I'd definitely suggest computational modelling as a supplement (but not replacement) to the above. Meeter & Murray 2004's TraceLink model is an example. 

Dear Gabriele,

            Many thanks for your reply and for the reference to the Bayesian Brain.  Reductionism is the golden rule of science, and I see you adhere to it in your admirable reasoning and research.  Reading one of your papers, I cannot avoid bragging a little about the fact that in probably the first-ever intracellular records from brain neurons, we were able to show a psychophysical power function—however compatible with logarithmic normality--in visual synaptic potentials (Fuster et al. ‑ Z. vergl. Physiol. 49:605‑622, 1965).  In any event, I can fully understand why you miss in our models the inclusive treatment of elementary neural structures and functions that would seem at least as relevant to cognition as our networks.  Indeed, these networks would not survive without the trophic support of glia, astrocytes and blood vessels, among other things.

            But, alas, when it comes to cognition and cognitive neuroscience, reductionism hits a ground floor of forbiddingly irreducible facts.  The reason, to put it simply, is that memory and knowledge, the veritable units of cognition, are defined by relations between percepts, acts, events, neurons and neuronal groups; just as words are defined by relations between letters and propositions between words. This fact strictly constrains the epistemology and methodology of cognitive neuroscience.  Hence the necessity of a network paradigm of cognition (of course you know which one I prefer!).  Any attempt to elucidate cognitive causality under its relational ground floor leads to a Laplacean demon of imponderable variables, in space and time.  It is like trying to understand the written letter of a friend by studying the chemistry of the ink.

            Cheers,  Joaquín        

 Dear Dr. Fuster,

I very much admire your work. Thank you for the interesting question. The following comments are based on work by Len Koziol and myself. I would offer an additional “tool in the toolbox” in our efforts to identify and research reliable neural network data. This tool is the incorporation of phylogenetic constraint into our models of network dynamics.

Data from all branches of neuroscience support the concept of ongoing “real time” competition for action selection (Cisek & Kalaska, 2010; Cisek & Pastor-Bernier, 2014). These data do not support a “perceive-think-respond” framework. Instead, action selection occurs through a process of elimination, based on rule-based input signals from the prefrontal cortex; reward prediction signals through cortico-striatal loops; and error correction signals processed by way of cortico-cerebellar and cerebellar-basal ganglia circuitry (Caligiore, Pezzulo, Miall, & Baldassarre, 2013; Cisek & Kalaska, 2010; Koziol, 2014; Koziol & Budding, 2009). Because decision making occurs during activity, embodiment emphasizes the interactive choices that occur in real time within the environment (Cisek & Kalaska, 2010; Koziol & Lutz, 2013; Lepora & Pezzulo, 2015).

The “perceive-think-respond” model functions much too slowly for this real time interaction. The primary evolutionary pressure required adjustment to a dynamically changing and unpredictable environment (Sterelny, 2003). Across the eons, there must have been considerable adaptive pressure to retain a system that was fast and accurate. Quickness and precision increases survival opportunity since it exploits the features of the environment we can count on while conserving resources (Cisek & Pastor-Bernier, 2014). Further progression up the phylogenetic chain reveals the predictive nature of adaptive functioning; flexible anticipatory control based on prediction of rewarding outcome enhances the ability to survive and thrive in a hostile world (Pezzulo, 2008, 2011).

Phylogenetic constraint is a useful tool to better understand neural networks. The evolution of brain structures is remarkably conserved across all mammals over 500 million years of evolution  (Grillner, Robertson, & Stephenson-Jones, 2013). Across time, functional changes occur through an expansion of neural circuitry within existing brain structures. These changes occur slowly, not by huge “leaps and bounds” that are made within the gene pool (Smaers & Soligo, 2013).  In fact, the basic architecture of the brain has been present since the earliest animals walked on land (Stephenson-Jones, Ericsson, Robertson, & Grillner, 2012). This fact should constrain any model of brain-behavior relationships; a model of neural networks must “work” within the constraints of phylogeny by incorporating the functional purposes of existing brain structures in earlier species.

