In this study, we step aside from traditional fusion strategies and propose to use a bidirectional associative memory to combine abstract representations of several modalities. The innovative contribution of our strategy is the fact that the fusion is performed neither at the input data level nor at the output score level. The model achieves outperforming results on depression evaluation from the videos of the AVEC2014 challenge corpus.

Deep Neural Networks (DNN) propose a new and efficient ML architecture based on the layer-wise building of several representation layers. A critical issue for DNNs remains model selection, e.g. selecting the number of neurons in each DNN layer. The hyper-parameter search space exponentially increases with the number of layers, making the popular grid search-based approach used for finding good hyper-parameter values intractable. The question investigated in this paper is whether the unsupervised, layer-wise methodology used to train a DNN can be extended to model selection as well. The proposed approach, considering an unsupervised criterion, empirically examines whether model selection is a modular optimization problem, and can be tackled in a layer-wise manner. Preliminary results on the MNIST data set suggest the answer is positive. Further, some unexpected results regarding the optimal size of layers depending on the training process, are reported and discussed.

Many complex systems, ranging from neural cell assemblies to insect societies, involve and rely on some division of labor. How to enforce such a division in a decentralized and distributed way, is tackled in this paper, using a spiking neuron network architecture. Specifically, a spatio-temporal model called SpikeAnts is shown to enforce the emergence of synchronized activities in an ant colony. Each ant is modelled from two spiking neurons; the ant colony is a sparsely connected spiking neuron network. Each ant makes its decision (among foraging, sleeping and self-grooming) from the competition between its two neurons, after the signals received from its neighbor ants. Interestingly, three types of temporal patterns emerge in the ant colony: asynchronous, synchronous, and synchronous periodic foraging activities - similar to the actual behavior of some living ant colonies. A phase diagram of the emergent activity patterns with respect to two control parameters, respectively accounting for ant sociability and receptivity, is presented and discussed.

Recent electrophysiological data show that synaptic weights are highly influenced by electrical activities displayed by neurons. Weights are not stable as assumed in classical neural network models. What is the nature of engrains, if not stored in synaptic weights? Adopting the theory of dynamical systems, which allows an implicit form of memory, we propose a new framework for learning, where synaptic weights are continuously adapted. Evolutionary computation has been applied to a population of dynamic neural networks evolving in a prey-predator environment. Each individual develops complex dynamic patterns of neuronal activity, underlied by multiple recurrent connections. We show that this method allows the emergence of learning capability through generations, as a byproduct of evolution, since the behavioural performance of the network is not a priori based on this property.

We present in this article a dynamic neural network that works as a memory for multiple associations. Heterogeneous pairs of patterns can be tied together through learning within this memory, and recalled easily. Starting from Kosko's bidirectional associative memory, we modify some fundamental features of the network (topology and learning algorithm). We show empirically that this network has a high storage capacity and is only weakly dependent upon learning hyperparameters. We demonstrate its robustness to corrupted or missing data. We finally present results from experiments where this network is used as a multisensory associative memory.

We introduce a support vector machine devoted to the approximation of multi-class discriminant functions. Its training procedure consists in minimizing an expression of the guaranteed risk. This bound is significantly tighter than the former ones, which should make the implementation of the structural risk minimization inductive principle in the context of multi-class discrimination better grounded.

Starting from a direct definition of the notion of margin in the multiclass case, we study the generalization performance of multiclass discriminant systems. In the framework of statistical learning theory, we establish on this performance a bound based on covering numbers. An application to a linear ensemble method which estimates the class posterior probabilities provides us with a way to compare this bound and another one based on combinatorial dimensions, with respect to the capacity measure they incorporate. Experimental results highlight their usefulness for a real-world problem.