Throughout the course of this book, we have dived deep into a specific realm of AI, nested within ML, that we call deep learning. This caveat of machine intelligence takes a connectionist approach, combining the predictive power of distributed representations, in turn learned by a deep neural network.
While deep learning neural networks have risen to prominence, since the advent of GPU, accelerated computing, and the availability of big data, many considerations have gone into improving the intuition and implementation behind these architectures, since their re-ascension to popularity, about a decade ago (Hinton et al, 2008). Yet, still, there exist many complex tasks that deep learning is not yet able to adequately tackle.