The goals of the vertebrate brain were achieved by implicit thinking coupled with action, alternating with episodes of adjustment that required more “cerebral” conscious cognitive awareness. The evolutionary expansion of the neocortex reflects its ability to perceive, organize, and manage increasingly complex environmental stimuli. The human brain contains no structures that are unique to humans. As we climb the phylogenetic chain, the cerebellum and the basal ganglia continue to perform their fundamental functions, but on more complex environmental stimuli/data (Shine & Shine, 2014; Smaers & Soligo, 2013).

By incorporating phylogenetic constraint, parameters can be placed on our theories of neural networks. Neuropsychological constructs, including models of working memory, have developed based on the idea that thought directs behavior. However, the vertebrate brain did not evolve for the purpose of thinking. The vertebrate brain evolved to meet the needs of interactive behavior.  This is critically important because we do not live in a static environment. Instead, the environment is dynamically changing. Researchers have begun to incorporate these ideas to enhance our understanding of neural networks. Two excellent papers incorporating these ideas include Haber’s examination of the impact of cortico-basal ganglia loops on neural circuits of reward and decision making (Haber, 2011) and a recent consensus paper of contributors examining the cerebellum, basal ganglia, and cerebral cortex as an integrated system (Caligiore et al., 2016).

Good Hunting! 

Arthur

Caligiore, D., Pezzulo, G., Baldassarre, G., Bostan, A. C., Strick, P. L., Doya, K., . . . Herreros, I. (2016). Consensus Paper: Towards a Systems-Level View of Cerebellar Function: the Interplay Between Cerebellum, Basal Ganglia, and Cortex. Cerebellum. doi:10.1007/s12311-016-0763-3

Caligiore, D., Pezzulo, G., Miall, R. C., & Baldassarre, G. (2013). The contribution of brain sub-cortical loops in the expression and acquisition of action understanding abilities. Neuroscience and Biobehavioral Reviews, 37(10 Pt 2), 2504-2515. doi:10.1016/j.neubiorev.2013.07.016

Cisek, P., & Kalaska, J. F. (2010). Neural mechanisms for interacting with a world full of action choices. Annual Review of Neuroscience, 33, 269-298. doi:10.1146/annurev.neuro.051508.135409

Cisek, P., & Pastor-Bernier, A. (2014). On the challenges and mechanisms of embodied decisions. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 369(1655). doi:10.1098/rstb.2013.0479

Grillner, S., Robertson, B., & Stephenson-Jones, M. (2013). The evolutionary origin of the vertebrate basal ganglia and its role in action-selection. Journal of Physiology. doi:10.1113/jphysiol.2012.246660

Haber, S. N. (2011). Neural circuits of reward and decision making: Integrative networks across corticobasal ganglia loops. In R. B. Mars, J. Sallet, M. Rushmore, & N. Yeung (Eds.), Neural basis of motivational and cognitive control (pp. 21-36). Cambridge, Mass.: MIT Press.

Koziol, L. F. (2014). The myth of executive functioning: Missing elements in conceptualization, evaluation, and assessment. New York, NY: Springer.

Koziol, L. F., & Budding, D. E. (2009). Subcortical structures and cognition : implications for neuropsychological assessment. New York: Springer.

Koziol, L. F., & Lutz, J. T. (2013). From movement to thought: the development of executive function. Appl Neuropsychol Child, 2(2), 104-115. doi:10.1080/21622965.2013.748386

Lepora, N. F., & Pezzulo, G. (2015). Embodied choice: how action influences perceptual decision making. PLoS Computational Biology, 11(4), e1004110. doi:10.1371/journal.pcbi.1004110

Pezzulo, G. (2008). Coordinating with the Future: The Anticipatory Nature of Representation. Minds Mach., 18(2), 179-225. doi:10.1007/s11023-008-9095-5

