X-Message-Number: 10418 Date: Sat, 12 Sep 1998 13:13:40 -0700 From: Jeff Davis <> Subject: Holographic memory in ye olde braine panne For Thomas Donaldson and the Cryonet, Thomas, you wrote me about the scrambled salamander brain experiments, wherein you said: "Years ago an interesting experiment was done with salamanders: their cortex was removed, chopped into several pieces, and then replaced. The salamander was comatose for some time afterwards, but then recovered its memory. (Cf A Hershkowitz et al "The acquisition of dark avoidance by transplantation of the forebrain of trained newts", BRAIN RESEARCH 48(1972) 366ff; and P Pietsch, "Scrambled salamander brains: a test of holographic theories of neural program storage" ANATOMICAL RECORD 172(1972) 383)." I followed it up, and found the scrambled salamander brain experiments documented in greater depth than one could hope for, and quite entertainingly, by Paul Pietsch in his book "Shufflebrain", which can be found at: http://www.indiana.edu/~pietsch/home.html#sample ---------------------------------------------------------------------- Looking for more recent work (the "Shufflebrain" work dates back to 1972), I found on the internet: Comparison between Karl Pribram's "Holographic Brain Theory" and more conventional models of neuronal computation By Jeff Prideaux Virginia Commonwealth University --------------------------------------------------------------------- Most recently I found the following, which I enclose in its entirety. It starts out as an optical computing article but transitions rapidly into a neuroholograpy article. As is often the case, the last paragraph is rich: Optical Processing & Computing Biomorphic neural networks are new links for information processing In 1985 researchers from Caltech and the University of Pennsylvania (Penn) drew attention to a valuable link between optics and the neural paradigm for information processing. In a seminal paper that appeared in Applied Optics 1, they demonstrated the advantages offered by optics for implementing the large number of interconnections between neurons in an artificial neural network that functioned as associative memory. Since that time, the use of optical interconnects in hardware implementation of neural net architectures has developed into an accepted methodology. Now Nabil Farhat and his students at Penn are finding another potentially useful link between optics and neurocomputing, this time heavily involving nonlinear dynamical systems, that could prove beneficial for all three fields. The new link is a conceptual connection between holography and neurocomputing which they have coined "neuroholography." The idea of a "holographic model of the brain" is not new. It has been around since the early sixties. What is new now, is that Farhat's concept of neuroholography emerges directly from a biomorphic (biology-like) model neuron conceived to mimic the behavior of cortical neurons2. (The cortex is that part of the brain that carries out all higher-level brain functions like cognition and complex motor control.) Their biomorphic-model neuron exhibits phase-locking between periodic firing patterns it produces and a periodic potential, that forms under certain "input" conditions, at the neuron's "hillock." The hillock is that part of a biological neuron where action potentials (nerve impulses or spikes) are initiated to be broadcast from the neuron via its axon and axon branches to other neurons. When the incident spike wavefront (ISW)-which is the aggregate of all trains of nerve impulses (spike trains) impinging on the neuron from other neurons in a prescribed time window (the input)-is coherent or partially coherent, i.e. the trains of incident nerve impulses from other neuron are correlated or partially correlated, the neuron's dendrites act to form a periodic potential at the hillock reflecting the coherence in the ISW. Frequency of Driving Signal, fs [Hz] The appearance of such "periodic activation" dramatically alters the neuron's behavior as an information processing element by enabling it to exhibit a variety of complex periodic firing patterns that can be phase-locked to the periodic activation or can be chaotic depending on the nature of the periodic activation (e.g. its amplitude and frequency if it is cosinusoidal). Under these circumstances the neuron acts as a detector and encoder of coherence in its input. When however the ISW is incoherent, the activation potential becomes steady or slowly varying in time and the neuron's behavior reverts to the usual sigmoidal dependence of firing frequency on activation without any phase-locking involved. Their model neuron is able thus to produce a wide range of phase-locked, nonphase-locked, and chaotic firing modalities and to bifurcate between them depending on the nature of the ISW and the associated activation potential. For this reason they are calling it the "bifurcation model neuron." It is interesting to note that behavior very similar to that of the bifurcation model neuron has been observed in several biophysical experiments performed with periodically stimulated biological neurons and excitable membranes (see reference 3 for example). Numerical simulations of a network of bifurcation neurons, show that neurons in the network tend to usually phase-lock their firing and this results in the emergence of coherence in the ISWs of the individual neurons. The feasibility of something similar occurring in cortical networks of the brain is supported by the relatively recent discovery of long-range oscillations and correlations in the spiking activity and local-field potentials (extracellular potentials) in the visual cortices of cat and monkey (see for example reference 4). This discovery has aroused intense interest in understanding its underlying neuronal mechanism and in exploring possible applications that seek to emulate higher-level brain functions such as scene segmentation, feature-binding, cognition, complex motor control and general development of dynamical computing (computing with all three types of attractors: point, periodic, and chaotic exhibited by high-dimensional dynamical systems like the brain. Bifurcation model neurons and networks fit well with an oscillation theory of cortical networks which hypothesizes that timing, phase-locking, synchronicity, and chaos might underlie higher-level brain functions. At first glance such hypothesis may induce skepticism. How could timing, phase-locking, and synchronicity occur in cortical networks when biological neurons are known to be noisy processing elements? Noise in