The proverb “it is forging that one becomes a blacksmith” now applies to machines, because many artificial intelligence devices today depend on repetition to learn. Deep learning algorithms are designed to enable artificial intelligence devices to gather knowledge from datasets and then apply what they have learned to real-world situations. For example, an artificial intelligence system receives data on how the sky is usually blue, which allows it to later recognize the sky in a series of images.
So, for a deep learning algorithm to recognize a cat with the accuracy level of a Siberian cat, it must be “fed” with hundreds of thousands of feline images, with variations in size, shape, texture, lighting and orientation. This complex work can be accomplished using this method, but could not the same results be obtained by providing less data? It would be much more effective if, a bit like a person, an algorithm could develop an idea of what makes a basic cat a Siberian cat.
A Boston-based startup, Gamalon, has developed a new technology to try to respond to this and has released two products that use its new approach. This type of technology allows a machine to learn effectively from fewer games and refine knowledge.
Gamalon uses a technique called “the Bayesian synthesis program”to create algorithms that can learn from less data. Re dictions about the world based on experience. Gamalon’s system uses probabilistic programming – or code that deals with probabilities rather than specific variables – to build a predictive model that explains a particular set of data. From a few examples, a probabilistic program can determine, for example, that it is very likely that cats have ears, whiskers and tails. As more examples are provided, the code can be rewritten to change the probabilities. This provides an effective way to learn important knowledge of the data.
If the underlying technique can be applied to many other tasks, this could have a significant impact. The ability to learn from less data could allow robots to explore and understand new environments very quickly or allow computers to know the user’s preferences without sharing their data.