Roman R. Poznanski, B.Sc (Hons), M.Sc., Ph.D.
Roman R. Poznanski is a leading authority on modeling in the neurosciences. He has written papers in prestigious high impact ISI-indexed neuroscience journals, including the Journal of Integrative Neuroscience: an interdisciplinary journal that combines theoretical models of information handling in the brain with experimental brain research.
In computational neuroscience, reductionist approaches span multiple levels of neural organization; however in integrative neuroscience, each level is seamlessly sculptured as part of a continuum of levels. Reductionism assumes a direct causal relationship between a molecular/cellular mechanism and a behavioral phenomenon, ignoring the constraints that higher-level properties exert on the possible functions of that mechanism. One of these constraints is dynamic continuity which is intrinsically difficult to harness computationally because compartmentalization and/or discretization is subject to dynamical misalignment, producing a false sense of biological reality.
In the dynamicist view, cognitive processes reflect upon dynamical transitions between regions of state-space, and rules are embedded in the dynamics of the system. Cognitive semantics are represented by trajectories in state-space. However, the neural basis of rule-guided behavior linking neurodynamics with cognitive semantics is still missing because current approximations to neural dynamics are based on artificial neural networks (i.e., spiking neural networks) that rely on the doctrine of biological computation . An alternative integrative approach is through relational biology where the functional relations between neurobiological processes depend on hierarchical and functional integration in the brain. From this perspective, hierarchical integration is structural involving dynamic continuity in spatiotemporal patterns, which brings about functional organization, while functional integration is relational enabling a relational organization to be mapped from the functional organization. Selectionism driven by dynamic continuity across scale produces an adaptive pressure that enables higher-order cognitive processes to spontaneously fathom unconscious experiential foundations for choices and preferences.
In computational neuroscience, reductionist approaches span multiple levels of neural organization; however in integrative neuroscience, each level is seamlessly sculptured as part of a continuum of levels. Reductionism assumes a direct causal relationship between a molecular/cellular mechanism and a behavioral phenomenon, ignoring the constraints that higher-level properties exert on the possible functions of that mechanism. One of these constraints is dynamic continuity which is intrinsically difficult to harness computationally because compartmentalization and/or discretization is subject to dynamical misalignment, producing a false sense of biological reality.
In the dynamicist view, cognitive processes reflect upon dynamical transitions between regions of state-space, and rules are embedded in the dynamics of the system. Cognitive semantics are represented by trajectories in state-space. However, the neural basis of rule-guided behavior linking neurodynamics with cognitive semantics is still missing because current approximations to neural dynamics are based on artificial neural networks (i.e., spiking neural networks) that rely on the doctrine of biological computation . An alternative integrative approach is through relational biology where the functional relations between neurobiological processes depend on hierarchical and functional integration in the brain. From this perspective, hierarchical integration is structural involving dynamic continuity in spatiotemporal patterns, which brings about functional organization, while functional integration is relational enabling a relational organization to be mapped from the functional organization. Selectionism driven by dynamic continuity across scale produces an adaptive pressure that enables higher-order cognitive processes to spontaneously fathom unconscious experiential foundations for choices and preferences.
 |
Roman R. Poznanski was awarded a certificate for teaching excellence from the University of Malaya where he was a visiting professor in 2009-2010. |
His main contribution to research began on the modeling of retinal neurons in visual perception with W.R.Levick, FRS from the John Curtin School of Medicine, Australian National University in Canberra. His modeling work was first to predict the locus of retinal direction selectivity in individual dendritic branches of starburst amacrine cells in the retina in 1990 (and published in 1992). He subsequently developed a more accurate model of a starburst amacrine cell in order to show how direction-selectivity is processed by a network of these cells. The unification of the yet unknown subcellular processing of retinal direction selectivity with starburst network topology remains one of his current research themes together with Amane Koizumi of the National Institute for Physiological Sciences, Okazaki, Japan. In partiuclar, on establishing the topology underlying a presynaptic scheme of directional selectivity, he proposed an acetylcholine (ACh) transmission model of directionally selective starburst amacrine cells (Poznanski, R.R. Cellular Inhibitory Behavior Underlying the Formation of Retinal Direction Selectivity, J. Integr. Neurosci. 9, 299-335, 2010). Although experimental studies have shown that during retinal development (prior to vision) the assembly of retinal cholinergic network connections are 'eliminated' in which the glutamatergic circuits involved in processing visual information are formed, such studies categorically do not rule out the possibility of the 'eliminated' ACh connections reconnecting after vision takes place in the mature retina.
