neurodudes

Computational neuroscience is a field where many successful researchers have a strong physics background. So far, the physics approach has provided a strong foundation from which to understand the brain. Recently, however, the influence of a computer science perspective has become more prominent. How can we understand the different perspectives that these disciplines bring to the field? Can we observe the influence of physics methodologies on the modern study of the brain? And if so, what is the consequence of our understanding of the brain through the lens of physics versus the lens of computer science?

One consequence may be the way that computational neuroscience models time in the brain. The study of physics generally conceptualizes time as continuous. Time is something to be plotted on the x-axis of a graph where some other quantity of interest is plotted on the y-axis.

In computer science, on the other hand, real time is rarely conceptualized explicitly. Computer scientists do not plot quantities against time unless they are profiling software for performance purposes, and even then, time is more generally thought of as number of operations. Thinking about operations generally leads computer scientists to think about time as discrete events.

I posit that the distinction between continuous and discrete time creates a foundational difference between the physics approach and the computer science approach to understanding how the brain works. Due to the discrete time conceptualization, computer scientists are more comfortable explaining the function of brain systems in terms of chains of events with definite beginnings and definite ends. Physicists, on the other hand, are more comfortable explaining the brain in terms of dynamics, which do not require definite beginnings or definite ends. Computer scientists care more what the consequence of an event is in the brain, whereas physicists are more concerned with an concise account of the dynamics of what is occurring.

This divide is visible in the distinct modeling approaches of neurons that derive from these two disciplines. The canonical neuronal model contributed by the physics philosophy is the multi-compartmental conductance based (Hodgkin-Huxley like) model. This model is concerned with matching waveforms of current and voltage traces with those that are measured in real neurons. This model helps us to understand how changes of the properties of excitable membranes over time result in changes of neuronal behavior over time. The computational complexity of these models is thought to prevent more than a few hundred neurons modeled in this way from being analyzed concurrently.

Alternatively, the canonical neuronal model contributed by the computer science philosophy is the integrate-and-fire neuron. This model does away with modeling conductances explicitly as functions of time and simply performs a weighted sum of its inputs at each time step. Here a time step is a discrete event whose duration is a parameter of the model. The simplicity of this model allows large networks to be constructed, which are useful for modeling systems of many thousands of neurons.

The physics approach provides insight into the activity of single cells and small networks, whereas the computer science approach provides insight into the activity of large networks. Neither approach is optimal and neither approach provides all the tools that are necessary to truly understand the brain. As these two perspectives are better understood, the field of computational neuroscience can benefit from finding creative ways to merge these two conceptions of time into models that capture both small scale and large scale neuronal activity.

In conclusion, I have demonstrated that what begins as a division between discrete and continuous time amounts to a divide between a bottom-up and a top-down approach. Furthermore, I have shown that understanding the relative contributions of different sciences to computational neuroscience is important for understanding the paradigms that pervade the field.

Transcranial magnetic stimulation (TMS) is a popular technology for stimulating human cortical neurons, due to its safety, noninvasiveness, and efficacy. A TMS device is just a little coil of wire, through which 10,000 Amps of current is cranked during a period of only a few hundred microseconds; the resultant rapidly-changing magnetic field induces eddy currents in the brain. Depending on the protocol used, TMS can drive/inhibit a region of cortex corresponding to roughly a cubic centimeter or two, and is being explored for the treatment of depression, the reduction of auditory hallucinations during schizophrenia, and the alleviation of tinnitus and migraines. Thousands of papers on medicine and psychology have been written using this tool.

Yet the device itself is expensive and rare — they can run from $20,000 to $50,000 or even more, despite the fact that they are, in essence, a coil, a switch, a bank of capacitors, and a power supply. Much of the art lies in making the devices safe and fail-proof. Is it possible to hack/engineer a system that is safe, fault-tolerant, efficacious, and inexpensive? And furthermore, can we facilitate a community that will devise such devices, and share information about protocols and approaches to brain hacking?

This past August at Foo Camp, a hackers’ conference in Northern California, a group of people got together and set out to do just that. We are designing a safe, noninvasive, modular, and “open source” brain stimulator that will open up the field of circuit modulation to a wider audience. Members of the group include therapists and mental health professionals, engineers, programmers, and others interested in either the development of such devices, or the sharing of information on this front. Key to the design is safety — we want to make sure that the devices we create are as safe as devices on the market. Also, all the information is released under the Creative Commons “Attribution and Sharealike” license. This is a new model for “open source” medical device development — which may move it beyond the domain of simply creating “cool toys,” and to creating real devices.

You can find out more information, or contribute to the project, or learn from the project, at
http://transcenmentalism.org/OpenStim/

-Ed

It seems Markram is again back to getting some interesting results. Recently a new discovery from the Brain Mind Institute of the EPFL shows that the brain adapts to new experience by unleashing a burst of new neuronal connections, and only the fittest survive. The research further shows that this process of creation, testing, and reconfiguring of brain circuits takes place on a scale of just hours, suggesting that the brain is evolving considerably even during the course of a single day.

The paper can be found Here.

This past May, the Almaden Research center, part of IBM research, invited some provocative speakers on the topic of “Cognitive Computing” to come and speak. Since IBM recently invested a lot of money into understanding the brain with the Blue Brain project, it seems like this meeting was a way to figure out the next step along this path.

Powerpoint presentations and videos of the event are available online.

From the synopsis:

The 2006 Almaden Institute will focus on the theme of “Cognitive Computing” and will examine scientific and technological issues around the quest to understand how the human brain works. We will examine approaches to understanding cognition that unify neurological, biological, psychological, mathematical, computational, and information-theoretic insights. We focus on the search for global, top-down theories of cognition that are consistent with known bottom-up, neurobiological facts and serve to explain a broad range of observed cognitive phenomena. The ultimate goal is to understand how and when can we mechanize cognition.

