The artificial intelligence topology that does not appear in the machine learning literature to my knowledge is that of officiated teams or round robins of them. The paradigm is a proven one in the world of sports. If the rules of the game are well designed, the result is sensational, in multiple meanings of that word.

Is anyone working on convergence in this topological space?

Does anyone want to discuss it with those considering it in my lab as a creative commons initiative?

Case One

Two teams of players engage in game play officiated by a team of officials. Team members (each a network itself in the machine learning context) collaborate to achieve a goal. The two teams collaborate to create the show of ability in goal achievement. The officials (each a network) make rulings in boundary cases. This is a network-ish way of achieving what fuzzy logic attempts to achieve.

In sports, the abilities are athletic, but that is arbitrary. The abilities could be linguistic, social and/or intellectual as in debate teams, hack-a-thons, or the competition between Google and FaceBook.

Case Two

Round robin or elimination tournaments exist in sports to create events of extended duration. Seasons are simply iterations. This is, in computer science, like a batch approach, but it could be reenterant and continuous as in ML reinforcement. In this way teams of neural networks could be used in real time learning and this may be what occurs in some of the structures of mammalian brains.

Humans may have projected its own inner workings onto playing fields, and that is why sports may be so popular. Enthusiasts are, in an unconscious sense, introspecting when they intensely following sports.

  • $\begingroup$ This seems neat. I don't have an answer, but some thoughts. This seems similar to co-evolutionary algorithms, especially your case 1. I think Jordan Pollack's work might be worth looking at in that vein. Case 2 also seems related (simple tournaments are used in Co-Evo algorithms for selection). $\endgroup$ – John Doucette Aug 8 '18 at 13:17
  • $\begingroup$ Team intelligence in a sports game can be realized with search in the game tree. Winning a game is equal to find the node with the highest reward. Before the solver can traverse through the nodes the game has to be modeled. Game modeling means, to describe the tournaments with abstract rules. One option in doing so is a formal grammar in the BNF notation. Another, more recent option, is to use the Game Description Language. In both cases, the allowed actions and game rules are expressed in a machine readable code. $\endgroup$ – Manuel Rodriguez Apr 17 '19 at 7:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.