Over the past several years, the global Data Science community has watched the rise and steady penetration of such concepts as neural networks, Deep Learning, and back propagation. As research in Artificial Intelligence continues to evolve, an increasing interest in Neural Networks and Deep Learning has caught the fancy of worldwide business owners and operators. So what is Deep Learning and how can it benefit the global business community?
An Introduction to Deep Learning provides a general view of the science of Deep Learning, but aptly describes how an algorithm is designed and how it learns through layers. This primer explains the Deep Learning technology through the analogy of a “thinking computer.”
Deep Learning can perfectly train a computer to solve intuitive problems in the following manner:
- Computers are systematically navigated through one layer of discovery to another, thus going through a hierarchical progression of knowledge from previous knowledge.
- The hierarchy of concepts is carefully built from bottom navigating upwards, without predefined rules.
This may sound somewhat like the “elephant and the six blind men” approach, but when the learner is a single entity, the different parts or components can be easily combined together to form the whole concept. In the case of an image, Deep Learning breaks the image down into a series of hierarchical layers, with each layer unfolding a part of the whole concept.
Although a lot of hype was generated around the above term, KD Nugget’s Deep Learning Flaws had the courage to come forward with the algorithmic flaws quite frequently discovered in Deep Learning and also in Machine Learning.
One claim has been made by Google’s Christian Szegedy, who revealed that images can be altered in a manner so that human brains cannot decipher the changes, but can induce misclassification by a Convolutional Neural Network (CNN). This startling discovery raises questions about the authenticity of applying Deep Learning techniques in image-recognition software. Another challenge put forth by Anh Nguyen of the University of Wyoming is the complete fabrication (doctored) of an authentically classified image.
Mathew Mayo. The author of Seven Steps to Deep Learning states that Deep Learning, while a distinct branch of Machine Learning, utilizes neural network architectures to propose solutions for fields like Natural Language Processing or Bioinformatics. Deep Learning, through intense research applications and proven results, has established its diverse beneficial effects on business situations. Deep Learning has suddenly begun to make its presence felt across all areas of Machine Learning and Data Science. The entire community of researchers, practitioners, and industry watchers are fascinated by this impactful research area of Artificial Intelligence.
So What is Deep Learning in Layman’s Terms?
In the world of programming, there may be problems for which writing a program is not easy. For example, writing a program to decipher handwriting. In such cases, the Data Scientist can take over from just a programmer, and try to develop an algorithm that the computer can use to study many examples and then use the acquired “learning” to solve the problem. In other words, the algorithm is teaching the computer to solve by example. Microsoft Research has really revealed the hidden promises of Deep Learning.
In the case of Deep Leaning, this acquired learning or experience comes from “hierarchical layers of discovery” in each layer. The computer learns from each layer, and then uses that learning in the next layer to learn more, till the learning reaches its full stage through cumulative learning in multiple layers. This was the only method possible to circumvent the complexity of feature extraction present in traditional Machine Learning. Thus, in Deep Learning the computer is not limited to “fed” or “supervised” logic; rather, the computer is trained to learn from each stage or layer of learning to use in the next.
Deep Learning technology mimics the human brain; hence this model is also known as a neural network, consisting of neurons. Similar to the structure of the human brain, neurons in neural networks are also organized in layers. Data Robots A Primer on Deep Learning can aid beginners gaining a deeper understanding of feature hierarchies and neural networks, which form the core of Deep Learning.
So Why Has Deep Learning Suddenly Become Important?
The article Why People Are Key to Data Science discusses the importance of human intervention in traditional Machine Learning projects.
In sharp contrast to the above, think of the competition held by Kaggle in 2013 to engineer an algorithm to classify sound clips to detect the location of whales on North Atlantic Sea. This competition was created with the sole purpose of preserving both the whales and the shipping traffic on the North Atlantic Sea. The takeaway from this competition was that Deep Learning promised highly successful solutions for problems that typically required a multi-layered structure of learning to produce meaning. This assumption further rests on Deep Learning’s ability to learn from data on its own without any human intervention. The whole objective of Deep Learning is to solve problems with no set rules.
The Founding Fathers of Deep Learning
The three Fathers of Deep Learning: Geoff Hinton of Google, Yann LeCun of AI Research at Facebook, and Yoshua Bengio of University of Montreal conducted frontier research in the field of Deep Learning and paved the way for a breakthrough discovery in 2006. These three pioneers were successful in training deep neural networks to learn from useful representations.
The underlying principle of this discovery is that “deep neural networks” can learn from representations of data in an “unsupervised” manner. This feature can prove very powerful in domains where a collection of data points must be analyzed together to convey meaning or information.
In The Role of Deep Learning in Data Science, the author points out that this fascinating research area of Machine Learning mimics the human brain. Data Scientists are essentially feeding the “machine” with data and smart algorithms, and helping the machine to act like a thinking brain to aid speech recognition, Natural Language Processing, or image detection tasks. The biggest difference between other Machine Learning techniques and Deep Learning is that while other Machine Learning techniques dealt with supervised data, Deep Learning techniques have the ability to function “unsupervised.” In other words, it can think and learn on its own based on the algorithms provided by the Data Scientist. Deep Learning gains knowledge from the many layers of experience in complex structures like a human face, just like the human brains do.
Current Market Applications of Deep Learning
Some of the well-known companies already utilizing Deep Learning technologies include
Apple, Facebook, Google, IBM, Microsoft, PayPal, Pinterest, Twitter, Yahoo, and others. This McKinsey tech article demonstrates that as organizations like Google, Facebook, Microsoft, or Baidu are investing in such technology, we can see a sudden acceleration in the industry applications of Deep Learning.
The three significant market applications are:
- Natural Language Processing Software: This tool helps the computer decipher messages or text.
- Image Recognition Software: This tool enables the computer to search, sort, and segment for object detection.
- Speech Recognition Software: This tool allows humans to interact with their smart gadgets!
Two products that have gained prominence in the recent times are Enlitic, a healthcare company that uses Deep Learning technology to diagnose diseases, and Cellscope, a healthcare device company, which allows parents to monitor the health condition of their children through a smart gadget instead of repeatedly visiting the doctor.
The top universities and research groups engaged in Deep Learning research are Carnegie Mellon, NYU, Stanford, Swiss Al Lab IDSIA, UC Berkeley, University of Michigan, University of Montreal (LISA Lab), University of Oxford, and University of Toronto among others.
Gartner Selects Deep Learning as a Top Tech Trend
In a recent Gartner article, they chose Deep Learning as one of the biggest strategic trends of 2016. There is a clear indication that Deep Learning is set to be one of the top trends moving through 2016 and beyond.
In the future, industry watchers are bound to see the impact of Deep Learning in other technologies such as the Internet of Things or Advanced System Architecture. More contributions from Machine Learning and Deep Learning to the larger business world of smart and connected products and services are certain. Deep Learning in no longer just a trend, it’s fast becoming an important technology used across multiple industries.