cortical neurons, as in other neurons, originates in the synapses, the very sites of communication between neurons. Synaptic noise has two causes. One is the probabilistic nature of "exocytosis" that describes events leading to the release of neurotransmitter molecules into the synaptic cap. These events occur with probability of less than one, meaning that not every nerve impulse reaching the terminal point of an axonal branch forming a synapse junction with another neuron, succeeds in the release of neurotransmitters into the synaptic gap separating a presynaptic neuron from a postsynaptic neuron. Neurotransmitter molecules released into the synaptic gap activate ionic channels in the postsynaptic membrane allowing ions to flow into the postsynaptic neuron altering thereby its potential. This, incidently, is the basic mechanism for electrochemical communication employed by neurons. The second cause of synaptic noise is the stochastic nature of the opening and closing (gating) of activated ionic channels. Recent work carried out by Farhat and Hernandez5, shows bifurcation neuron networks exhibit a phenomenon they call clustering that neutralizes the noise caused by the probabilistic nature of exocytosis. In reference 5 they show that an externally driven network of bifurcation neurons exhibits under certain input conditions clustering, synchronicity, and phase-locking. The processing elements (neurons) in their network, group themselves into clusters of unequal size. Neurons within a cluster fire in unison, i.e. neurons within a cluster are synchronized. Different clusters have distinct period-m firing patterns but all such patterns are phase-locked i.e. there is a fixed temporal relation between them. (Period-m firing is cyclic firing in which the neuron fires repeatedly a pattern of m spikes that are not necessarily equally spaced). What this means for the probabilistic exocytosis issue, is that neurons in such a network would be receiving at their synapses identical spike trains from all other neurons belonging to the same cluster. The ISWs of the neurons are now not only coherent but also highly redundant because of clustering. The redundancy means that if exocytosis at one synapse does not occur when a nerve impulse arrives, there is more than ample chance that the same spike in the identical spike train arriving at another synapse will. This redundancy suppresses the noise caused by probabilistic exocytosis and the effectiveness of suppression improves rapidly with redundancy, i.e. with the size of the clusters. Farhat's group is now focusing their attention on noise caused by the stochastic gating of activated ion channels: Ongoing modeling and simulation work is showing, amazingly enough, that channel noise helps rather than hinders the conversion of coherence in the ISW into a periodic activation potential at the neuron's hillock because of Stochastic Resonance (SR). SR is a mechanism by which noise in a bistable or thresholding process such as believed to exist in ion channels or in excitable biological membranes, can amplify a weak periodic signal rather than degrade it6. Withstanding further scrutiny, this latter result would validate the bifurcation neuron concept which ignored synaptic noise from the start, and would go a long way towards explaining the nature of the "neuronal code," i.e. the way cortical neurons encode information they receive from other neurons while participating in carrying out higher-level brain functions. Furthermore, the availability of a periodic activation potential at each neuron can be viewed as a local reference signal serving to phase-lock the periodic spike pattern produced by the neuron. This, together with the possibility that the efficiency with which synapses transmit information can be modified by the degree of correlation between the neuron's spike train and the spike train received at each synapses from other neurons, i.e. correlations between the pre- and post-synaptic spike trains, furnish the basic ingredients needed for the formulation of a self-consistent neuroholography concept. This hopefully will be useful for both optical information processing, neurocomputing, and for the optical implementation of adaptive nonlinear dynamical systems. References 1. N. Farhat, D. Psaltis, A. Prata, and E. Paek, "Optical implementation of the Hopfield Model," Appl. Opt., vol. 24, pp. 1469-1475, 1985. 2. N. Farhat, S-Y Lin and M. Eldefrawy, "Complexity and chaotic dynamics in a spiking neuron embodiment," in Adaptive Computing, S. Chen and J. Caulfield, (Eds.), vol. CR55, pp. 77-88, SPIE, Bellingham, Wash., (1994). 3. K. Aihara and G. Matsumoto, "Chaotic oscillations and bifurcation in squid giant axon," in Chaos, A.V. Holden (Ed.), Princeton Univ. Press, Princeton, N.J., pp. 257-269, (1986). 4. R. Eckhorn, et. al., "Coherent oscillations: A mechanism of feature linking in the visual cortex," Biol. Cybern., vol. 60, pp. 121-130, (1988). 5. N. Farhat and E. Del Moral Hernandez, "Recurrent neural networks with recursive processing elements," presented at SPIE '96, Denver, August (1996). 6. J. Collins, et. al., "Stochastic resonance without tuning," Nature, vol. 376, pp. 126-135, July (1995). SPIE Web Home | OE Reports Feb. =A9 1997 SPIE - The International Society for Optical Engineering ---------------------------------------------------------------------- Holographic storage and retrieval is an analog process. The brain is an analog processor. Holographic storage provides the ultimate in redundancy by the "equipotential" distribution of the "data set", and consequently the ultimate in durability/survivability. Survivability is THE fundamental evolutionary fitness criteria. I read somewhere--unfortunately I can't find it again--that cells from tissues frozen with cryoprotectant and thawed, and then cultured, showed survival rates between 50% and 95%. Higher order structures might be unable to recover viability, but individual cells did, in high percentages. (Could someone help me to relocate the source for this?) How is this relevant for cryonics? High cell survival rates and an information storage paradigm characterized by high inherent durability, bodes well for the prospect of memory/personality preservation. Does this "prove" anything? Nope. Should anyone ease up in their efforts to achieve reversible suspension, damage-free suspension, or cell-repair technologies? I don't think so Tim! Nevertheless, I find it quite heartening. Best, Jeff Davis "Everything's hard till you know how to do it." Ray Charles =09 Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=10418