Here we propose an artificial neural network approach to solve two player zero sum games. The paper is divided into three chapters: the first is an introduction about the topics of zero sum games and artificial neural networks; the second concerns the development of several nets, some convolutional while others dense, able to perform the desired task; and, finally, the third one contains an analysis of the results as well as considerations for further developments. Where does the idea come from? We know that a finite zero sum game is represented by a simple two dimensional finite matrix and always has a rational outcome in mixed strategies; we know that convolutional neural networks perform well in image recognition but an image is nothing more than a set of three matrices to a computer. So the idea is to use these networks to solve zero sum games.
Qui proponiamo un approccio basato sulle reti neurali per risolvere giochi a somma zero tra due giocatori. Il lavoro è diviso in tre capitoli: il primo consiste in un’introduzione agli argomenti dei giochi a somma zero e delle reti neurali; il secondo riguarda la progettazione e le caratteristiche di diverse reti, alcune convoluzionali mentre altre dense, in grado di risolvere i giochi menzionati; infine il terzo contiene l’analisi e un confronto dei risultati ottenuti e considerazioni su possibili sviluppi. Da dove arriva l’idea? Sappiamo che i giochi a somma zero finiti sono rappresentati da una semplice matrice bidimensionale e finita e hanno sempre un risultato razionale in strategie miste; sappiamo che le reti neurali convoluzionali funzionano bene nel campo del riconoscimento delle immagini ma queste, per i computer, non sono altro che un in- sieme di tre matrici. Dunque l’idea è di usare tali reti per risolvere i giochi in questione.
Solving zero sum games with artificial neural networks
RAPONI, DAVIDE
2023/2024
Abstract
Here we propose an artificial neural network approach to solve two player zero sum games. The paper is divided into three chapters: the first is an introduction about the topics of zero sum games and artificial neural networks; the second concerns the development of several nets, some convolutional while others dense, able to perform the desired task; and, finally, the third one contains an analysis of the results as well as considerations for further developments. Where does the idea come from? We know that a finite zero sum game is represented by a simple two dimensional finite matrix and always has a rational outcome in mixed strategies; we know that convolutional neural networks perform well in image recognition but an image is nothing more than a set of three matrices to a computer. So the idea is to use these networks to solve zero sum games.File | Dimensione | Formato | |
---|---|---|---|
2025_04_Raponi.pdf
accessibile in internet per tutti a partire dal 03/03/2026
Descrizione: Tesi completa
Dimensione
1.79 MB
Formato
Adobe PDF
|
1.79 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/235495