Publications

Artificial emotions of different varieties have been used for controlling behavior, e.g. in cognitive architectures and reinforcement learning models. We propose to use artificial emotions for a different purpose: controlling concept development. Dynamic networks with mech anisms for adding and removing nodes are more flexible than networks with a fixed topology, but if memories are added whenever a new sit uation arises, then these networks will soon grow out of proportion. Therefore there is a need for striking a balance that ideally ensures that only the most useful memories will be formed and preserved in the long run. Humans have a tendency to form and preserve memories of situa tions that are repeated frequently or experienced as emotionally intense (strongly positive or strongly negative), while removing memories that do not meet these criteria. In this paper we present a simple network model with artificial emotions that imitates these mechanisms.

@incollection{strannegaard2015,
  title = {Emotional Concept Development},
  author = {Stranneg{\aa}rd, Claes and Cirillo, Simone and Wessberg, Johan},
  booktitle = {Proceedings of the 8th International Conference on Artificial General Intelligence. {AGI} '15},
  publisher = {Springer},
  address = {Berlin, Germany},
  month = {July},
  year = {2015},
  pages = {362--372},
  doi = {http://dx.doi.org/10.1007/978-3-319-21365-1_37}
}

We propose a Genetic Programming architecture for the generation of foreign exchange trading strategies. The system's principal features are the evolution of free-form strategies which do not rely on any prior models and the utilization of price series from multiple instruments as input data. This latter feature constitutes an innovation with respect to previous works documented in literature. In this article we utilize Open, High, Low, Close bar data at a 5 minutes frequency for the AUD.USD, EUR.USD, GBP.USD and USD.JPY currency pairs. We will test the implementation analyzing the in-sample and out-of-sample performance of strategies for trading the USD.JPY obtained across multiple algorithm runs. We will also evaluate the differences between strategies selected according to two different criteria: one relies on the fitness obtained on the training set only, the second one makes use of an additional validation dataset. Strategy activity and trade accuracy are remarkably stable between in and out of sample results. From a profitability aspect, the two criteria both result in strategies successful on out-of-sample data but exhibiting different characteristics. The overall best performing out-of-sample strategy achieves a yearly return of 19%.

@article{cirillo2014evolving,
  title = {Evolving intraday foreign exchange trading strategies utilizing multiple instruments price series},
  author = {Cirillo, Simone and Lloyd, Stefan and Nordin, Peter},
  journal = {arXiv.org {[Preprint]}},
  year = {2014},
  month = {November},
  url = {http://arxiv.org/abs/1411.2153}
}

In this paper we describe a novel implementation of the Interpreter class for the metaheuristic optimization framework HeuristicLab, comparing it with the three existing interpreters provided with the framework. The Interpreter class is an internal software component utilized by HeuristicLab for the evaluation of the symbolic expression trees on which its Genetic Programming (GP) implementation relies. The proposed implementation is based on the creation and compilation of a .NET Expression Tree. We also analyze the Interpreters' performance, evaluating the algorithm execution times on GP Symbolic Regression problems for different run settings. Our implementation results to be the fastest on all evaluations, with comparatively better performance the larger the run population size, dataset length and tree size are, increasing HeuristicLab's computational efficiency for large problem setups.

@inproceedings{cirillo2014,
  title = {A Scalable Symbolic Expression Tree Interpreter for the HeuristicLab Optimization Framework},
  author = {Cirillo, Simone and Lloyd, Stefan},
  booktitle = {Proceedings of the Companion Publication of the 2014 Annual Conference on Genetic and Evolutionary Computation. {GECCO} '14},
  year = {2014},
  address = {Vancouver, Canada},
  month = {July},
  pages = {1141--1148},
  doi = {http://dx.doi.org/10.1145/2598394.2605692},
  owner = {Spooky},
  timestamp = {2014.07.20}
}

