@Article{Gallistel2017,
  author       = {Gallistel, C R},
  journal      = {Trends in cognitive sciences},
  title        = {The Coding Question.},
  year         = {2017},
  issn         = {1879-307X},
  pages        = {498--508},
  volume       = {21},
  abstract     = {Recent electrophysiological results imply that the duration of the stimulus onset asynchrony in eyeblink conditioning is encoded by a mechanism intrinsic to the cerebellar Purkinje cell. This raises the general question - how is quantitative information (durations, distances, rates, probabilities, amounts, etc.) transmitted by spike trains and encoded into engrams? The usual assumption is that information is transmitted by firing rates. However, rate codes are energetically inefficient and computationally awkward. A combinatorial code is more plausible. If the engram consists of altered synaptic conductances (the usual assumption), then we must ask how numbers may be written to synapses. It is much easier to formulate a coding hypothesis if the engram is realized by a cell-intrinsic molecular mechanism.},
  comment      = {From Gallistel (2017). The coding Question. Trends in Cognitive Sciences, 21, 498-508.

# Desiderata of a memory mechanism

*	Carry large amounts of information forward in time in a computationally accessible form. 
*	High thermodynamic stability (last a long time) 
*	Low or negligible energy costs (not drain the batteries). 
*	Realizable in a maximally compact volume, using as few elements as possible (store a great many bits in very little space). 
*	Capable of representing a huge range of quantities (store very small and very big numbers).
*	Addressable on a short timescale – that is, it is quickly readable on the basis of location in memory, without reference to informational content.

# Problems with synaptic plasticity as a memory mechanism
*	Synaptic machinery, as commonly conceived is designed for the transmission and modulation of transient signals.
*	Rate codes are highly inefficient and provide less than 1bit/spike. To see this, consider the contrast between the hash mark code for twenty (‘||||||||||||||||||||’; that is, 20 tally strokes), as contrasted with the digital code (20) or the binary code (10100).
*	Rate codes are extremely costly energetically. Cortical spike conduction together with transmitter release and reuptake requires the hydrolysis of 108 ATP molecules per spike (Laughlin, S.B. (2004) The implications of metabolic energy requirements for the representation of information in neurons. In The Cognitive Neurosciences III (Gazzaniga, M.S., ed.), pp. 187–196, MIT Press. ) 
•	Spike trains transmit more information than encoded by the rate:
	-	Spike trains convey between three and seven bits of information per spike (Gerstner, W. et al. (2014). Neuronal Dynamics: From Single Neurons to Networks and Models of Cognition, Cambridge University Press ; Rieke, F. et al. (1997). Spikes: Exploring the Neural Code, MIT Press).
	-	Information about quantitative variation in an environmental variable is conveyed by the interspike intervals; reordering the intervals changes the message (Lazar, A.A. and Zhou, Y. (2014) Reconstructing natural visual scenes from spike times. Proc. IEEE 102, 1500–1519). The precise timing of the spikes has also recently been shown to determine a motor result (Srivastava, K.H. et al. (2016) Motor control by precisely timed spike patterns. Proc. Natl. Acad. Sci. U. S. A. 114, 1171–1176).
	-	Single axons in hippocampal neuropil make ‘two or more synaptic contacts on the same dendrites, having shared histories of presynaptic and postsynaptic activity. The spine heads and neck diameters ... were nearly identical in size’ (Bartol, T.M. et al. (2015) Nanoconnectomic upper bound on the variability of synaptic plasticity. eLife 4, e10778). There was a minimum of 26 distinguishable synaptic strengths (head sizes). They suggest that a synapse can therefore store log2(26) = 4.7 bits of information.
	-	Individual cerebellar Purkinje cells time and remember the interval between the onset of an artificial conditioned stimulus (CS) and the onset of an artificial unconditioned stimulus (US) (Johansson, F. et al. (2014) Memory trace and timing mechanism localized to cerebellar Purkinje cells. Proc. Natl. Acad. Sci. U. S. A. 111, 14930–14934). As a result, there must be some molecular memory mechanism.

Candidate molecular memory mechanisms
*	Polynucleotides
	-	DNA carries information at remarkable volumetric densities (more bits per cubic micron than in the best currently available memory chips), with negligible energetic cost (the code can be read thousands of years after the animal has died), and with an address-addressable logic-based reading mechanism [7,30].	