Pezzulo, G. (2011). Grounding Procedural and Declarative Knowledge in Sensorimotor Anticipation. Mind & Language, 26(1), 78-114. doi:10.1111/j.1468-0017.2010.01411.x

Shine, J. M., & Shine, R. (2014). Delegation to automaticity: the driving force for cognitive evolution? Frontiers in Neuroscience, 8, 90. doi:10.3389/fnins.2014.00090

Smaers, J. B., & Soligo, C. (2013). Brain reorganization, not relative brain size, primarily characterizes anthropoid brain evolution. Proceedings: Biological Sciences, 280(1759), 20130269. doi:10.1098/rspb.2013.0269

Stephenson-Jones, M., Ericsson, J., Robertson, B., & Grillner, S. (2012). Evolution of the basal ganglia: dual-output pathways conserved throughout vertebrate phylogeny. Journal of Comparative Neurology, 520(13), 2957-2973. doi:10.1002/cne.23087

Sterelny, K. (2003). Thought in a hostile world : the evolution of human cognition. Malden, MA: Blackwell.

Dear Fuster,

First of all, I would like to say I am a great fellow of your work, that inspire some of mine. Second, thank you for sharing your thoughts with the scientific community. I think this channel could be a good path to promote scientific discussion, and provide a wonderful tool to young and novel scientists as me to get in contact with great scientists as you. 

Regarding your post, I have to admit the challenges that scientific community is facing to measure brain dynamic and structure with an unique neuroimaging technic/algorithm with a high temporal and spatial resolution at the same time. 

For instance, fMRI and fNIRS is an indirect measured (BOLD) of brain activity, because they just measure the cerebral vascular tissue consumption, and not the Cerebral Metabolic Oxygen Rate (CMOR2). Besides, the increased oxygen transported to the brain area typically exceeds the local CMOR21,2. So, it is even more complicated, because we "fail" to have an accurate measured of brain structure, dynamics and activity in a single, reliable and valid technic/algorithm.

However, there are promising technics and methods that can be developed to overcome these amazing challenges:

1. Best in Temporal Resolution: MAGNETOENCEPHALOGRAPHY

2. Best Spatial Resolution: fMRI/DTI

3. Best Forthcoming Neuroimaging Technic: Optogenetic (as soon as we can apply this technic on human, we could sort out the methodological limitations we are facing today).

4. An algorithm that I like to measure the dynamics of brain networks is dynamic functional network connectivity (d-FNC).

5. One important advance in neuroscience would be to develop the measured by fNIRS of the cytochrome c oxidase (Cyt-Ox). It is the last enzyme in the respiratory electron transport chain of mitochondria or bacteria located in the mitochondrial or bacterial membrane. So, with this enzyme we could have a directed measured of CMOR2, and fNIRS enables its detection.

6. Nanotechnology and Neuroimaging  http://www.ncbi.nlm.nih.gov/pubmed/23514423 

Summarizing, I think we are living in amazing time that would be revolutionized in the next years by the synergies of basic science (neuroscience, materials, energy, quantum) + information technology + engineering + market3.

Best,

References

1.Fox, P.T., Raichle, M.E., Mintun, M.A. & Dence, C. (1988). Nonoxidative glucose

consumption during focal physiologic neural activity. Science, Vol. 241, No. 4864,

(July 1988), pp. 462-4, ISSN 1095-9203.

2. Lin, A.L., Fox, P. T., Yang, Y., Lu, H., Tan, L.H. & Gao, J.H. (2008). Evaluation of MRI models in the measurement of CMRO2 and its relationship with CBF. Magnetic resonance in medical sciences, Vol. 60, No. 2, (August 2008), pp. 380-9, ISSN 1880-2206.

3. Alivisatos et al., Nanotools for neuroscience and brain activity mapping. (2013). ACS Nano, 7(3), 1850-66. doi: 10.1021/nn4012847.

Dear Joaquin, 

You raised a very important question, hard to answer though. The coming debate between Tony and Moritz may be interesting for a further question. Even if we characterize every spatial and temporal feature of the network, are we there to explain cognition?

