Search

Ten years ago we started a course on video game design and development. It was the first course on video games in our university and possibly in our country. We were immediately daunted by two main decisions: (i) the selection of the projects to be developed during the course and (ii) the evaluation of students’ projects. We wanted to give students the maximum freedom and no limit to their creativity. We wanted them to focus on the creation of a game that people would love to play without worrying about some score objectives to maximize and without caring about their instructors’ game design preferences. Accordingly, we decided to ask students to do all the job, starting from the submission and the selection of the game concepts to develop during the course, up to the final evaluation of the projects, the evaluation of their teammates, and thus basically the grading. In this paper, we discuss our experience over the last ten years with our course organization and grading model that, we believe, gives students complete freedom to express themselves and leaves them most of, if not all, the agency.

The RecSys challenge 2014 focuses on the engagement generated by the tweets posted by the users of the IMDb application for smartphones. Such engagement depends on attributes concerning: the user who posts the message (e.g., his role in the social network), the tweet content (e.g., the rating), and the movie object of the tweet (e.g., the popularity of the movie). In this work we provide an analysis of the dataset and of the task to help participants better understand the challenge. Furthermore, we propose a baseline prediction algorithm. We split our analysis into three stages: (i) data enrichment, (ii) knowledge extraction, and (iii) engagement prediction. Initially, we enriched the dataset with additional movie attributes extracted from IMDb and Free-base. Subsequently, we analyzed the statistics of the main tweet attributes and we applied some machine learning techniques to extract additional knowledge from the data. Finally, we defined a predictor on the basis of the main outcomes of these analyses. We define a linear regression model on attributes such as: user rating score, the presence of mentions, and whether the tweet is a retweet or it has already been retweeted. Such predictor led to an nDCG@10 equals to 0.8352.

Computational Intelligence and Games is a rather young but fast growing application domain where evolutionary computation is been widely applied. This tutorial aims at giving a brief overview of this area, highlighting the several opportunities for genetic and evolutionary computation methods, and providing some in-depth discussion of specific class of problems where evolutionary computation has proved very successful. The proposed tutorial is related to the Digital Entertainment Technologies and Arts (DETA) track, an attempts to bring together people and concepts from related areas, e.g. music and games, but also visual arts. Connecting these different domains is of growing importance for game production and CI techniques can play an important role here.

Since Learning Classifier Systems (LCSs) were introduced by Holland (1976, 1978) with the aim of creating cognitive systems which used evolutionary computation to learn a given task by interacting with its environment, the LCS paradigm has broadened greatly into a framework encompassing many representations, rule discovery mechanisms, and credit assignment schemes. LCSs are a very active area of research, with interesting, newer approaches that have shown not only to be competitive with respect to state-of-the-art machine learning techniques, but also to be very flexible approaches capable of solving a wide variety of real-world problems that range from data mining to automated innovation and online control.The International Workshop on Learning Classifier Systems was initiated in 1992 at the NASA Johnson Space Center in Houston, Texas. Then, since 1999 the workshop has been held yearly in conjunction with PPSN in 2000 and 2002 and with GECCO in 1999, 2001 and from 2003 to 2010.The workshop aims at bringing together researchers in Learning Classifier Systems and at providing a forum for researchers to discuss recent developments and future challenges of LCSs. We wish to thank all the members of the program committee and of the advisory committee for their help and support. We wish you will enjoy the 15th edition of the International Workshop on Learning Classifier Systems.

Computing the system prediction is one of the most important and computationally expensive tasks in Learning Classifier Systems. In this paper, we provide a parallel solution to the problem of computing the prediction array in XCS using the NVIDIA's Compute Unified Device Architecture (CUDA). We performed several experiments to test our parallel solution using two different types of GPUs and to study how performances are affected by (i) the problem size, (ii) the number of problem actions, and (iii) the number of classifiers in the population. Our experimental results show a speedup that ranges from slightly less than 2X up to 32X.