	-	Adding one nucleotide to a polynucleotide string adds 2 bits of information, and it requires the hydrolysis of only one ATP
	-	DNA cannot be used as a memory mechanism, but cells are full of extranuclear microRNAs. As they are important in intracellular signal cascades, they are readable, and might be used for cognitive cascades as well
*	Isomerization
-		Opsin have to thermodynamically stable isomers, and influence signaling
*	“These advantages – compactness, energy efficiency, and ability to adapt and match – all suggest the principle compute with chemistry. It is cheaper.” (Sterling, P. and Laughlin, S.B. (2015) Principles of Neural Design, MIT Press, p. 124).},
  country      = {England},
  created      = {2017-05-19},
  doi          = {10.1016/j.tics.2017.04.012},
  file         = {:/Users/endress/Articles/The Coding Question.pdf:PDF},
  issn-linking = {1364-6613},
  issue        = {7},
  nlm-id       = {9708669},
  pmid         = {28522379},
  pubmodel     = {Print-Electronic},
  revised      = {2017-06-17},
}

{"_id":"ax7oryvpWoYgCvK4G","bibbaseid":"gallistel-thecodingquestion-2017","author_short":["Gallistel, C R"],"bibdata":{"bibtype":"article","type":"article","author":[{"propositions":[],"lastnames":["Gallistel"],"firstnames":["C","R"],"suffixes":[]}],"journal":"Trends in cognitive sciences","title":"The Coding Question.","year":"2017","issn":"1879-307X","pages":"498–508","volume":"21","abstract":"Recent electrophysiological results imply that the duration of the stimulus onset asynchrony in eyeblink conditioning is encoded by a mechanism intrinsic to the cerebellar Purkinje cell. This raises the general question - how is quantitative information (durations, distances, rates, probabilities, amounts, etc.) transmitted by spike trains and encoded into engrams? The usual assumption is that information is transmitted by firing rates. However, rate codes are energetically inefficient and computationally awkward. A combinatorial code is more plausible. If the engram consists of altered synaptic conductances (the usual assumption), then we must ask how numbers may be written to synapses. It is much easier to formulate a coding hypothesis if the engram is realized by a cell-intrinsic molecular mechanism.","comment":"From Gallistel (2017). The coding Question. Trends in Cognitive Sciences, 21, 498-508. # Desiderata of a memory mechanism * Carry large amounts of information forward in time in a computationally accessible form. * High thermodynamic stability (last a long time) * Low or negligible energy costs (not drain the batteries). * Realizable in a maximally compact volume, using as few elements as possible (store a great many bits in very little space). * Capable of representing a huge range of quantities (store very small and very big numbers). * Addressable on a short timescale – that is, it is quickly readable on the basis of location in memory, without reference to informational content. # Problems with synaptic plasticity as a memory mechanism * Synaptic machinery, as commonly conceived is designed for the transmission and modulation of transient signals. * Rate codes are highly inefficient and provide less than 1bit/spike. To see this, consider the contrast between the hash mark code for twenty (‘||||||||||||||||||||’; that is, 20 tally strokes), as contrasted with the digital code (20) or the binary code (10100). * Rate codes are extremely costly energetically. Cortical spike conduction together with transmitter release and reuptake requires the hydrolysis of 108 ATP molecules per spike (Laughlin, S.B. (2004) The implications of metabolic energy requirements for the representation of information in neurons. In The Cognitive Neurosciences III (Gazzaniga, M.S., ed.), pp. 187–196, MIT Press. ) • Spike trains transmit more information than encoded by the rate: - Spike trains convey between three and seven bits of information per spike (Gerstner, W. et al. (2014). Neuronal Dynamics: From Single Neurons to Networks and Models of Cognition, Cambridge University Press ; Rieke, F. et al. (1997). Spikes: Exploring the Neural Code, MIT Press). - Information about quantitative variation in an environmental variable is conveyed by the interspike intervals; reordering the intervals changes the message (Lazar, A.A. and Zhou, Y. (2014) Reconstructing natural