I wrote an article on “My First and Best Lessons Learned in Writing Blogs”, where my third and last lesson was on Accidental Topic Scouting, in which I “accidentally” realize that I already wrote most of an article when I was writing content for another purpose — responding to an email from a colleague. The following three paragraphs are precisely an example of this. These paragraphs are my reply to an email inquiry that I received asking what’s so special about deep learning. I sent this reply to my colleague:
Deep learning represents a remarkable technological convergence. Specifically, deep learning lives and thrives at the convergence of new disruptive problem-solving approaches, scientific techniques, algorithmic methods, real world applications, advanced mathematics, computational tools, computing resources, and the best minds in the computer and data sciences. Some would say that neural networks are not new. That’s true. Some would say computer vision has been around long before convolutional neural networks rose to everyone’s attention. That’s also true. And some would say that machine learning and AI have already gone through multiple springs and winters over the past few decades, and we will probably see another long winter when the current hype cycle fades. That’s — maybe — a winning hat trick for the naysayers. Right? I say, “wrong!”
The ability to solve hard problems with computational algorithms and the capabilities to automate those solutions are now at a level in depth and breadth that we have never seen. Previously insurmountably hard problems can now be solved, such as those required for safe operations of autonomous vehicles, or real-time language translation in live conversations, or conversational chatbots that come close to passing the Turing Test, or the relatively easy generation of deep visual and textual “fakes” that both entertain and frighten us. What makes all that possible is the convergence in the maturity and accessibility of advanced mathematical algorithms, ubiquitous fast computing resources, universally adoptable coding languages, and oceans of data, data, data everywhere (that AI and deep learning devour)!
Deep learning brings these many tools, techniques, and talents together (that’s the convergence) in a myriad of diverse real-world applications (that’s a healthy complement of divergence). Deep neural networks do a remarkable job of succinctly auto-encoding the salient features in complex data (images, video, audio, documents, spoken language), which we may call dimensionality reduction or explanatory feature generation, then applying those implicit (latent) hyper-patterns to inform decisions and actions fueled from those complex data sources. Whether the challenge is in image understanding, language understanding, or context understanding, the new deep learning techniques and components enable exciting functionalities in data-rich environments: object detection and recognition, behavior detection and recognition, anomaly detection, content (image, video, audio) generation, and context (attention) determination.
Follow me on Twitter at @KirkDBorne
Learn more about my freelance consulting / training business: Data Leadership Group LLC
See what we are doing at AI startup DataPrime.ai