DEEP LEARNING IS READY FOR PRIMETIME
Couple of months back Google owned Deep Mind designed a machine called AlphaGo able to beat the world`s best Go player. Machines beating humans in various games such as backgammon or chess is nothing new, this particular success however was something entirely different. In comparison to chess the game of Go has exponentially more possible moves. To illustrate this, in example after the first two moves in Chess there are 400 possible moves, in Go the number is close to 130,000. Following these mechanics the number of possible scenarios to map in a classic AI brute-force solution (like the one applied by IBM`s Deep Blue in 1996) becomes impossible to process. In fact Google wrote in a January post: "The search space in Go is vast... a number greater than there are atoms in the universe". In comparison to such brute-force solution AlphaGo applies a technique called deep learning. It replaces the classic exhaustive search tree mapping with neural networks, setting down the rules of the game. After that the algorithm analyzes vast amounts of data and tries to derive sequences of moves likely to lead to positive outcome. Essentially the machine learns to recognize abstract patterns, a feat that can be considered as one of the first instances of true A.I.
The investment opportunities in deep learning are vast. Virtually every industry can benefit from the efficient solving of a general pattern recognition problem. A task at which deep learning algorithms excel given enough data. And data is become available at ever increasing rates. The IoT, cheap sensors, social networking and cheap cloud computing all lead to the generation of huge amounts of structured data effectively turning almost every company into a big data company. Opportunities for disruption are everywhere ranging from financial services to healthcare. Healthcare in particular is of great interest considering the large size of the industry and relative recent developments like the declining cost of DNA sequencing. Following the trend it is no surprize that every big software service provider is investing heavily in deep learning. A.I. IBM`s Watson has now split into several separate APIs, SAP has it`s own deep learning projects in the pipeline and Google has already outsourced it`s latest machine learning system.
Besides the big players there are several startups already offering effective solutions:
Ayasdi is a Valley company with effective solutions in financial services and healthcare. Here is a nice presentation by Ayasdi`s Gurjeet Singh from 2014:
Israeli based SparkBeyond is offering an automated pattern recognition software, which according to the company's website is already applied by several Fortune 500 companies.
Another big area for deep learning algorithms is automated video recognition. Ranging from automated advertisement to content matching to automated surveillance the opportunities are huge. Clarifai offers automated recognition of objects in any video allowing i.e. the matching of advertisement to the content being displayed at a specific moment.
Enlitic specializes in using deep learning algorithms to spot abnormalities in medical images. I.E it can highlight abnormal areas in a CT scan and help doctors give a diagnosis.