Francisco Coelho

Towards Individual Power Design

We started the setup of a virtual laboratory for the study of power struggles where autonomous agents embody different capabilities. The scenario is a caricature of hierarchical production societies which are driven to be structured into different organizational forms. Under the assumption of limited rationality, we look at diverse mechanisms able to drive theagent choice of action and to act. In our simple setup, an agent will look to shift to a new leader when placed in a poor individual situation. On the other hand, if the agent is comfortable, it might lower the pression imposed to its subordinates. Our agent models are yet too naive to distinguish motivation from will: the decision procedures aren’t sophisticated enough to enable the consideration of reasoning processes running at each deliberation step. Thus we can’t identify reasons (or motives) driving action selection. However, there is a causal relation from agent’s state and its choosing process (but not its action).

The euclid abstract machine: trisection of the angle and the halting problem

What is the meaning of hypercomputation, the meaning of computing more than the Turing machine? Concrete non-computable functions always hide the halting problem as far as we know. Even the construction of a function that grows faster than any recursive function — the Busy Beaver — a more natural function, hides the halting function, that can easily be put in relation with the Busy Beaver. Is this super-Turing computation concept related only with the halting problem and its derivatives? We built an abstract machine based on the historic concept of compass and ruler construction which reveals the existence of non-computable functions not related with the halting problem. These natural, and the same time, non-computable functions can help to understand the nature of the uncomputable and the purpose, the goal, and the meaning of computing beyond Turing.

Predictability for autonomous decision support

The experimental scientist need tools to quantify and classify collected data. This paper proposes to give meaning and measure to the intuitive concept of predictability. It is a global and time dependent real valued quantity that, we argue, indicates how hard it is to make a forecast for the next value on a time series. We start with a a definition of predictability for binary words and show properties about its growth and computational cost. Our measure evaluates in time On3, what is an acceptable performance specially for supporting bounded time decisions. Then, we investigate application procedures illustrated with data achieved from iterations of the logistic map, economic simulations and the Portuguese GDP (Gross Domestic Product).

Discussing Agent Power-Of by Experimentation

Does an agent mental space constitute the environment for a second agent? Can we use a second agent to percept and manipulate the individual entities that compose a certain agent? What advantages can be achieved? And with what costs? This paper covers a research path on the advantage side questions. We did again an experiment proposed by Steels (Mars explorer subsumption agent): Its list of reaction pairs is initially shuffled and a meta-agent can promote or demote those pairs in response to individual-power local variations. After a while the observed list of reaction pairs (and the implied agent behavior) is very close to the original and intended design

A method for regularization of evolutionary polynomial regression

Display Omitted A regularization term is proposed to control complexity in polynomial regression using genetic algorithms.Regularization reduces out-of-sample error with respect to polynomials found by non-regularized methods.Regularization improves convergence speed.Error performance is empirically evaluated on some common datasets versus standard regression methods. While many applications require models that have no acceptable linear approximation, the simpler nonlinear models are defined by polynomials. The use of genetic algorithms to find polynomial models from data is known as evolutionary polynomial regression (EPR). This paper introduces evolutionary polynomial regression with regularization, an algorithm extending EPR with a regularization term to control polynomial complexity. The article also describes a set of experiences to compare both flavors of EPR against other methods including linear regression, regression trees and support vector regression. These experiments show that evolutionary polynomial regression with regularization is able to achieve better fitting and needs less computation time than plain EPR.

Probabilistic Perception Revision in AgentSpeak(L)

Agent programming is mostly a symbolic discipline and, as such, draws little benefits from probabilistic areas as machine learning and graphical models. However, the greatest objective of agent research is the achievement of autonomy in dynamical and complex environments — a goal that implies embracing uncertainty and therefore the entailed representations, algorithms and techniques. This paper proposes an innovative and conflict free two layer approach to agent programming that uses already established methods and tools from both symbolic and probabilistic artificial intelligence. Moreover, this method is illustrated by means of a widely used agent programming example, GoldMiners.

A recursive process related to a partisan variation of Wythoff

Abstract. Wythoff Queens is a classical combinatorial game related to very interesting mathematical results. An amazing one is the fact that the -positions are given by and where . In this paper, we analyze a different version where one player (Left) plays with a chess bishop and the other (Right) plays with a chess knight. The new game (call it Chessfights) lacks a Beatty sequence structure in the -positions as in Wythoff Queens. However, it is possible to formulate and prove some general results of a general recursive law which is a particular case of a Partizan Subtraction game.

Meta-agency and individual power

The state of the art in the agent field is at the frontier of deploying real intelligent agents, able to build their own understanding of the world and of associated processes. Maybe that goal can be achieved through individual power: a framing concept for the agents inner capabilities to achieve dominant goals. The research on individual power started from the BDI model and led to the distinction between external and internal factors in the agents behaviour: while the former depends on the relation the agent has with its environment (including peers) and ties in with social power, the latter links up to the agent control and is rather less understood. In the individual power study, the focus is set on the integration of the agents will, commitment and control structure. Such integration can be interpreted as an extension of the classic belief-desire-intention trilogy. This paper describes an experiment about the integration of will and commitment management in the agents control structure. The experiment consists of two teams of agents, competing for resource gathering. While one team features reactive control, the other team uses new ideas exploring meta-agency for agent control and will for behaviour design. This experiment illustrates the integration and a virtual application of previous work on these subjects. The meta-agent and mental states based control is proven able to define flexible and effective behaviours. The agents will, by centralising the management of commitments, simplifies the implementation, debugging and interpretation of such behaviours.