CNS 2016 Debate

Connectomics: ‘Missing link’ for understanding neural computation, or ‘minutiae of implementation’?

Monday, April 4, 5:00 – 6:30 pm, Sutton Center

A debate between Anthony Movshon, NYU and Moritz Helmstaedter, Max-Planck-Institute.

Moderator: David Poeppel, Max-Planck-Institute & NYU

Limited seating, reservation required.

Anthony Movshon, NYU – Biologists are taught the basic principle that form follows function. The idea of connectomics goes further, to propose that form determines function. From this it would follow that knowing the form of neural circuits in detail would by itself allow us to deduce their function. The reality is different. The form of neural circuits undoubtedly constrains their function, but a knowledge of form alone is neither necessary nor sufficient to understand brain function. Neuroanatomy has always been one of the central tools of neuroscience. But except in very special cases like the retina, where the behavior of a circuit is well understood, anatomy alone tells us little of function. Rather, functional models that try to capture the purpose of neural computation rather than dwelling on the minutiae of its implementation offer a better path to understanding the brain.

Moritz Helmstaedter, Max-Planck-Institute – Since brains drive rich time-varying behaviors, the focus of scientific investigation has always been the dynamics of neurons, synapses, neuronal ensembles. Decades later we have a myriad of theoretical proposals how brains could compute, especially in the cerebral cortex of mammals. But few of these models are unequivocally refutable by functional experiments. Could the lack of knowledge about neuronal circuit structure, connectomes, be the missing link? Connectomic experiments are more efficient than functional ones in finding rare but relevant computations of a neuronal ensemble. Connectomes are required to explain computations in the mammalian retina and fly optical system. The degree to which connectomes shape computations in the mammalian cerebral cortex will remain a matter of scientific debate until finally, cerebral cortex connectomes will be measured.

It seems to me that what is needed are theoretical models of the minimal brain mechanisms that are specified in sufficient structural and dynamic detail to enable us to specify their patterns of activation under natural conditions. Then we need to demonstrate that these theoretical models can successfully predict relevant mental phenomena on the basis of analogical correspondences between the models' performance and mental phenomena. See, for example, The Cognitive Brain, "Space, self, and the theater of consciousness", and "A Foundation for the Scientific Study of Consciousness" on my RG page.

I dont know why nobody listed the method/technique that affords the best brain correlates of conscious processing: calcium imaging with fluorescence microscopy, with or without optogenetics (see an example in the link below). 

http://www.nature.com/nprot/journal/v11/n3/full/nprot.2016.021.html#videos

Unfortunately, simple correlative information does not expose the underlying neuronal mechanisms that generate our mental experiences. I have suggested this bridging principle between the the biophysical activity of relevant brain mechanisms and phenomenal content:

*For any instance of conscious content, there is a corresponding analog in the biophysical state of the brain.*

The scientific problem is to specify the putative brain mechanisms that have the competence to generate such analogs of conscious content. The neuronal mechanisms that I have proposed have been put to empirical tests and have been able to predict/postdict many previously unexplained conscious experiences. Are there alternative specified models that can perform as well or better?

Dr Trehub, your principle is truly fascinating. I take that principle may mean that, for every instance of conscious content, (in the neocortex) that there is a corresponding analog in the brainstem. That truly integrates conscious and unconscious neural activities and makes a lot of sense. In fact, I propose that the brainstem may mirror, in analogous networks, the pattern of circuitry found in the neocortex, helping it to do the simple analysis needed to enable repairs and maintenance on that region. Just as emotion centers in the forebrain are critical for decision-making, because they give "weight" to various choices of action, there are probably centers in the brainstem that play an "emotional" role, too, helping to make basic physiological responses in the rest of the body more fluid, predictable, and faster.

Dr. Fuster,

In answer to your original question, my own preference for methods and models is multifractal analysis to address the cascade-like (e.g., "context-sensitive") nature of goal-directed behavior. This approach aligns wit Don Ingber and Michael Turvey's (independent but consonant) views of organic tissue as proceeding according to tensegrity principles at both the cellular and perceptual levels of organization. To my knowledge multifractal analysis is the gold standard method for modeling cascade processes, and tensegrity structures are generally understood to follow multifractal organization.

