Martin Spano is the author of Artificial Intelligence in a Nutshell, a book that explores the mystified subject of artificial intelligence (AI) with simple, non-technical language. Spano’s passion for AI began after he watched 2001: A Space Odyssey, but he insists this ever-changing technology is not just the subject for sci-fi novels and movies; artificial intelligence is present in our everyday lives.
The story of Alex KrizhevskyRead also:Read more
Alex Krizhevsky was born in Ukraine but lived most of his life in Canada. After finishing his undergraduate studies, he continued as a postgraduate under the supervision of Geoffrey Hinton, legendary computer scientist and cognitive psychologist, one of the foremost advocates of using artificial neural networks for artificial intelligence. Krizhevsky stumbled upon an algorithm by Hinton that used graphics cards instead of processors for its execution. He applied this tweak to artificial neural networks with more layers, so-called deep neural networks. Soon after that, Sutskever, another student under the supervision of Hinton, learned about the Krizhevsky algorithm and proposed to use it for the ImageNet competition. In this contest, teams try to achieve the highest possible accuracy in image classification, or what a computer sees in a picture. It is very easy for humans but immensely tricky for computers. With Hinton as an advisor, Krizhevsky and Sutskever entered the competition in 2012 with this exotic idea of using deep neural network designed by Krizhevsky and devastated all competitors. These events gave birth to a new, unprecedented boom in artificial intelligence.
Child-like thinkingRead also:Read more
Like any computer program, artificial intelligence is based on the use of algorithms. An algorithm is a set of clear instructions executed on a computer leading to the desired result. It is like a precise recipe with which you can prepare a computer program. The pioneers of artificial intelligence thought that the most suitable approach to creating artificial intelligence was to program it similarly as we do in the case of digital computers, to instruct it step-by-step on what it should do using if-then-else statements. For simple tasks this approach is feasible, but in real life, we quickly run into the problem of the exponential explosion as there are many possibilities in each routine decision. What should we do? Let's look at how the only object in the universe we know of possessing human-level intelligence - the human brain - works. Consider how a child obtains knowledge. You do not instruct it how to behave in each situation. It would also be unfeasible when the child orders its own instructions, as it rarely does. You tell them some basic guidelines and let them learn on their own, correcting them if they misbehave. This approach is the principle of machine learning.
Similarly as a child in the example above needs a lot of real-world input to learn successfully, a machine that learns needs a large amount of training data. If this data is labeled, the learning is called supervised, unsupervised otherwise. Consider an example of a machine learning algorithm that recognises a dog on a picture. Training data consists of a vast database of images. For supervised learning to work, the data needs to be labeled; we need to tell which images from the training set are dogs and which are not.
Deep learning, large dataRead also:Read more
In contrast, if an algorithm can find patterns in input without prior data labeling, it is called unsupervised learning. Similar to children, we can give a machine a reward for good responses or punish it for doing something wrong. This is an example of reinforcement learning.
The most advanced subset of machine learning is using artificial neural networks. In this approach, we simulate how the brain works by imitating its structure. We create a software representation of neurons and connections between them. The artificial neural network is divided into multiple layers, with input passing through the input layer then through hidden inner layers, each layer extracting more abstract information, until the output is passed out through an output layer. If there are multiple inner layers present we called the network the deep neural network and the approach deep learning.
One of the critical contributors to the advancements of recent years was the availability of big data. There is a massive amount of data generated by social networks and data outputs from an ever more increasing number of gadgets connected to the internet. Deep learning needs large data sets to train artificial intelligence. These trained networks are then used to classify or find patterns in other large datasets.
The Slovak Spectator will publish more extracts from Spano’s book in the coming weeks.
16. Aug 2019 at 10:30 | Martin Spano