His other research direction has been on the establishment of a new generation of neural networks, in particular, the biophysical foundations of neural network theory (as embodied in his book, Biophysical Neural Networks, Mary Ann Liebert, 2001). He was the first to reveal how microscopic-level biophysical properties (e.g., endogenous structures, ion channels; neuronal geometries) may be explicitly incorporated into an analytical formalism that predicts mesoscopic-level functionality. This approach has two major advantages: (1) avoids entirely the mathematical errors and uncertainties inevitable in iterative computational models that necessarily discretize time and space; (2) provides a framework for generating complete and exact solutions for network output enabling dynamical continuity to be reflected through spatiotemporal patterns as a field of influence for dynamic cognitive processes. This led him to consider more sophisticated artificial systems, like the 'cognitive' brain-computer interface (embodied in the book, Modeling in the Neurosciences: From Biological Systems to Neuromimetic Robotics, CRC Press, 2005) .
He has spent years in research at the interface of neuroscience and mathematical modeling with particular emphasis on developing neuronal cable models across scale. His recent collaboration with Dorian Aur (Stanford University, CA) focuses on the development of a cable theory for understanding the precise effect of protein polarization on membrane potential and on the excitability process. Cable theory finds its true foundations in Maxwell's theory of the electromagnetic field. In classical Nernst-Planck theory, the membrane has no structure and therefore, any attempt at combining the dynamics described by time-dependent Nernst-Planck equation for the spatial distribution of ionic concentration with cable theory is fortuitous. Ionic current flow based on Maxwellian approach is not physico-chemical and currents caused by concentration gradients are neglected. However, this new cable theory of protein polarization approximates electrodiffusion in physico-chemical systems, and is comparable with, yet resilient to the epistemological limitations inherent in the classical Hodgkin-Huxley system. It was Hodgkin who said: “electrodiffusion is like a flea hopping in the storm” meaning that ionic current flow in the presence of an electric field has no coherency compared with an action potential, but Hodgkin never envisaged ionic current flow as a propagating 'shock wave'. It was Huxley who in 1959 suggested that a subthreshold disturbance can be initiated by numerically solving Hodgkin-Huxley equations, but these subthreshold oscillations as envisaged by Andrew F. Huxley in the 1950s are not the only subthreshold responses. Subthreshold oscillations do not have a rapid leading-edge reminiscent of a 'shock wave' for decoding memory. Dr. Poznanski has embarked on the physics of memory and how it depends on: (i) changes in protein metabolism accompanying learning; (ii) memory trace formation (encoding) and storage (consolidation) in assemblies at the subcellular level through phosphorylation; and (iii) decoding through dendritic protein polarization by subthreshold signals at the subcellular level. The prevalent hypothesis is that memory is subserved by modulation in gene expression.
His other research direction has been on the establishment of a new generation of neural networks, in particular, the biophysical foundations of neural network theory (as embodied in his book, Biophysical Neural Networks, Mary Ann Liebert, 2001). He was the first to reveal how microscopic-level biophysical properties (e.g., endogenous structures, ion channels; neuronal geometries) may be explicitly incorporated into an analytical formalism that predicts mesoscopic-level functionality. This approach has two major advantages: (1) avoids entirely the mathematical errors and uncertainties inevitable in iterative computational models that necessarily discretize time and space; (2) provides a framework for generating complete and exact solutions for network output enabling dynamical continuity to be reflected through spatiotemporal patterns as a field of influence for dynamic cognitive processes. This led him to consider more sophisticated artificial systems, like the 'cognitive' brain-computer interface (embodied in the book, Modeling in the Neurosciences: From Biological Systems to Neuromimetic Robotics, CRC Press, 2005) .