We present a framework for the procedural generation of tracks for a high-end car racing game (TORCS) using interactive evolution. The framework maintains multiple populations and allow users to work both on their own population (in single-user mode) or to collaborate with other users on a shared population. Our architecture comprises a web frontend and an evolutionary backend. The former manages the interaction with users (e.g., logs registered and anonymous users, collects evaluations, provides access to all the evolved populations) and maintains the database server that stores all the present/past populations. The latter runs all the tasks related to evolution (selection, recombination and mutation) and all the tasks related to the target racing game (e.g., the track generation). We performed two sets of experiments involving five human subjects to evolve racing tracks alone (in a single-user mode) or cooperatively. Our preliminary results on five human subjects show that, in all the experiments, there is an increase of users' satisfaction as the evolution proceeds. Users stated that they perceived improvements in the quality of the individuals between subsequent populations and that, at the end, the process produced interesting tracks.

Modern computer games are at the same time an attractive application domain and an interesting testbed for the evolutionary computation techniques. In this paper we apply NeuroEvolution of Augmenting Topologies (NEAT), a well known neuroevolution approach, to evolve competitive non-player characters for a racing game. In particular, we focused on The Open Car Racing Simulator (TORCS), an open source car racing simulator, already used as a platform for several scientific competitions dedicated to games. We suggest that a competitive controller should have two basic skills: it should be able to drive fast and reliably on a wide range of tracks and it should be able to effectively overtake the opponents avoiding the collisions. In this paper we apply NEAT to evolve separately these skills and then we combined them together in a single controller. Our results show that the resulting controller outperforms the best available controllers on a challenging racing task. In addition, the experimental analysis also confirms that both the skills are necessary to develop a competitive controller.

XCS with computed prediction, namely XCSF, has been recently extended in several ways. In particular, a novel prediction update algorithm based on recursive least squares and the extension to polynomial prediction led to significant improvements of XCSF. However, these extensions have been studied so far only on single step problems and it is currently not clear if these findings might be extended also to multistep problems. In this paper we investigate this issue by analyzing the performance of XCSF with recursive least squares and with quadratic prediction on continuous multistep problems. Our results show that both these extensions improve the convergence speed of XCSF toward an optimal performance. As showed by the analysis reported in this paper, these improvements are due to the capabilities of recursive least squares and of polynomial prediction to provide a more accurate approximation of the problem value function after the first few learning problems.

We investigate rule matching in learning classifier systems for problems involving binary and real inputs. We consider three rule encodings: the widely used character-based encoding, a specificity-based encoding, and a binary encoding used in Alecsys. We compare the performance of the three algorithms both on matching alone and on typical test problems. The results on matching alone show that the population generality influences the performance of the matching algorithms based on string representations in different ways. Character-based encoding becomes slower and slower as generality increases, specificity-based encoding becomes faster and faster as generality increases. The results on typical test problems show that the specificity-based representation can halve the time required for matching but also that binary encoding is about ten times faster on the most difficult problems. Moreover, we extend specificity-based encoding to real-inputs and propose an algorithm that can halve the time require for matching real inputs using an interval-based representation.

XCSF is an extension of XCS in which classifier prediction is computed as a linear combination of classifier inputs and a weight vector associated to each classifier. XCSF can adjust the weight vector of classifiers to evolve accurate piecewise linear approximations of functions. The Widrow-Hoff rule, used to update the weight vectors, prevents (when some conditions hold) XCSF from exploiting the expected piece-wise linear approximation. In this paper we replace the Widrow-Hoff rule with linear least-squares and we show that with this improvement XCSF can fully exploit its generalization capabilities.

XCSF is the extension of XCS in which classifier prediction is computed as a linear combination of classifier inputs and a weight vector associated to each classifier. XCSF can exploit classifiers' computable prediction to evolve accurate piecewise linear approximations of functions. In this paper, we take XCSF one step further and show how XCSF can be easily extended to allow polynomial approximations. We test the extended version of XCSF on various approximation problems and show that quadratic/cubic approximations can be used to significantly improve XCSF's generalization capabilities.