AI News, Why Neural Networks Look Set to Thrash the Best Human Go Players for the First Time
- On 4. oktober 2018
- By Read More
Why Neural Networks Look Set to Thrash the Best Human Go Players for the First Time
The best algorithms only achieve the skill level of a very strong amateur player which the best human players easily outperform.
These guys have applied the same machine learning techniques that have transformed face recognition algorithms to the problem of finding the next move in a game of Go.
The players alternately place black and white stones on the grid in an attempt to end up occupying more of the board than their opponent when the game finishes.
Go players have 19 x 19 = 361 possible starting moves and there are usually hundreds of possible moves at any point in the game.
Many experts believe that the secret to human’s Go-playing mastery is pattern recognition—the ability to spot strengths and weaknesses based on the shape that the stones make rather than by looking several moves ahead.
These advances have used massive databases of images to train deep convolutional neural networks to recognize objects and faces with the kind of accuracy that now matches human performance.
The question that these guys have trained a deep convolutional neural network to answer is: given a snapshot of a game between two Go experts, is it possible to predict the next move in the game?
They used almost 15 million of these position-move pairs to train an eight-layer convolutional neural network to recognize which move these expert players made next.
Clark and Storkey say that the trained network was able to predict the next move up to 44 percent of the time “surpassing previous state of the art on this task by significant margins.” That’s Interesting not least because the new approach does not use any of the previous moves to make its decision;
“Even though the networks are playing using a ‘zero step look ahead’ policy, and using a fraction of the computation time as their opponents, they are still able to play better then GNU Go and take some games away from Fuego,” they say.
One idea that Clark and Starkey suggest is to run the convolutional neural network in parallel with the conventional approach to help prune the tree of possible moves that need to be explored.
- On 30. september 2020
Evolution - 7 layer neural net Boids evolved with genetic algorithm 79 Hours
Using the original code of John Hess's Evolution simulator, with more "creatures" and bigger brains. This version used a number of starts until some movement ...
Deep Convolutional Neural Network for Segmentation and Classification of Off-Road Scenes
From On-Road to Off: Transfer Learning within a Deep Convolutional Neural Network for Segmentation and Classification of Off-Road Scenes (C.J. Holder, T.P. ...
Software Defined Networking - Computerphile
Audible 30 day free Trial: Software Defined Networking takes the control away from basic protocols and gives it to the ..
The Cognitive and Computational Neuroscience of Categorization, Novelty-Detec...
Google Tech Talks November, 15 2007 ABSTRACT Neurocomputational models provide fundamental insights towards understanding the human brain circuits ...
Metal Gear Solid 4 - The Movie [HD] Full Story
Movie version of the Metal Gear Solid 4: Guns of the Patriots. Haven't played MGS4 before or just want to see the epic game again? Then this video might be for ...
Essence of calculus, chapter 1
I want you to feel that you could have invented calculus for yourself, and in this first video of the series, we see how unraveling the nuances of a simple geometry ...
CARTA: Imagination and Human Origins - Sheldon Brown, Agustín Fuentes, Caren Walker
Visit: 1:39 - Introduction - Alysson Muotri 3:10 - What is Imagination? - Sheldon Brown 22:10 - Dream It, Be It: How Imagination and Creativity ..