He has spent years in research at the interface of neuroscience and mathematical modeling with particular emphasis on developing neuronal cable models across scale. His recent collaboration with Dorian Aur (Stanford University, CA) focuses on the development of a cable theory for understanding the precise effect of protein polarization on membrane potential and on the excitability process. Cable theory finds its true foundations in Maxwell's theory of the electromagnetic field. In classical Nernst-Planck theory, the membrane has no structure and therefore, any attempt at combining the dynamics described by time-dependent Nernst-Planck equation for the spatial distribution of ionic concentration with cable theory is fortuitous. Ionic current flow based on Maxwellian approach is not physico-chemical and currents caused by concentration gradients are neglected. However, this new cable theory of protein polarization approximates electrodiffusion in physico-chemical systems, and is comparable with, yet resilient to the epistemological limitations inherent in the classical Hodgkin-Huxley system. It was Hodgkin who said: “electrodiffusion is like a flea hopping in the storm” meaning that ionic current flow in the presence of an electric field has no coherency compared with an action potential, but Hodgkin never envisaged ionic current flow as a propagating 'shock wave'. It was Huxley who in 1959 suggested that a subthreshold disturbance can be initiated by numerically solving Hodgkin-Huxley equations, but these subthreshold oscillations as envisaged by Andrew F. Huxley in the 1950s are not the only subthreshold responses. Subthreshold oscillations do not have a rapid leading-edge reminiscent of a 'shock wave' for decoding memory. Dr. Poznanski has embarked on the physics of memory and how it depends on: (i) changes in protein metabolism accompanying learning; (ii) memory trace formation (encoding) and storage (consolidation) in assemblies at the subcellular level through phosphorylation; and (iii) decoding through dendritic protein polarization by subthreshold signals at the subcellular level. The prevalent hypothesis is that memory is subserved by modulation in gene expression.
 |
| FORTHCOMING BOOK MATHEMATICAL NEUROSCIENCE |
He has also embarked on a research collaboration with Stanislaw Brzychczy (AGH University of Science and Technology) on the development of a nonlinear analysis methods to better understand the intricate fallacies of methodological reductionism in neuroscience. One recent example of this research is the application of functional analysis in nonlinear analysis of the cable equation. The functional analysis proved that 'compartmental models of spiking neurons' are dynamically implausible representations of real neurons. This also includes 'multi-scale' computational approaches that claim to 'integrate' through the use of computational methodologies. In simple layman's terms, the approach of multi-scale computational modeling is a constructionist attempt at modeling the brain. The constructionist brain model postulates that as long as the algorithms work it suffices for the purpose of reverse engineering or building a brain model. However, there are severe problems with this approach. One limitation, is that multi-scale models are incapable of harnessing dynamical continuity across scale by means of discretization. As a result, the 'brain in a supercomputer' paradigm is wishful thinking.
Integration across scale requires continuity in the spatiotemporal dynamics, which is manifested through the electric field inside neurons and across synapses that results in a field of influence for augmenting cognitive processes in assemblies of networks. If dynamic continuity in spatiotemporal patterns is essential for integration then why is the brain not a syncytium? Cognition resides an in an infinite dimensional state-space and the brain is considered being dynamically unbounded. This means that synapses are essential for dynamic continuity to prevail across large networks of assemblies through the act of changing spatiotemporal patterns in addition to disrupting the spatiotemporal pattern in order for distinct associable representation to emerge. In such a continuum, the brain is modeled as an infinite-dimensional dynamical system. Infinite systems represent hierarchical levels in a way that will not delude the continuous dynamics of the neuronal systems across spatiotemporal scales. Neural organization across various multi-hierarchical levels yields complexities through infinite systems that are not disjoint sets (as in algebraic geometry or category theory) and therefore, different to ‘multi-scale computational modeling’ approaches that remain a constructionist attempt at modeling the brain.