visual scenes from spike times. Proc. IEEE 102, 1500–1519). The precise timing of the spikes has also recently been shown to determine a motor result (Srivastava, K.H. et al. (2016) Motor control by precisely timed spike patterns. Proc. Natl. Acad. Sci. U. S. A. 114, 1171–1176). - Single axons in hippocampal neuropil make ‘two or more synaptic contacts on the same dendrites, having shared histories of presynaptic and postsynaptic activity. The spine heads and neck diameters ... were nearly identical in size’ (Bartol, T.M. et al. (2015) Nanoconnectomic upper bound on the variability of synaptic plasticity. eLife 4, e10778). There was a minimum of 26 distinguishable synaptic strengths (head sizes). They suggest that a synapse can therefore store log2(26) = 4.7 bits of information. - Individual cerebellar Purkinje cells time and remember the interval between the onset of an artificial conditioned stimulus (CS) and the onset of an artificial unconditioned stimulus (US) (Johansson, F. et al. (2014) Memory trace and timing mechanism localized to cerebellar Purkinje cells. Proc. Natl. Acad. Sci. U. S. A. 111, 14930–14934). As a result, there must be some molecular memory mechanism. Candidate molecular memory mechanisms * Polynucleotides - DNA carries information at remarkable volumetric densities (more bits per cubic micron than in the best currently available memory chips), with negligible energetic cost (the code can be read thousands of years after the animal has died), and with an address-addressable logic-based reading mechanism [7,30]. - Adding one nucleotide to a polynucleotide string adds 2 bits of information, and it requires the hydrolysis of only one ATP - DNA cannot be used as a memory mechanism, but cells are full of extranuclear microRNAs. As they are important in intracellular signal cascades, they are readable, and might be used for cognitive cascades as well * Isomerization - Opsin have to thermodynamically stable isomers, and influence signaling * “These advantages – compactness, energy efficiency, and ability to adapt and match – all suggest the principle compute with chemistry. It is cheaper.” (Sterling, P. and Laughlin, S.B. (2015) Principles of Neural Design, MIT Press, p. 124).","country":"England","created":"2017-05-19","doi":"10.1016/j.tics.2017.04.012","file":":/Users/endress/Articles/The Coding Question.pdf:PDF","issn-linking":"1364-6613","issue":"7","nlm-id":"9708669","pmid":"28522379","pubmodel":"Print-Electronic","revised":"2017-06-17","bibtex":"@Article{Gallistel2017,\n author = {Gallistel, C R},\n journal = {Trends in cognitive sciences},\n title = {The Coding Question.},\n year = {2017},\n issn = {1879-307X},\n pages = {498--508},\n volume = {21},\n abstract = {Recent electrophysiological results imply that the duration of the stimulus onset asynchrony in eyeblink conditioning is encoded by a mechanism intrinsic to the cerebellar Purkinje cell. This raises the general question - how is quantitative information (durations, distances, rates, probabilities, amounts, etc.) transmitted by spike trains and encoded into engrams? The usual assumption is that information is transmitted by firing rates. However, rate codes are energetically inefficient and computationally awkward. A combinatorial code is more plausible. If the engram consists of altered synaptic conductances (the usual assumption), then we must ask how numbers may be written to synapses. It is much easier to formulate a coding hypothesis if the engram is realized by a cell-intrinsic molecular mechanism.},\n comment = {From Gallistel (2017). The coding Question. Trends in Cognitive Sciences, 21, 498-508.\n\n# Desiderata of a memory mechanism\n\n*\tCarry large amounts of information forward in time in a computationally accessible form. \n*\tHigh thermodynamic stability (last a long time) \n*\tLow or negligible energy costs (not drain the batteries). \n*\tRealizable in a maximally compact volume, using as few elements as possible (store a great many bits in very little space). \n*\tCapable of representing a huge range of quantities (store very small and very big numbers).\n*\tAddressable on a short timescale – that is, it is quickly readable on the basis of location in memory, without reference to informational content.