Skip to main content

OpenAI Universe - New Way of Training AIs

Until now, the way a neural network worked, was to supply it with millions of pre-classified data, in the so called supervised learning scheme, which resulted in neural networks only learning what we've instructed them to do.

But there's also another technique, that of reinforcement learning where you let the AI  discover by itself what it's supposed to do, without prior knowledge of its surroundings or any other data fed to it.
Microsoft was one of the first to employ this technique in a gaming environment, in trying to make a Minecraft character climb a virtual hill in the so called AIX Minecraft Project. In there, you let the algorithm explore the Minecraft world it was dropped in, let it freely move and interact with its surroundings, and force it to learn by rewarding it when it does something right so that it understands the goal of the game, the goal that it should be aiming for. Of course for us humans it's easy to see that we must climb that hill, or that when Super Mario touches a troll he instantly dies; but not so for an algorithm. Its strength instead, lies in the fact that it can try a billion combinations in the span of a microsecond in order to discover the same thing that humans intuitively had already in possession.

full article on i-programmer

Comments

Popular posts from this blog

Book Review : How To Create Pragmatic, Lightweight Languages

At last, a guide that makes creating a language with its associated baggage of lexers, parsers and compilers, accessible to mere mortals, rather to a group of a few hardcore eclectics as it stood until now.

The first thing that catches the eye, is the subtitle:

The unix philosophy applied to language design, for GPLs and DSLs"
What is meant by "unix philosophy" ?. It's taking simple, high quality components and combining them together in smart ways to obtain a complex result; the exact approach the book adopts.
I'm getting ahead here, but a first sample of this philosophy becomes apparent at the beginnings of Chapter 5 where the Parser treats and calls the Lexer like  unix's pipes as in lexer|parser. Until the end of the book, this pipeline is going to become larger, like a chain, due to the amount of components that end up interacting together.

The book opens by putting things into perspective in Chapter 1: Motivation: why do you want to build lan…

Deep Angel-The AI of Future Media Manipulation

Undeniably, we live in the era of media manipulation. Such powerful and accessible tools exist today that nearly everyone can do it. Now add to this collection Deep Angel, an artificial intelligence that can erase objects from photographs and videos.

I was notified of Deep Angel around the time I was watching Kill Switch, a futuristic and dystopic movie about our Earth getting cloned in order to suck the resources of the cloned planet, something that would sustain our world's energy needs for at least another millennium. To cut a long story short... full article on i-programmer.info

SAP's Creating Trustworthy and Ethical Artificial Intelligence

With the ink hardly dry on the pages of the EU Ethical AI Guidelines manifest, a free online course exploring the issues they raise is already in prospect on the openSAP platform. Run by members of the very same group, the European Union’s High-Level Expert Group on Artificial Intelligence, who wrote the guidelines and in cooperation with SAP's online education platform, a course with the titleCreating Trustworthy and Ethical Artificial Intelligence has been made accessible to anyone with an interest on AI or ML:
full article on i-programmer.info