Integration across scale requires continuity in the spatiotemporal dynamics, which is manifested through the electric field inside neurons and across synapses that results in a field of influence for augmenting cognitive processes in assemblies of networks. If dynamic continuity in spatiotemporal patterns is essential for integration then why is the brain not a syncytium? Cognition resides an in an infinite dimensional state-space and the brain is considered being dynamically unbounded. This means that synapses are essential for dynamic continuity to prevail across large networks of assemblies through the act of changing spatiotemporal patterns in addition to disrupting the spatiotemporal pattern in order for distinct associable representation to emerge. In such a continuum, the brain is modeled as an infinite-dimensional dynamical system. Infinite systems represent hierarchical levels in a way that will not delude the continuous dynamics of the neuronal systems across spatiotemporal scales. Neural organization across various multi-hierarchical levels yields complexities through infinite systems that are not disjoint sets (as in algebraic geometry or category theory) and therefore, different to ‘multi-scale computational modeling’ approaches that remain a constructionist attempt at modeling the brain.
L-R: Drs. Masumi Ishikawa, Roman Poznanski, and Chee Peng Lim. At far right,
Dr. Jonathan Chan. Discussion on higher brain function at a meeting in Kuala Lumpur.
He has advised several graduate students: (1) Hiroshi Yamamoto 1996 M.Sc. “Computer Simulation of Bipolar Cell Coupling in the Teleost Retina,” Faculty of Information Sciences, Toho University, Japan; (2) Tirad Almalahmeh 2009 Ph.D. "Directional Selectivity by Network of Starburst Amacrine Cells in Retina", Faculty of Computer Science and Information Technology, University of Malaya, Malaysia; (3) Seyed Maysam Torabi 2010 M.Sc. "Noisy Cables", Faculty of Computer Science and Information Technology, University of Malaya, Malaysia; (4) Chan Siow Cheng 2013 Ph.D "Neural Activity in a Morris-Lecar Population Density Model", Faculty of Engineering and Science, UTAR, Malaysia.
His most significant achievements include:
(i) First to pinpoint the locus underlying retinal direction selectivity in mammals, circa 1992. Through modeling starburst amacrine cells he was first to predict that direction selectivity is linked to their individual dendritic branches in a way that is still unknown with precision [5,6,13, 32].
(ii) First to show that conduction velocities in dendrites are nonconstant [20]. This theoretical result showed that sparse distribution of ionic channels will determine how information is processed differently in the dendrites of neurons as opposed to those in axons [21, 22].
(iii) First to find approximate analytical solutions to the Frankenhaeuser-Huxely equations [14, 38].
(iv) First to construct synaptically and gap-junctionally connected neural networks with ionic channels in dendritic cable structures. Such models have been applied to brain function through the development of large-scale brain cell assemblies [17, 23, 24, 27, 39, 40].
(v) First to introduce the conceptual idea that cognition is determined by how the distribution of endogenous proteins (e.g., ion channels) and synaptic inputs along the dendrites of neurons is integrated with the collective behavior of a large population of neurons grouped together as assemblies [15].
(vi) First to propose and develop nested neural network models for fMRI [12].
(vii) First to debunk the assumption of isopotentiality of small compartments (under 0.2λ) as a result of significant thermal noise [10].
(viii) First to propose a model-based framework for the development of a cognitive brain-computer interface [11].
(viii) First to propose a model-based framework for the development of a cognitive brain-computer interface [11].
(ix) Together with Brzycyhczy first to use functional analysis to prove how neural responses differ when continuous space is discretized in computational models [2,9].
(x) New theories of long-term memory away from synapses and dependent on ionic charge configurations [3,4].
(xi) First to elucidate the precise effect of protein polarization on membrane potential and on the excitability process through propagating subthreshold shock waves that are conducive of rich-logic requirements in dendrites underlying memory decoding [1].
(xi) First to elucidate the precise effect of protein polarization on membrane potential and on the excitability process through propagating subthreshold shock waves that are conducive of rich-logic requirements in dendrites underlying memory decoding [1].
(xii) First book in mathematical neuroscience that goes beyond unrealistic dynamics of physical abstractions into the realm of integrative modeling and dynamic continuity in spatiotemporal patterns [ to appear].