Confirmed speakers include Toby Berger (Cornell), Gerald Edelman (The Neurosciences Institute), Joaquin Fuster (UCLA), Jeff Hawkins (Palm/Numenta), Robert Hecht-Nielsen (UCSD), Christof Koch (CalTech), Henry Markram (EPFL/BlueBrain), V. S. Ramachandran (UCSD), John Searle (UC Berkeley) and Leslie Valiant (Harvard). Confirmed panelists include: James Albus (NIST), Theodore Berger (USC), Kwabena Boahen (Stanford), Ralph Linsker (IBM), and Jerry Swartz (The Swartz Foundation).

From the Apr 20 issue of Neuron: Integrative Properties of Radial Oblique Dendrites in Hippocampal CA1 Pyramidal Neurons (or, for those who want just the N&V’s summary: Dendritic Enlightenment: Using Patterned Two-Photon Uncaging to Reveal the Secrets of the Brain’s Smallest Dendrites)

The technology is essentially high-speed two photon uncaging of glutamate, but the authors have used it here to create “realistic” patterns of dendritic input in an attempt to see just how dendritic arithmetic works. Although I haven’t read the paper closely, they claim to work out the spatiotemporal parameters underlying dendritic spike generation for pyramidal neurons.

A related methodology paper from a recent J. Neurophys. also uses fast acousto-optic deflectors and two-photon but for imaging purposes. It’s more descriptive about the setup and techniques for those interested in doing this type of work.

The lab of David McCormick at Yale has released a paper that shows neurons operating in both analog and digital modes simultaneously.

From an article about the finding:

“McCormick’s group demonstrated that the analog signal present in the cell body also propagates down the axon and influences synaptic transmission onto other neurons. As the voltage on the sending cell becomes more positive, the amplitude of the subsequent transmission to the receiving cell, mediated by an action potential, is enhanced. This means that the waveform generated in the receiving neuron is not just determined by the digital pattern of action potentials generated, but also by the analog waveform occurring in the sending neuron.”

McCormick is a big name in the field. Is it time to start creating a new field of artificial neural networks that has both analog and digital modes?

How do crickets know when they are chirping?

These questions appear to be answered with the discovery of a motor interneuron in the cricket that is resposible for “corallary discharge” or forwarding neural signals from motor systems to sensory systems. By inhibiting auditory neurons during chirping, the animal can “counter the expected, self-generated sensory feedback”.

Over at the synapse blog, it is pointed out that the cerebellum may have this function in vertebrates.

Henry Markram, the director of the IBM-sponsored Blue Brain Project has written an article in the latest issue of Nature Reviews: Neuroscience that provides the most technical details about the project to date.

From the article:

“The three-dimensional neurons are then imported into BlueBuilder, a circuit builder that loads neurons into their layers according to a ‘recipe’ of neuron numbers and proportions. A collision detection algorithm is run to determine the structural positioning of all axo-dendritic touches, and neurons are jittered and spun until the structural touches match experimentally derived statistics. [...] Probabilities of connectivity between different types of neuron are used to determine which neurons are connected, and all axo-dendritic touches are converted into synaptic connections. The manner in which the axons map onto the dendrites between specific anatomical classes and the distribution of synapses received by a class of neurons are used to verify and fine-tune the biological accuracy of the synaptic mapping between neurons. It is therefore possible to place 10–50 million synapses in accurate three-dimensional space, distributed on the detailed three-dimensional morphology of each neuron.”

–Stephen

This project was announced several months ago, but I didn’t see a post here so I thought I would add it.

The project, dubbed “Blue Brain“, represents a team up between Henry Markram, (who co-authored the chapter on the neocortex in the acclaimed reference The Synaptic Organization of the Brain), and IBM’s Blue Gene super computer.

From the New Scientist article: For over a decade Markram and his colleagues have been building a database of the neural architecture of the neocortex, the largest and most complex part of mammalian brains.

Very thin slices of mouse brain were kept alive under a microscope and probed electrically before being stained to reveal the synaptic, or nerve, connections. “We have the largest database in the world of single neurons that have been recorded and stained,” says Markram.

–Stephen

A neat paper from 1999 that I saw.

This post is mostly identical to the corresponding page on NeuroWiki. You may wish to read/discuss it there instead.

This paper presents a leaky integrate and fire model which adapts to the average rate of incoming spikes. The model has two modes, integration mode and coincidence detection mode.

Specifically, the model is an extension to the Morris-Lecar model in which maximal conductance changes over time according to a simple calcium dynamics model. This change allows the neuron to adapt to different average rates of input.

Interestingly, directly after you change the average rate of presynaptic input, the neuron may be transiently pushed into a different mode. Specifically, if you bump up the level of activity, the neuron is pushed towards coincidence detection mode, and if you suddenly decrease the level of activity, the neuron is pushed towards integration mode.

The paper also contains a bunch of citations to introduce the spike rate vs spike timing code debate.

I’m not quite sure if the neuron’s “default” mode (that is, when the average rate of incoming spikes is fixed) is always the integrator mode, or if you can change that by changing the parameters. I don’t think there’s any hysteresis (that is, I don’t think that the neuron gets “stuck” in one mode or another; I think it only switches modes transiently and then returns to its default slowly over time as it adapts), but I’m not sure. Anyone care to clarify?

Michele Giugliano, Marco Bove, Massimo Grattarola. Activity-Driven Computational Strategies of a Dynamically Regulated Integrate-and-Fire Model Neuron. Journal of Computational Neuroscience 7(3): 247-254 (1999)