Neill R. Taylor

2006 Special Issue: Attention as a controller

We investigate, by constructing suitable models, the manner in which attention and executive function are observed to interact, including some aspects of the influence of value/emotion on this interaction. Attention is modelled using the recent engineering control model (Corollary Discharge of Attention Movement, CODAM), which includes suitable working memory components. We extend this model to take account of various executive functions performed in working memory under attention control, such as rehearsal, substitution and transformation of buffered activity. How these are achieved is specified in suitable extension of CODAM. Further extensions are then made to include emotional values of stimuli. All of these extensions are supported by recent experimental brain imaging data on various working memory tasks, which are simulated with reasonable accuracy. We conclude our analysis by a discussion on the nature of cognition as seen in terms of the resulting extended attention model framework.

Hard-wired models of working memory and temporal sequence storage and generation

We develop hard-wired simulations of temporal sequence storage and generation by multi-modular networks based on the frontal lobe system (cortex, basal ganglia and thalamus). Single cell activity is shown to have been constructed so as to mimic results measured in monkeys on a similar task, indicating that a suitable form of chunking had been achieved in the models. The mathematical nature of these processes is discussed, from the viewpoint of bifurcation theory.

Occlusion, attention and object representations

Occlusion is currently at the centre of analysis in machine vision. We present an approach to it that uses attention feedback to an occluded object to obtain its correct recognition. Various simulations are performed using a hierarchical visual attention feedback system, based on contrast gain (which we discuss as to its relation to possible hallucinations that could be caused by feedback). We then discuss implications of our results for object representations per se.

Understanding spike-time-dependent plasticity: A biologically motivated computational model

Abstract Long-term synaptic plasticity underlies many important learning processes in the brain. Recent physiological data have shown that the precise relative timings of pre- and post-synaptic neuron firings at a synapse determine both the direction of certain types of modification (potentiation or depression), and magnitude of this modification. We propose a neurophysiological mechanism by which this spike-time-dependent plasticity (STDP) could arise, and support this hypothesis using a model involving calcium dynamics. We show that, in addition to reproducing experimental data for paired spikes, the model can explain differences in experimentally observed STDP forms. We also demonstrate that the model provides a good match to recent data for the triplet and quadruplet paradigms, and that a simulated network of reciprocally connected neurons using this learning rule can store and recall a simple temporal sequence. In conclusion we make predictions from the model both on the plasticity effects of quadruple spike interactions and manipulations of concentrations of components involved at the synapse.

A hierarchical attention-based neural network architecture, based on human brain guidance, for perception, conceptualisation, action and reasoning

We present a neural network software architecture, guided by that of the human and more generally primate brain, for the construction of an autonomous cognitive system (which we have named GNOSYS). GNOSYS is created so as to be able to attend to stimuli, to conceptualise them, to learn their predicted reward value and reason about them so as to attain those stimuli in the environment with greatest predicted value. We apply this software system to an embodied version in a robot, and describe the activities in the various component modules of GNOSYS, as well as the overall results. We briefly compare our system with some others proposed to have cognitive powers, and finish by discussion of future developments we propose for our system, as well as expanding on the arguments for and against our approach to creating such a software system.

Attention as sigma-pi controlled ACh-based feedback

We analyse experimental data on attention to indicate that any attention feedback control signals to lower order cortical sites will lead to a quadratic sigma-pi form of output in its dependence on the lower-order input and the feedback signal. The manner by which this structure works is shown by a brief simulation. We then discuss how such a structure could arise from the action of diffuse acetylcholine signals from the NBM, especially involving nicotinic receptors. We deduce certain structural regularities which should be expected both at local-and at micro-circuit level, mainly in cortical layer V (the output layer).

Coding of objects in low-level visual cortical areas

We develop a neural network architecture to help model the creation of visual temporal object representations. We take visual input to be hard-wired up to and including V1 (as an orientation-filtering system). We then develop architectures for afferents to V2 and thence to V4, both of which are trained by a causal Hebbian law. We use an incremental approach, using sequences of increasingly complex stimuli at an increasing level of the hierarchy. The V2 representations are shown to encode angles, and V4 is found sensitive to shapes embedded in figures. These results are compared to recent experimental data, supporting the incremental training scheme and associated architecture.

Learning to generate temporal sequences by models of frontal lobes

Possible architectures for the storage and retrieval of temporal sequences are investigated using a cartoon version of the frontal lobes (the ACTION net), based loosely on actual frontal architecture. A particular version of the architecture is developed which gives similar temporal patterns of single neurone activity as observed experimentally in monkeys. The underlying principles involved in the manner that chunking is achieved of these sequences is deduced in the final part of the paper.

Experimenting with Models of the Frontal Lobes

We develop simulations of temporal sequence storage and generation by multi-modular networks based on the frontal lobe system (cortex, basal ganglia and thalamus). Single cell activity is shown to develop so as to mimic results measured in monkeys on a similar task, indicating that a suitable form of chunking had been achieved.

2006 Special Issue: The micro-structure of attention

We investigate three possible methods of specifying the microstructure of attention feedback: contrast gain, additive and output gain, using simple single node and 3-layer cortical models composed of graded or spiking neurons. Contrast gain and additive attention are also tested in a spiking network which is simplified by mean field methods. The simulation task uses two stimuli, probe and reference, presented singly or together within the neuronal receptive fields whilst attention is directed towards or away from the receptive field. Model neurons are differentially activated in the different stimuli and attention and equilibrium potentials or average firing rates recorded depending on neuron type are recorded. We compare results for the different modes of attention and architectures with experimental single cell recordings which show how neuronal firing rates change in response to attention, with a bias towards neurons that respond more effectively to the attended stimulus, to investigate which attentional method best fits the experimental data. The simulation results are also mathematically analysed. We conclude that there is most experimental support for contrast gain, although some additional feedback gain would be possible. We propose a tentative method by which attention as contrast gain may occur in the primate brain using acetylcholine and nicotinic receptors.