\n\n# Problems with synaptic plasticity as a memory mechanism\n*\tSynaptic machinery, as commonly conceived is designed for the transmission and modulation of transient signals.\n*\tRate codes are highly inefficient and provide less than 1bit/spike. To see this, consider the contrast between the hash mark code for twenty (‘||||||||||||||||||||’; that is, 20 tally strokes), as contrasted with the digital code (20) or the binary code (10100).\n*\tRate codes are extremely costly energetically. Cortical spike conduction together with transmitter release and reuptake requires the hydrolysis of 108 ATP molecules per spike (Laughlin, S.B. (2004) The implications of metabolic energy requirements for the representation of information in neurons. In The Cognitive Neurosciences III (Gazzaniga, M.S., ed.), pp. 187–196, MIT Press. ) \n•\tSpike trains transmit more information than encoded by the rate:\n\t-\tSpike trains convey between three and seven bits of information per spike (Gerstner, W. et al. (2014). Neuronal Dynamics: From Single Neurons to Networks and Models of Cognition, Cambridge University Press ; Rieke, F. et al. (1997). Spikes: Exploring the Neural Code, MIT Press).\n\t-\tInformation about quantitative variation in an environmental variable is conveyed by the interspike intervals; reordering the intervals changes the message (Lazar, A.A. and Zhou, Y. (2014) Reconstructing natural visual scenes from spike times. Proc. IEEE 102, 1500–1519). The precise timing of the spikes has also recently been shown to determine a motor result (Srivastava, K.H. et al. (2016) Motor control by precisely timed spike patterns. Proc. Natl. Acad. Sci. U. S. A. 114, 1171–1176).\n\t-\tSingle axons in hippocampal neuropil make ‘two or more synaptic contacts on the same dendrites, having shared histories of presynaptic and postsynaptic activity. The spine heads and neck diameters ... were nearly identical in size’ (Bartol, T.M. et al. (2015) Nanoconnectomic upper bound on the variability of synaptic plasticity. eLife 4, e10778). There was a minimum of 26 distinguishable synaptic strengths (head sizes). They suggest that a synapse can therefore store log2(26) = 4.7 bits of information.\n\t-\tIndividual cerebellar Purkinje cells time and remember the interval between the onset of an artificial conditioned stimulus (CS) and the onset of an artificial unconditioned stimulus (US) (Johansson, F. et al. (2014) Memory trace and timing mechanism localized to cerebellar Purkinje cells. Proc. Natl. Acad. Sci. U. S. A. 111, 14930–14934). As a result, there must be some molecular memory mechanism.\n\nCandidate molecular memory mechanisms\n*\tPolynucleotides\n\t-\tDNA carries information at remarkable volumetric densities (more bits per cubic micron than in the best currently available memory chips), with negligible energetic cost (the code can be read thousands of years after the animal has died), and with an address-addressable logic-based reading mechanism [7,30].\t\n\t-\tAdding one nucleotide to a polynucleotide string adds 2 bits of information, and it requires the hydrolysis of only one ATP\n\t-\tDNA cannot be used as a memory mechanism, but cells are full of extranuclear microRNAs. As they are important in intracellular signal cascades, they are readable, and might be used for cognitive cascades as well\n*\tIsomerization\n-\t\tOpsin have to thermodynamically stable isomers, and influence signaling\n*\t“These advantages – compactness, energy efficiency, and ability to adapt and match – all suggest the principle compute with chemistry. It is cheaper.” (Sterling, P. and Laughlin, S.B. (2015) Principles of Neural Design, MIT Press, p. 124).},\n country = {England},\n created = {2017-05-19},\n doi = {10.1016/j.tics.2017.04.012},\n file = {:/Users/endress/Articles/The Coding Question.pdf:PDF},\n issn-linking = {1364-6613},\n issue = {7},\n nlm-id = {9708669},\n pmid = {28522379},\n pubmodel = {Print-Electronic},\n revised = {2017-06-17},\n}\n\n","author_short":["Gallistel, C R"],"key":"Gallistel2017","id":"Gallistel2017","bibbaseid":"gallistel-thecodingquestion-2017","role":"author","urls":{},"metadata":{"authorlinks":{}}},"bibtype":"article","biburl":"https://endress.org/publications/ansgar.bib","dataSources":["xPGxHAeh3vZpx4yyE","TXa55dQbNoWnaGmMq"],"keywords":[],"search_terms":["coding","question","gallistel"],"title":"The Coding Question.","year":2017}