The application of artificial intelligence is a constant in the world of technology, but there is still a lot of ground to explore. For example, Facebook wants to train an AI to be able to “forget” irrelevant data. Researchers from the social network presented a very interesting project that is based on establishing expiration dates for all the information.
Not All Memories are Created Equal: Learning to Forget by Expiring, is the name of the work published in the last hours. Basically, what Facebook proposes is to incorporate tools so that artificial intelligence can identify which data deserves more attention. A) Yes, everything that becomes irrelevant can expire and does not unnecessarily consume analysis time or resources.
Facebook’s project in relation to AI is very interesting, especially in times where the volume of information available is increasing. For this reason the researchers propose a method called Expire-Span. It has been designed so that artificial neural networks sort and store more efficiently information related to the tasks to be performed.
Facebook trains an AI to forget irrelevant data
Photo by Markus Winkler on Unsplash
According to Facebook, Expire-Span predicts what information will be most useful to the network while developing a task and assigns an expiration date. If the system understands that certain information is very important for the work being carried out, the expiration date will be longer. On the contrary, if the data produces a smaller impact, the expiration will be closer.
The ability to forget irrelevant elements will not only allow artificial intelligence to focus on what really matters. Also will help free up memory space, something very relevant when the technical resources are not infinite.
Facebook research also mentions that every time new information is added, Expire-Span does not only analyze the relevance of this. The system reviews the data that previously existed to determine whether to change the expiration dates already set. “Calculate a prediction based on the context learned from the data and influenced by the memories that surround it,” the authors of the article explained.
The specialists of the social network pursue an ambitious goal, applying tools that are committed to being revolutionary: they want Artificial Intelligence systems to are every day “more human-like”.