In two studies about farming practices, the respondents who are particularly favorable to organic farming tend to have a higher intention to convert their farm to organic when they perceive other farmers as not very favorable to this practice. This intention can be considered as anticonformist, as it is in opposition to the general view of others. This article hypothesizes that this phenomenon can be explained by some biases on the perceptions of attitudes. It proposes an agent-based model which computes an intention based on the theory of reasoned action (TRA) and assumes some biases in the perception of others' attitudes according to the social judgment theory. It investigates the conditions on the model parameter values for which the simulations reproduce the features observed in the studies. The results show that perceptual biases are a possible explanation of anticonformist intentions.

This article focuses on an artificial life approach to some important problems in machine learning such as statistical discrimination, curve approximation, and pattern recognition. We describe a family of models, collectively referred to as semi-algebraic networks (SAN). These models are strongly inspired by two complementary lines of thought: the biological concept of autopoiesis and morphodynamical notions in mathematics. Mathematically defined as semi-algebraic sets, SANs involve geometric components that are submitted to two coupled processes: (a) the adjustment of the components (under the action of the learning examples), and (b) the regeneration of new components. Several examples of SANs are described, using different types of components. The geometric nature of SANs gives new possibilities for solving the bias/variance dilemma in discrimination or curve approximation problems. The question of building multilevel semi-algebraic networks is also addressed, as they are related to cognitive problems such as memory and morphological categorization. We describe an example of such multilevel models.

The perceptron membrane is a new connectionist model that aims at solving discrimination (classification) problems with piecewise linear surfaces. The discrimination surfaces of perceptron membranes are defined by the union of convex polyhedrons. Starting from only one convex polyhedron, new facets and new polyhedrons are added during learning. Moreover, the positions and orientations of the facets are continuously adapted according to the training examples. Considering each facet as a perceptron cell, a geometric credit assignment provides a local training domain to each perceptron of the network. This enables one to apply statistical theorems on the probability of good generalization for each unit on its learning domain, and gives a reliable criterion for perceptron elimination (using Vapnik-Chervonenkis dimension). Furthermore, a regularization procedure is implemented. The model efficiency is demonstrated on well-known problems such as the 2-spirals or waveforms.

The neural units recruitment algorithms (NEURAL) are algorithms mixing techniques from neural networks and symbolic machine learning. The size and architecture of the network are not specified before the learning process. Basic cells (perceptron units or delta rule units) are recruited and organized in a structure similar to the one of a decision tree which grows during the learning process (beginning with only one initial cell). As the tree grows, the leaf cells are specialized in smaller and smaller parts of the initial training set. Convergence is guaranteed for any set of patterns, with real inputs and Boolean outputs. Learning can be incremental: learning a new pattern set only alters the parts of the structure concerned with the differences between the old and the new sets