I appreciate that my approach is not popular or well received, even by current followers of this question of yours, but you asked, and since this is RG, I have contributed my two cents. Somehow, I manage to be a cognitive scientist with my admittedly strange views.

Best wishes,

Damian

Dear Damian,

Multifractal analysis of what? 

Dear Alfredo,

Multifractal analysis of fine-grained, densely-sampled, movement variability, usually at the point(s) of contact between organism and ambient energy distribution. My organism of choice is the human, so the foot or the hand or the eyes or the head end up offering highly textured variability for analysis. I then draw on predictor sets including standard task parameters and also multifractal statistics drawn from movement variability  during the task, and I use those items as predictors in a regression model of trial-by-trial cognitive/perceptual judgments or responses.

Thank you for your kind interest. I hope this description helps. I am happy to answer more questions as they arise.

Best wishes,

Damian

It is not clear how we get from the biological neuronal network level to the complex computations of the conscious and subconscious brain. While there has been great discoveries, such as receptive fields, Hebbian learning, STDP, cell assemblies etc which offer many insights, it is still not clear at all how some brain function is "compiled" into the language of biological neural networks.

It is my understanding that the `cognitive sciences' adopt  David Marr's Tri-level hypothesis when attempting to explain some brain function. That is, we break the problem into three subproblems :

The problem you describe is somewhere in between level 2 and 3, lets call it level 2.5. Subproblem 2.5 his how one gets from the functional/computational description of how some brain function is executed (level 2) to the actual biological neurons implementing this function (level 3). This level can be seen as the "Compiler level" and I believe is the concern of theoretical neuroscientists. We try to describe the "machine code" of the brain, what are the building blocks of computation.  I think to answer your question fully, more research and insight is needed on this level.

Well, I my incipient (if you want) opinion, we dont need to bet everything on technology to explain the life. In the past, we answered many questions with strong ideas and very simple techniques. We can perfect  some of the same techniques and keep increasing our knowledge. In this case, I don't see why not the use of selective brain damage by esterotaxic surgery/lasser, and combination with optogenetics/pharmacology, behavioural tasks, microdialysis and tractography can answer any question related with structure and dynamics of cognitive networks.

On the other hand, any level of knowledge has elements which cannot been explained with the sum of the whole elements of the inferior level. In other words, cognitive networks cannot explain behaviour itself.

I think that technology and fragmentation of science is not the way to solve problems of the society related with the researcher. Thus, there is not a more suitable method (seen as the sum of techniques used to answer a scientific question) than another. Instead, there are more effective ideas to follow than others.

Only an opinion.

Best regards. Bruno

Hello,

First seems to me not to persecute prevent or avoïd by reducing ( in a bad sens ) people minding researching and imaginating. So will it be good in my opinion to favor a "by all ( people means and A.B. ( Artificial Beings ) ) research  and self questionning to overflow your present question/idea by making as if we go radically beyond behind and ahead not forgetting the past.

Every people and things count ( are important ) in this challenge.

In french it sounds like " ma-thématique, ta-thématique, sa-thématique " that is an "ourthematical" question.

So let me suggest we commonly organize and concern " Young* " people in philosophical worldly wisdom and self concerned or by consciousness approaches

of, so that our unconscious thousand faces participate as lightened by a Pierre Buser book in my méditations

Of course any theoretical " logical " and practical domain is to be involved in by artists, inventors, professionnals and scholars as they feel and experiment deeply these questions and the world. " Love has his reasons motives grounds that { what ? } don't know about " . Because it is not so correct but at least puzzling to say : " L'amour a ses raisons que la raison ne connaît pas "...

Must go to eat,

Regards

* " Il n'y a pas d'âge pour être jeune ". Autocitation.

There is no age to be young.

