Universal computation by networks of model cortical columns

Journal of Physiology-Paris, 1994

We describe a biologically plausible modelling framework based on the architectural and processing characteristics of the cerebral cortex. Its key feature is a multicellular processing unit (cortical column) reflecting the modular nature of cortical organization and function. In this framework, we describe a neural network model of the neuronal circuits of the cerebral cortex that learn different functions associated with different parts of the cortex: I) visual integration for invariant pattern recognition, performed by a cooperation between temporal and parietal areas; 2) visual-to-motor transformation for 3D arm reaching movements, performed by parietal and motor areas; and 3) temporal integration and storage of sensorimotor programs, performed by networks linking the prefrontal cortex to associative sensory and motor areas. The architecture of the network is inspired from the features of the architecture of cortical pathways involved in these functions. We propose two rules which describe neural processing and plasticity in the network. The first rule (adaptive tuning if gating) is an analog of operant conditioning and permits to learn to anticipate an action. The second rule (adaptive timing) is based on a bistable state of activity and permits to learn temporally separate events forming a behavioral sequence. neural modelling / cerebral cortex / cortical column / learning / sensorimotor programs

2011

Abstract We argue that neural processes are neither analog nor digital computations; they constitute a third kind of computation. Analog computation is the processing of continuous signals; digital computation is the processing of strings of digits. But current neuroscientific evidence indicates that typical neural signals, such as spike rates, are graded like continuous signals but are constituted by discrete elements (spikes); thus typical neural signals are neither continuous signals nor strings of digits.

Brain-inspired computing structures,technologies, and methods offer innovative approaches to the future of computing. From the lowest level of neuron devices to the highest abstraction of consciousness, the brain drives new ideas (literally and conceptually) in computer design and operation. This paper interrelates three levels of brain inspired abstractions including intelligence, abstract graph data structures, and neuron operation and interconnection. An abstract machine architecture is presented from which a lower bound on resource requirements for intelligence is to be derived. At the lowest level a new use of cellular automata architecture is discussed that mimics the fine-grain locality of action and high degree interconnectivity of neurons and their structures. Graph structures serve as a brain inspired intermediary abstraction between these two as the neocortex is organized as a directed graph. This paper shows how all of the pieces tie together and opens a new way of considering future computing structures through brain inspired concepts.

Science, 2006

Biologically Based Computational Models of

Qeios

The authors discuss a functional unit of neuronal circuits, crucial in both brain structure and artificial intelligence (AI) systems. These units recognize objects denoted by single words or verbal phrases and are responsible for object perception, memorization of image patterns, and retrieval of those patterns as imagination. Functional neuronal units, activated by the working memory system, contribute to problem-solving. The paper highlights the authors' achievements in developing a theory for these functional units, known as 'the equimerec - units', which combine 'the threshold logic unit' and 'the feedback control loop'. The authors emphasize the importance of this theory, especially in the context of highly advanced language-based AI systems. Additionally, the authors note that the functional units' essence relies on backpropagation connections, causing impulse circulation in closed circuits, ultimately leading to the emergence of an electromagne...

The brain's architecture is usually discussed as a hierarchy where information flows in a bottom-up manner. However, local and recurrent cortical connections form an important aspect of the brain's architecture and it is not quite clear what is the function of this "horizontal" form of connectivity. In this theoretical paper, we aim to provide one possible explanation to this question using the perspective of abstract algebra and more specifically the one of Category Theory. We propose that recurrent interactions can be modeled by using an abstract structure which is the Groupoid. This abstract structure may explain the role of recurrent local circuits in natural computing machines with multi layers of representations.

Natural Computing, 2003

The first two main rounds of neural computing focused on adaptation and selforganization in neural networks, and on use of analog VLSI for compartmental modeling of the neuron, respectively. This paper is a prospectus for a third round of neural computing: analyzing the architecture of the primate brain to extract neural information processing principles and translate them into biologically-inspired operating systems and computer architectures. The way in which the cerebellum interacts with other brain regions in learning how to better control and coordinate movements provides a case study to introduce key ideas for these three rounds of neural computation. It is argued that the third round will develop and exploit general insights into brain architectures, their function and dynamics, that will provide a principled combination of cooperative computation, learning and perceptual robotics (i.e., the integration of action, perception and computation). This new effort is motivated by recent advances in computational neuroscience research, studying examples of system evolution ranging from low-level vision to how the interactions between frontal and parietal cortices serve action recognition (the mirror system), and language. The paper also notes the vast difference between the slow biological evolution of diverse brains and the needs of computer technology for explicit tools for the design and testing of novel programs and architectures, and suggests the challenges of developing a reflection methodology for wrapping modules with descriptions that can be automatically updated as the module itself adapts through learning. The Appendix provides a road map for approaching the voluminous literature on brain theory and neural networks.

Unification-based approaches have come to play an important role in both theoretical and applied modeling of cognitive processes, most notably natural language. Attempts to model such processes using neural networks have met with some success, but have faced serious hurdles caused by the limitations of standard connectionist coding schemes. As a contribution to this effort, this paper presents recent work in Infinite RAAM (IRAAM), a new connectionist unification model. Based on a fusion of recurrent neural networks with fractal geometry, IRAAM allows us to understand the behavior of these networks as dynamical systems. Using a logical programming language as our modeling domain, we show how this dynamical-systems approach solves many of the problems faced by earlier connectionist models, supporting unification over arbitrarily large sets of recursive expressions. We conclude that IRAAM can provide a principled connectionist substrate for unification in a variety of cognitive modeling domains.

Proceedings of the XXVII Annual Conference of the …, 2005

I present a cognitive model, dubbed BioSLIE, that integrates and extends recent advances in: 1) distributed, structuresensitive representation; 2) neurocomputational modeling; and 3) our understanding of the neuroanatomy of inference. As a result, BioSLIE is biologically detailed, learns different behaviors in different contexts, and exhibits systematic, structuresensitive generalization. Here, BioSLIE is applied to the Wason card selection task. Its performance meets Cosmides' (1989) challenge to mechanistically define domain-general procedures that can use induction to produce the observed domain specific performance on the Wason task. As well, it demonstrates the relevance of neural computation to understanding cognition, despite claims to the contrary by and Jackendoff .

2012

DRAFT – Please do not quote or share without permission We argue that neural processes are neither analog nor digital computations; they constitute a third kind of computation. Analog computation is the processing of continuous signals; digital computation is the processing of strings of digits. But current neuroscientific evidence indicates that typical neural signals, such as spike rates, are graded like continuous signals but are constituted by discrete elements (spikes); thus typical neural signals are neither continuous signals nor strings of digits. It follows that neural computation is sui generis. This has three important consequences. First, understanding neural computation requires a specially designed mathematical theory (or theories) rather than the mathematical theories of analog or digital computation. Second, several popular views about neural computation turn out to be incorrect. Third, computational theories of cognition that rely on non-neural notions of computatio...