In this paper, an adaptation of the k-means algorithm and related methods to non-Euclidian topology is presented.The paper introduces a rationale for approximating the geodesics of that topology, as well as a learning rule that is robust to noise.The first results on artificial but very noisy distributions presented here are promising for further experimentation on real cases.

The motivation of our work is the instantiation of a computational view of the cerebral cortex. Kohonen’s early definition of self-organizing maps was inspired by the cortical substrate on a local scale and is now a widely used learning algorithm. Following the same path, from biology to computation, the cortex can be interpreted as an architecture made of similar self-organizing modules connected together. To our knowledge, there are no such algorithmic derivation of large architectures of self-organizing modules. This paper presents the behavior of several maps connected one to another as a step towards wider networks of self-organizing maps and shows that this architecture learns a model of inputs and generates predictions in a map without using an additional algorithm. This prediction ability is applied to the control of a quadcopter flying in a corridor.

This paper introduces CxSOM, a model to build modular architectures based on self-organizing maps (SOM). An original consensus driven approach enables to adress non-hierarchical architectures where SOMs get organized jointly. The paper aims at showing how the modules are able to store the association between data, and evaluating, by a mutual information criterion, the resulting organization. These results stand as preliminary work to study bigger architectures.

In this paper, the ability of 1D-SOMs to address the Euclidian Travelling Salesperson problem is extended to more irregular topologies, in order to compute short closed paths covering an indoor environment. In such environments, wall constraints makes the topology of the area to be visited by a patroller very irregular. An application to indoor unmanned aerial vehicule (UAV) security patrols is considered.

This paper introduces representations and measurements for revealing the inner self-organization that occurs in a 1D recurrent self-organizing map. Experiments show the incredible richness and robustness of an extremely simple architecture when it extracts hidden states of the HMM that feeds it with ambiguous and noisy inputs.

Quand les ´etats d’un processus ne sont pas Markoviens (POMDP par exemple), la convergence des algorithmes d’apprentissage par renforcement n’est pas garantie. Une solution est de reconstruire un processus Markovien en partant de la s´equence des ´etats. Dans ce but, nous explorons les capacit´es d’architectures r´ecurrentes qui s’appuient sur des cartes neuronales autoorganisatrices pour apprendre `a pr´edire des s´equences d’observations issues de HMM.

We introduce a new rule based system for belief tracking in dialog systems. Despite the simplicity of the rules being considered, the proposed belief tracker ranks favourably compared to the previous submissions on the second and third Dialog State Tracking challenges. The results of this simple tracker allows to reconsider the performances of previous submissions using more elaborate techniques.

We introduce a new rule based system for belief tracking in dialog systems. Despite the simplicity of the rules being considered, the proposed belief tracker ranks favourably compared to the previous submissions on the second and third Dialog State Tracking challenges. The results of this simple tracker allows to reconsider the performances of previous submissions using more elaborate techniques.

This paper presents a vector quantization process that can be applied online to a stream of inputs. It enables to set up and maintain a dynamical representation of the current information in the stream as a topology preserving graph of prototypical values, as well as a velocity field. The algorithm relies on the formulation of the accuracy of the quantization process, that allows for both the updating of the number of prototypes according to the stream evolution and the stabilization of the representation from which velocities can be extracted. A video processing application is presented.

This paper presents a multi-map joint self-organizing architec- ture able to represent non-markovian temporal sequences. The proposed architecture is inspired by previous works based on dynamic neural fields. It provides a faster and easier to handle architecture making it easier to scale to higher dimensional machine learning problems.

This research projects takes inspiration from the human brain as a fruitful hint towards the design of a computational architecture able to understand natural scenes. More precisely, in the case of visual understanding, many works in biology and psychology stressed that vision is rather a sequential palpation of details than a global computation of the whole surrounding scene. This palpation is driven by neural structures, based on highly coupled populations of elementary computing units, i.e. the neurons, whose dynamics leads to a global, robust and harmonious signal processing resulting in a given animal behaviour. Such properties are, on the one hand, the result of evolution which brought the anatomical structure of our brains to maturity. They are also due, on the other hand, to the massive self-organization processes inside the brain, at the timescale of an individual’s

This paper shows how partial differential problems can be solved thanks to cellular computing and an adaptation of the Least Squares Finite Elements Method. As cellular computing can be implemented on distributed parallel architectures, this method allows the distribution of a resource demanding differential problem over a computer network.

This paper introduces the rllib as an original C++ template-based library oriented toward value function estimation. Generic programming is promoted here as a way of having a good fit between the mathematics of reinforcement learning and their implementation in a library. The main concepts of rllib are presented, as well as a short example.

In this paper, a distributed recurrent self-organizing architecture is presented. It can extract the current state of a dynamical system from the sequence of the recent observations provided by this system, even if they are ambiguous. The recurrent network is an adaptation of RecSOM to the context of the simulation of large scale distributed neural architectures, since it relies on a strictly local fine-grained computation. The experiments show the ability of the recurrent architecture to capture the states, but also exhibit some unexpected dynamical effects, like some instabilities of the learned mappings. The presented architecture addresses the cognitive ability to set up representations from sequences at a mesoscopic level. At that intermediate level, between cognition and neurons simulation, some complex dynamics is unveiled. It needs to be identified and understood in order to bridge the gap between neuronal activities and high level cognition.

InterCell is an open and operational software suite for implementation, code generation and interactive simulation of fine grained parallel computational models. This article describes the software architecture, some use cases from physics and cortical networks as well as first performance measurements.

This paper presents a multi-map recurrent neural architecture, exhibiting self-organization to deal with the partial observations of the phase of some dynamical system. The architecture captures the dynamics of the system by building up a representation of its phases, coping with ambiguity when distinct phases provide identical observations. The architecture updates the resulted representation to adapt to changes in its dynamics due to self-organization property. Experiments illustrate the dynamics of the architecture when fulfilling this goal.

—This paper presents a self-organizing architecture made of several maps, implementing a recurrent neural network to cope with partial observations of the phase of some dynamical system. The purpose of self-organization is to set up a distributed representation of the actual phase, although the observations received from the system are ambiguous (i.e. the same observation may correspond to distinct phases). The setting up of such a representation is illustrated by experiments, and then the paper concludes on extensions toward adaptive state representations for partially observable Markovian decision processes.

In this paper, dynamic neural fields (DNFs) are used to develop key features of a cortically-inspired computational module. Under the perspective of designing computational systems that can exhibit the flexibility and genericity of the cortical substrate, using neural field as the competition layer for self-organising modules has to be considered. However, despite the fact that they serve as a biologically-inspired model, applying DNFs to drive self-organisation is not straightforward. In order to address that issue, an original method for evaluating neural field equations is proposed, based on statistical measurements of the field behaviour in some scenarios. Limitations of classical neural field equations are then quantified, and an original field equation is proposed to overcome these difficulties. The performance of the proposed field model is discussed in comparison with some previously considered models, leading to the promotion of the proposed model as a suitable mean for processing competition in cortex-like computation for cognitive systems.