Progressive matrix problems are frequently used in modern IQ tests. In a progressive matrix problem, the task is to identify the missing element that completes the pattern of a pictorial matrix. We present a method for solving progressive matrix problems. The method is not limited to problems that are on the multiple choice format, which makes it potentially useful for solving real-world pattern discovery problems that do not come with predefined answer alternatives. The method is anthropomorphic in the sense that it uses certain problem solving strategies that were reported by high-achieving human solvers. We also describe a computer program implementing this method. The computer program was tested on the sets C, D, and E of Raven's Standard Progressive Matrices test and it produced correct solutions for 28 of the 36 problems considered. This score corresponds roughly to an IQ of 100. Finally, we conclude that it is possible to solve progressive matrix problems without analyzing potential answer alternatives and discuss some implications of this finding.

@article{strannegaard2013,
  title = {An Anthropomorphic Method for Progressive Matrix Problems},
  author = {Stranneg{\aa}rd, Claes and Cirillo, Simone and Str\"om, Victor},
  journal = {Cognitive Systems Research},
  year = {2013},
  month = {June},
  pages = {35--46},
  volume = {22--23},
  doi = {http://dx.doi.org/10.1016/j.cogsys.2012.08.002},
  keywords = {Anthropomorphism, Cognitive modeling, Inductive reasoning, Raven's Progressive Matrices},
  publisher = {Elsevier},
  timestamp = {2013.08.07}
}

The report describes a computer program for solving Raven's Progressive Matrices (RPM), a multiple choice test of abstract reasoning introduced by Dr. John C. Raven in 1936. Each RPM problem consists of a grid (or matrix) of 2x2 or 3x3 cells with graphical content, where the cell content in the bottom right corner is omitted; the solver's task is to pick the missing content from a set of eight solution candidates. We argue these problems are not only mathematical, but also psychological in nature. Due to this and other considerations such as algorithmic transparency, the program makes use of a simple cognitive model. The program solves RPM problems in a fully automatic fashion, without taking the solution candidates into account. The input is an RPM problem represented as a vector graphics file; the output is a complete or partial solution for the missing entry, represented in the same format. Internally we use multi-layered structures which enable the perception of the problems' different organizational levels. The program was tested on sections C, D and E of the Standard Progressive Matrices(SPM) and produced correct solutions for 28 of the 36 considered problems.

@MastersThesis{cirillo2010,
  Title                    = {An Anthropomorphic Solver For Raven's Progressive Matrices},
  Author                   = {Simone Cirillo and Victor Str\"om},
  School                   = {Chalmers University of Technology},
  Year                     = {2010},
  Address                  = {G\"oteborg, Sweden},
  Month                    = {June},
  Abstract                 = {The report describes a computer program for solving Raven's Progressive Matrices (RPM), a multiple choice test of abstract reasoning introduced by Dr. John C. Raven in 1936. Each RPM problem consists of a grid (or matrix) of 2x2 or 3x3 cells with graphical content, where the cell content in the bottom right corner is omitted; the solver's task is to pick the missing content from a set of eight solution candidates.
We argue these problems are not only mathematical, but also psychological in nature. Due to this and other considerations such as algorithmic transparency, the program makes use of a simple cognitive model.
The program solves RPM problems in a fully automatic fashion, without taking the solution candidates into account. The input is an RPM problem represented as a vector graphics file; the output is a complete or partial solution for the missing entry, represented in the same format. Internally we use multi-layered structures which enable the perception of the problems' different organizational levels.
The program was tested on sections C, D and E of the Standard Progressive Matrices(SPM) and produced correct solutions for 28 of the 36 considered problems.},
  Keywords                 = {Anthropomorphic Artificial Intelligence, Cognitive Model, General Artificial Intelligence, Intelligence Tests, Raven's Progressive Matrices},
  Timestamp                = {2013.08.07},
  Url                      = {http://studentarbeten.chalmers.se/publication/123536-an-anthropomorphic-solver-for-ravens-progressive-matrices}
}