. However one major challenge is that they require very high computation power and GPU support since its pre-processing, feature learning and training phases take large time course
Business Impact of Deep Learning:
In just the last couple of years, deep learning software from giants like Google, Facebook, and China’s Baidu as well as a bundle of startups, has led to big breakthroughs in image detections, speech recognitions, stock trading, and much more.
An example of deep implementation is voice recognition like Google Now and Apple’s Siri. Most of this was developed from the works of Dahl, whose 2012 paper is “Context-dependent pre-trained deep neural networks for large-vocabulary speech recognition”. Companies like Facebook has also launched its own AI groups to find meaning in its feeds and posts.
And not just for big businesses. Lots of startups are also doing research in this field. Google recently acquired “DeepMind Technologies”, a startup based in London that had one of the biggest concentrations of researchers anywhere working on deep learning. Elliot Turner, founder and CEO of AlchemyAPI, said his company’s mission is to “democratize deep learning.” AlchemyAPI is a deep-learning platform in the cloud. This company is working in many domains from advertising to business intelligence, helping its customers to apply it to their businesses.
Apparently the worth of a dozen deep learning researchers is more than $400 million!
So deep learning is indeed revolutionizing businesses whether it involves providing better user interfaces, building better apps, generating money from placing advertisements or making sense of text mining from posts and news feeds. If Deep learning can make such an impact from speech and object recognition, then this could be a very important development in terms of value.