The Internet of Things (IoT) has brought about the opportunity to establish a new architecture for Smart Buildings that decentralizes systems and pushes analytics processing to the “edge”, or sensor unit, rather than the cloud or a central server. This ‘Edge’ or ‘fog’ computing provides real-time intelligence and enhances system agility while simultaneously offloading the heavy communications traffic.

The hardware is ready for this change with the development of cheap and energy-efficient embedded processors. On the software side this distributed system lends itself to a deep learning approach to data analysis, intelligently selecting relevant data to be transferred and processed. What was for a long time a theoretical future is now being put into practice in our smart buildings, cities and the world of IoT.

Traditionally, rule based systems have always been considered easier to analyze, but as patches of rules begin to be stacked on top of one another its simplicity dissipates. Furthermore, rule creation and modification are inherently human tasks and therefore subject to error or less-than-ideal structure, especially in response to changes.

More recently, through machine learning, we have transferred the task of creating rules to a pre-designed algorithm. By using machine learning, engineers must only define the features of the relevant raw data and provide enough data sets for the algorithm to “learn”. Machine learning can then extract relevant data from the data set, before applying learned rules to achieve a certain outcome.

From machine learning has emerged “deep learning” an almost futuristic system in which engineers do not even need to define features. Deep learning begins with labeling samples; from that humble starting point the algorithm itself can then determine the ideal process to turn raw data into desired output.

The level of computation achieved through deep learning is considered significantly more effective than that of machine learning. Using a neural network, a computer system modeled on the human brain and nervous system, deep learning employs millions of parameters, which it uses to refine the system until the ideal process is identified. With deep learning, the engineer does not even need a significant knowledge of the application, fundamentally changing system design and implementation.

In fact, the deep learning system’s engineer need only define the neural network’s core architecture and ensure the network is large enough to identify best process. Hardware developments have allowed increasing processing power throughout the system and a neural network would be able to adapt to further increases, in order to re-optimize the system, with minimal human involvement.

This self-learning element is what makes the step to deep learning different from previous advances. In essence, deep learning can be described as the first true practical form of artificial intelligence, a system that can adapt to changing circumstances and environments without human intervention. Removing human involvement is the basis of the advancement, accepting that machines are better at identifying the most relevant data and processing, given the right objectives.

All of this means that machines, using deep learning, can be used to solve ‘intuitive’ problems – problems characterized by high dimensionality and no clear rules. Shifting from a traditional top-down system where all enough rules are provided for all possible circumstances, to a bottom-up system where the machine learns from experiences.

Ajit Jaokar, founder of the London based tech research company futuretext describes the process as similar to the way in which a child learns what a dog is, for example. By understanding the sub-components of a concept, such as the behavior (barking), shape of the head, the tail, the fur and so on, then putting these concepts in one bigger idea i.e. the dog itself.”

“Knowledge representation incorporates theories from psychology which look to understand how humans solve problems and represent knowledge. The idea is that: if like humans, Computers were to gather knowledge from experience, it avoids the need for human operators to formally specify all of the knowledge that the computer needs to solve a problem,” explains Jaokar.

Deep learning is most appropriate for applications that entail considerable amounts of data and intricate relationships between different parameters. For a Neural network to learn it needs to be shown that “given an input, this is the correct output” over and over again. With enough repetitions, an adequately trained network will reproduce the desired results while ignoring inputs that are irrelevant to the solution.

“I think AI is akin to building a rocket ship. You need a huge engine and a lot of fuel. If you have a large engine and a tiny amount of fuel, you won’t make it to orbit. If you have a tiny engine and a ton of fuel, you can’t even lift off. To build a rocket you need a huge engine and a lot of fuel. The analogy to deep learning [one of the key processes in creating artificial intelligence] is that the rocket engine is the deep learning models and the fuel is the huge amounts of data we can feed to these algorithms,”suggests Andrew Ng, chief scientist at China’s search giant Baidu.

In the real world we are seeing sensor rich, smart buildings and smart cities, which have “considerable amounts of data and intricate relationships between different parameters.” As large data sets begin to take shape, these IoT entities will present ideal proving grounds for this emerging technology. In this sense, what happens in the smart building and smart city sector will have huge implications on the world of tomorrow.

[contact-form-7 id="3204" title="memoori-newsletter"]