Human cognition is unique in the way in which it relies on combinatorial (or compositional) structures. Language provides ample evidence for the existence of combinatorial structures, but they can also be found in visual cognition. To understand the neural basis of human cognition, it is therefore essential to understand how combinatorial structures can be instantiated in neural terms. In his recent book on the foundations of language, Jackendoff formulated four fundamental problems for a neural instantiation of combinatorial structures: the massiveness of the binding problem, the problem of 2, the problem of variables and the transformation of combinatorial structures from working memory to long-term memory. This paper aims to show that these problems can be solved by means of neural ‘blackboard ’ architectures. For this purpose, a neural blackboard architecture for sentence structure is presented. In this architecture, neural structures that encode for words are temporarily bound in a manner that preserves the structure of the sentence. It is shown that the architecture solves the four problems presented by Jackendoff. The ability of the architecture to instantiate sentence structures is illustrated with examples of sentence complexity observed in human language performance. Similarities exist between the architecture for sentence structure and blackboard architectures for combinatorial structures in visual cognition, derived from the structure of the visual cortex. These architectures are briefly discussed, together with an example of a combinatorial structure in which the blackboard architectures for language and vision are combined. In this way, the architecture for language is grounded in perception. 2 Content

In the theoretical framework described in this thesis, attention is part of the inference process that solves the visual recognition problem of what is where. The theory proposes a computational role for attention and leads to a model that predicts some of its main properties at the level of psychophysics and physiology. In our approach, the main goal of the visual system is to infer the identity and the position of objects in visual scenes: spatial attention emerges as a strategy to reduce the uncertainty in shape information while feature-based attention reduces the uncertainty in spatial information. Featural and spatial attention represent two distinct modes of a computational process solving the problem of recognizing and localizing objects, especially in difficult recognition tasks such as in cluttered natural scenes. We describe a specific computational model and relate it to the known functional anatomy of attention. We show that several well-known attentional phenomena – including bottom-up pop-out effects, multiplicative modulation of neuronal tuning

Cortical organization of vision appears to be divided into two pathways: the ventral pathway and the dorsal pathway. Models of vision have generally adopted this separation into a functional division such that recognition is supposed to be located in the ventral pathway and spatial attributes are processed in the dorsal pathway. I suggest a less distinct separation. According to my model the ventral pathway contributes to the selection of the location of an object by feedback connections.

Abstract—Traditional connectionist classification models place an emphasis on learned synaptic weights. Based on neurobiological evidence, a new approach is developed and experimentally shown to be more robust for disambiguating novel combinations of stimuli. It requires less training, variables and avoids many training related questions. Instead of determining all connection weights a-priori based on the training set, only positive binary associations are encoded (i.e. X has Y). Negative associations (i.e. X does not have Z) are not encoded, but inferred during the test phase through feedback connections. This allows the network to function outside its training distribution. For example, the network is able to recognize multiple stimuli even if it is only trained on single stimuli. We compare the accuracy and generalization of this network with traditional weight learning networks. Q I.

In the theoretical framework described in this thesis, attention is part of the inference pro-cess that solves the visual recognition problem of what is where. The theory proposes a computational role for attention and leads to a model that predicts some of its main prop-erties at the level of psychophysics and physiology. In our approach, the main goal of the visual system is to infer the identity and the position of objects in visual scenes: spa-tial attention emerges as a strategy to reduce the uncertainty in shape information while feature-based attention reduces the uncertainty in spatial information. Featural and spatial attention represent two distinct modes of a computational process solving the problem of recognizing and localizing objects, especially in difficult recognition tasks such as in clut-tered natural scenes. We describe a specific computational model and relate it to the known functional anatomy of attention. We show that several well-known attentional phenom-ena- including bottom-up pop-out effects, multiplicative modulation of neuronal tuning curves and shift in contrast responses- emerge naturally as predictions of the model. We also show that the bayesian model predicts well human eye fixations (considered as a proxy

Abstract—In this paper, we present a novel network to separate mixtures of inputs that have been previously learned. A significant capability of the network is that it segments the components of each input object that most contribute to its classification. The network consists of amplitude-phase units that can synchronize their dynamics, so that separation is determined by the amplitude of units in an output layer, and segmentation by phase similarity between input and output layer units. Learning is unsupervised and based on a Hebbian update, and the architecture is very simple. Moreover, efficient segmentation can be achieved even when there is considerable superposition of the inputs. The network dynamics are derived from an objective function that rewards sparse coding in the generalized amplitude-phase variables. We argue that this objective function can provide a possible formal interpretation of the binding problem and that the implementation of the network architecture and dynamics is biologically plausible. Index Terms—Binding problem, deconvolution, oscillations, phase correlation, separation of mixtures, synchronization.

In this paper, we investigate the relation between Artificial Neural Networks (ANNs) and networks of populations of spiking neurons. The activity of an artificial neuron is usually interpreted as the firing rate of a neuron or neuron population. Using a model of the visual cortex, we will show that this interpretation runs into serious difficulties. We propose to interpret the activity of an artificial neuron as the steady state of a cross-inhibitory circuit, in which one population codes for 'positive' artificial neuron activity and another for 'negative' activity. We will show that with this interpretation it is possible, under certain circumstances, to transform conventional ANNs (e.g. trained with 'backpropagation ') into biologically plausible networks of spiking populations. However, in general, the use of biologically motivated spike response functions introduces artificial neurons that behave differently from the ones used in the classical ANN paradigm. 2001 Elsevier Science Ltd. All rights reserved.

Abstract. We present a neural architecture of sentence representation. Words are represented with neural cell assemblies. Relations between words are represented with ‘structure ’ assemblies. Word and structure assemblies are bound temporarily to form a sentence representation. We show how multiple sentences can be represented simultaneously, and we simulate how specific information can be retrieved from the architecture. The assemblies are simulated as populations of spiking neurons, in terms of the average firing rate of the neurons in the population. 1.

ABSTRACT. The problem of visual feature binding and the unity of an object in visual consciousness is discussed in relation to the account of these phenomena presented by interactive hierarchical structuralism. It is argued that the binding problem should be studied and solved from the intrinsic perspective, given by the information that is available within the system (i.e., brain) itself. This kind of information is always local. Therefore, the intrinsic perspective induces a process approach to solving the binding problem which depends on global information processed in different areas within the brain. The interaction between feedforward and feedback activity in the visual cortex is a process that solves the binding problem of visual features. A similar process could underlie visual awareness and the unity of an object in visual consciousness. It results in a sequential form of awareness, in which the awareness of one of the features of an object induces the awareness of its other features. KEY WORDS: attention, awareness, binding, consciousness, intrinsic perspective,

The number of objects that can be maintained in visual working memory without interference is limited. We present simulations of a model of visual working memory in ventral prefrontal cortex that has this constraint as well. One layer in ventral PFC constitutes a 'blackboard ' representation of all objects in memory. These representations are used to bind the features (shape, color, location) of the objects. If there are too many objects, their representations will interfere in the blackboard and therefore the quality of these representations will degrade.