What if "localist representation" is not in digital spikes, but is in electrical patterns? I see nothing wrong with connectionism, if redefined within a biological context where integration and dynamic continuity are part of the equation. This is what McClelland calls distributed processing afforded by a network style of representation. Sure, connectionism has relied in the past on computational modeling (machine learning, reinforcement learning, Bayesian models), but the dichotomy between connectionism and continuity is pivotal for extracting the correct theories of the brain. It is continuity that drives integration and if connectionism is used as a metaphor for dynamic continuity and integration then McClelland's assertion is correct. If there is no continuity then there is a set of discrete rules and symbols that govern rule-guided behaviour. Without dynamic continuity you have a 'binding' problem. The paper of Poggio in 1990 (ref.1) has considered and put emphasis on!
how the brain might work based on symbolic processing which is an extension of grandmother theories via look-up tables or representations in the brain. In such scenario, localist processing would be viable construct underlying the neural basis of cognition. Thus, theories of cognition residing in a single neuron have been applied from motor and sensory processing parts of the brain that have certain specificity, where a firing-rate analysis may be enough, but not necessarily in other parts of the brain, as for example, in MTL where the firing-rate alone doesn't discriminate well between presented images or behaviour. Taking the firing-rate idea from sensory and motor neurons, indicated that extending the model to MTL neurons and deriving the 'concept cell' is the 'grandmother' concept. What if the same 'grandmother' concept is carried over and over as an explanatory attribution for neural constructivism?
In Spivey's continuity of the mind thesis (ref.2), dynamics in single neurons give no reminiscence of cognition while trajectories in assemblies of neurons allows for a synthesis of dynamic continuity for higher-level cognition, without using symbolic processing. While localist representation if in digital spikes depends on symbolic processing for binding precepts, the case is not so, however, if localist representation is in electrical patterns, which includes synaptic interactions and electrical interactions inside and between neurons (non-synaptic). What if a localist representation is in electrical patterns? For example, retinal mechanisms of visual perception have been associated with spiking patterns in directionally selective ganglion cells, but presynaptically starburst amacrine cells have been shown to exhibit directionally biased electrical patterns. Therefore, spiking in neurons is insufficient for elucidating robust cognitive computation, and we may need to take !
into consideration computation by physical interactions, but the question that lingers is how does biological computation leading to cognitive computation generate meaning and concepts? The top-down approach, as for example, advocated by David Marr has been pivotal in fostering further developments of connectionist vistas beyond Rosenblatt's neurodynamics (ref. 3) without any success at answering this question. More recent examples coming from neuroimaging to understand cognition require a cortical homunculus (ref. 4), and deciphering brain connectivity based on the connectome is also unable to answer this question.
Fallacies about neural information processing capabilities of neurobiological systems are common, but understanding higher-level cognition rests on a different platform to information processing. Based on selectionism (ref.5), cognitive semantics that are fluidly engrained within the neural structure as a field of influence of dynamic continuity requires no information processing. This is where McClelland's views on distributed processing become entangled (ref. 6). A half century of biological computation based on information science has reached a cross-road. New brain theories (ref. 7) together with modern mathematics of the brain (ref.8) are starting to weed out these anomalies of the past. Rosen's relational biology extended by Chauvet (ref.9) to include concepts of hierarchical and functional integration is a guidepost for theoretical neuroscience as an alternative to constructionist attempts at building the brain.
References
1. Poggio, T. (1990) A theory of how the brain might work. Cold Spring Harbor Symp. Quant. Biol., 55, 899-910.
2. Spivey, M.J. (2007) The Continuity of Mind. Oxford University Press.
3. Rosenblatt, F. (1962) Principles of Neurodynamics. Spartan Books.
4. Dehaene, S. (1997) The Number Sense: How the Mind Creates Mathematics. Oxford University Press.
5. Edelman, G.M. (1987) Neural Darwinism. The Theory of Neuronal Group Selection. Basic Books.
6. Rogers, T. T. and McClelland, J. L. (2004). Semantic Cognition: A Parallel Distributed Processing Approach. MIT Press.
7. Aur, D. and Jog, M.S (2010) Neuroelectrodynamics. Understanding the Language of the Brain. IOS Press.
8. Brzychczy, S. and Poznanski, R.R. (2013) Mathematical Neuroscience. Academic Press.
9. Chauvet, G.A. (1996) Theoretical Systems in Biology: Hierarchical and Functional Integration. Pergamon Press.
--------
Roman R. Poznanski
Professor
Office: D218(Block D)2nd Floor
Universiti Tunku Abdul Rahman (UTAR)
Department of Physical & Mathematical Science
Faculty of Science
31900 Kampar, Perak
Malaysia
Office: (+605)468-8888 Ext. 2289
E-mail: roman@utar.edu.my
http://romanpoznanski.blogspot.com
and
Chief-Editor,
Journal of Integrative Neuroscience
http://www.worldscinet.com/jin/mkt/editorial.shtml
+++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++
This message is confidential and is solely intended for the
individual named. If you have received this e-mail in error
please notify the sender. Universiti Tunku Abdul Rahman
and/or the sender accepts no liability for any damages, claims
or losses caused by any viruses that may be transmitted by
this e-mail or any attachment. Any views or opinions
presented in this e-mail are solely those of the sender. No
contract or official order shall be concluded by means of this
e-mail.
+++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++
+++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++
This message is confidential and is solely intended for the
individual named. If you have received this e-mail in error
please notify the sender. Universiti Tunku Abdul Rahman
and/or the sender accepts no liability for any damages, claims
or losses caused by any viruses that may be transmitted by
this e-mail or any attachment. Any views or opinions
presented in this e-mail are solely those of the sender. No
contract or official order shall be concluded by means of this
e-mail.
+++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++
_______________________________________________
Comp-neuro mailing list
Comp-neuro@neuroinf.org
http://www.neuroinf.org/mailman/listinfo/comp-neuro
Very nice, inspiring!
ReplyDelete