Dear Yixuan,

         You ask: even if we characterize every spatial and temporal feature of the network, can we explain cognition?  My answer is yes, at least in principle.  But, of course, we must properly define both the network and cognition.  Here I will summarize my definitions and working assumptions on both:

 1. The large-scale cognitive network (cognit for short) is the basic neural unit of memory and knowledge (i.e., semantic memory).  One such cognit is a distributed array of cortical neuron assemblies representing discrete sensory and motor events that have been associated by experience.  The associations that form a cognit (new or enhanced synaptic connections) result from spatial and temporal relationships of contiguity, coincidence or near-coincidence between the experienced events.  In the cortex of the individual brain there is an infinite number of overlapping and interconnected cognits of great many sizes, some connected to or embedded in others; all of them are plastic, modifiable by experience, with variable synaptic strength between their neurons. One neuron or group of neurons, practically anywhere in the cortex, can be part of many cognits, thus many memories or items of knowledge. The entire conglomerate of cognits forms an immense web whose individual components are defined in cortical space by differences in distribution and synaptic strength.  The all-encompassing web or “connectome,” even if its description were structurally complete and perfect, would be functionally meaningless and useless (and a waste of money).

2.  A cognitive function consists in the timely and orderly activation of cognits in the course of behavior or reasoning.  From that activation (and the accompanying reciprocal inhibitions) emerge all cognitive functions that guide sensors and effectors in the perception-action cycle (attention, perception, memory, language and intelligence).  There is no need for networks specialized in each of those functions.  Every cognitive function results from that timely and orderly activation of cognits. What is most important is that their structure and functions are relational (associative) by definition and irreducible to their elementary constituents.  The cognitive “code” is essentially a relational code (like that of Gestalt).   

3.  Indeed, as Tony and Moritz remark or imply, the structure of a given cognitive network does not reveal its physiological function in any cognitive function.  To reemphasize my point #2, I should add that, as in primary sensory and motor systems, that function is defined and determined by a spatial-temporal pattern of excitation and inhibition within and between cognits.  The key to understanding its neural operations is to use good operational definitions of cognitive function and to combine methods to obtain enough spatial and temporal resolution to reveal those patterns (fMRI, NIRS, EEG, EMG, single- or multi-unit analysis, etc.).  But remember, all analysis, even single-neuron analysis, must be conducted with reference to a particular cognit or set of cognits.  Because of its relational character and definition, the cognit is the methodological floor of cognition.  As I have said elsewhere in present context, any reductionist attempt to penetrate that floor downward (e.g., toward synapses, channels or molecules) is like trying to understand the written manuscript by studying the chemistry of the ink.

4.  In he light of these refutable ideas (Hayek used to say, “without a theory the facts are silent”), the computational task looks less daunting.  After all, the transactions within and between cognits are essentially analog and probabilistic (e.g., Bayesian), not digital and deterministic, because they occur within a system of widely ranging and variable synaptic strengths. 

         Without those principles, no one will ever discover any conceivable neurocognitive causality by staring at the cold connectome, however exact the latter may be.  In my opinion, he analogy with the genome is spurious beyond any level of credibility.

         All this may sound implausible to some and, I humbly admit, may be wrong.  But it is the way I see it after many years of toiling with these matters.

With kind regards, Joaquín   

J.M. Fuster - Cortex and memory: emergence of a new paradigm.  Journal of Cognitive Neuroscience,  21:2047-2072, 2009.

Dear Roman,

            I agree wholeheartedly with your statement, “a continuous hierarchical field allows for integrating physiological functions across spatial scales.”  I believe my cognit paradigm conforms quite well to that description:

 J.M. Fuster - Cortex and memory: emergence of a new paradigm. Journal of Cognitive Neuroscience, 21:2047-2072, 2009.

             The late Walter Freeman was right.  I have a letter from him commenting on the commonalities between his field approach and my cognit.

            Trying to understand cognition in the connectome is like trying to understand the economy of a nation by studying its roadmap.

 Cheers,

 Joaquín