The Future is Machine Learning
In recent years, machine learning and artificial intelligence have become prominent in industry and the minds of the public. The availability of more data than ever before, combined with the unwavering growth in capabilities of hardware systems, has allowed machine learning techniques to be scaled with unprecedented ambition.
The field of deep learning, in particular, has seen meteoric success. Deep neural networks have overturned the status quo in the domains of image recognition, voice synthesis, machine translation, among many others. Algorithms powered by machine learning have displaced humans as champions of the most intellectually demanding games on the planet. Even if machine learning research stopped today, the percolation of techniques available right now would continue to overturn industrial practices for decades to come.
Faces generated by a deep learning-based Generative Adversarial Network.
But progress is not stopping, it is accelerating. The successes of machine learning have spurred a tidal wave of attention, research and development. Every month brings more sophisticated and ambitious algorithms and architectures into the arena. Manufacturers of the GPUs that allowed deep learning to take off have turned to creating hardware specifically designed to speed up training and increase the scope deep nets by orders of magnitude. The momentum behind the artificial intelligence transformation is growing.
Though artificial intelligence has gone through similar spurts of interest in the past, only to be followed by “AI winters“, none of the previous phases had achieved nearly as much applied success as the current. It is safe to say that no industry intersecting with IT is likely to evade the reaching transformative effects of what has been called the “fourth industrial revolution“. Even outside of IT, the ongoing transformation will affect all of us. In a few decades life without self-driving cars, artificial conversation agents and intelligent personal assistants will seem as alien as life without the internet does now.
Understand the Future
It is my firm belief that anything as important and transformative as the current wave of machine learning should be understood by those who might be affected by it – ideally, all of us. But for IT professionals, knowledge of the lexicon, basic principles and most prominent sub-domains of machine learning is not just an ideal, it is becoming indispensable.
Style can be separated from content by adversarial autoencoders to generate fully custom images.
Those in a position to decide where and how to apply machine learning tools in their business must understand these tools. Knowing how to answer questions like, “how does a deep neural net work? what is a Bayesian method? how much data do I need for this task? can this be automated? with what effort?” can help not only to tackle outstanding practical problems, but also to find hitherto unseen opportunities to analyze, automate and profit from the AI transformation.
Inculcating machine learning intuition is important, and I have given a number of presentations aiming to give an introduction to machine learning that would allow someone unfamiliar with the field to begin to answer such questions. I have included the slides for one of these. These slides give an introduction to the basic domains of machine learning for the uninitiated reader.