Deep Learning by John D. Kelleher. John is the Academic Leader of the Information, Communication and Entertainment research institute at the Technological University Dublin. He has previously taught at Dublin City University, Media Lab Europe, and DFKI (the German Centre for Artificial Intelligence Research).
This is a very good introduction to specific subsets of artificial intelligence that are indeed powering imaging, speech recognition, machine translation, and autonomous cars today.
Consumers may forget as they are engaging various technologies, their interactions are via Deep Learning systems. This includes interactions with Siri on iPhones, and Alexa on all things from Amazon. To a lesser extent is Cortana from Microsoft. Actually, John provides a wonderful glossary. This serves the reader well in helping to further develop their understanding of Deep Learning systems.
Likewise, his introduction illustrates how Deep learning delivers data-driven decisions from very large datasets. The key is Deep Learning deliver immediate ‘learning’ as the large datasets grow.
In addition, his insights on autoencoders, recurrent neural networks, and Generative Adversarial Networks (GAN) are very stimulating. At the same time, addressing gradient descent and especially backpropagation is amazing in of itself.
Above all, I must admit there is a bit of a deep dive into mathematical concepts. Addressing neurons, classic neural networking, auto-encoders, gradient-decent and back-propagation algorithms is no easy task.
Chapter Three: Neural Networks
John’s detailed overview of Neural Networks is certainly fascinating and well worth the cost of this book alone. This technology is really advancing from the 1950’s original foundation of Artificial Intelligence. Besides, understanding that by aligning the human brain to advanced networks servers the reader well for the reminder of the book.
Mapping neurons and learning how scientists also created nonlinear mapping is a key understanding to how Neural Networks can learn to diagnosis patient x-rays. It is perhaps telling that while John is certainly outlining how the power of Deep Learning can be applied to healthcare, he misses the real world application.
Neurons are the basic building blocks of neural networks, and therefore they are the basic building blocks of the mapping a network defines. The overall mapping from inputs to outputs that a network defines is com- posed of the mappings from inputs to outputs that each of the neurons within the network implement. The implication of this is that if all the neurons within a network were restricted to linear mappings (i.e., weighted sum calculations), the overall network would be restricted to a linear mapping from inputs to outputs. However, many of the relationships in the world that we might want to model are nonlinear, and if we attempt to model these relationships using a linear model, then the model will be very inaccurate.
pg. 77
Deep Learning systems can provide tens of thousands of x-ray scan outcomes to doctors in rural countrysides. The advanced diagnosis no longer requires patients to travel to large metropolitan medical centers. This not only saves time and money, but provides more efficient delivery of treatment. This is a great addition to Mark Coeckelbergh’s AI Ethics
Chapter Four: History of Deep Learning
Whereas most readers would are expecting the historical developments within chapters one or two, John jumps this into Chapter Four. In fact, his development and required math tell quite a detailed history of developments. For this purpose, John is addressing how projects can stumble early on, with corrections and technology advances that permitted advanced neural network and GANs to develop over time.
Welcome to your future
John finishes the book with a short overview regarding on the “future of AI” and it’s not disappointing. From scientific events (competitions) that disclose new advances in Deep Learning to new revisions to traditional CPU designs to increase efficiency and processing power. This is pushing me to read about GANs as well.
In conclusion, John is delivering a wonderful book on Deep Learning that should be standard curriculum across country.