100 Days Blog

Day 019 - Emergence

Submitted by Sam on 8 June, 2011 - 22:35

The induced procedural behaviour of optogenetically engineered fruit flies and the invariantly instinctive behaviour of the Sphex wasp are examples of how complex behaviour can be encoded programmatically, with fixed stimuli producing fixed responses. These fixed pathways can be seen as discrete 'programs' in the brain, a subset of the overall neural network devoted to a particular function. Through the interaction of many such discrete components, extremely complex behaviour can emerge; though the components themselves are not intelligent, their combined interaction can produce intelligent behaviour. This is the theory of emergence, and it is the animating force behind attempts to create artificially intelligent neural networks.

The classic example of high-level artificially emergent behaviour is Reynolds' Boids, a computer model developed in 1986 to simulate coordinated animal motion. Through the interaction of three basic rules, each simulated boid is steered into a flock. The simple rules which produce this behaviour are:

  • Separation: steer to avoid crowding local flockmates
  • Alignment: steer towards the average heading of local flockmates
  • Cohesion: steer to move toward the average position (centre of mass) of local flockmates

Unexpected behaviours such as flocks dividing and then reuniting to avoid obstacles emerge from the interaction of these rules, showing once more how simple instructions can produce organized behaviour.

Day 018 - Natural programmatic behaviour

Submitted by Sam on 7 June, 2011 - 23:05

Gero Miesenböeck produced a programmatic behavioural response in fruit flies by selectively re-wiring their neural circuitry, establishing an evidential link between the electrical states of individual neurons and specific behaviour. His findings offer micro-scale support to our understanding of instinctive behaviour, in particular the abundance of examples of fixed action paths that can be found in nature.

Fixed action patterns are hard-wired, invariant sequences which run until completion in response to a sensory stimulus, known as a sign stimulus or releaser. Whilst very similar to reflex actions, which are clearly instinctive, fixed action paths can be processed by the brain. Reflex actions, conversely, do not go through the brain, but instead trace a reflex arc which involves no processing by the neural networks of the brain at all.

A favourite example of an easily triggered fixed action pattern is found in some species of Sphex digger wasp. Adult females lay their eggs in a burrow, leaving them to hatch on their own. Before leaving them, the mother wasp captures and paralyzes a cricket, dragging it to the edge of the burrow. She then goes in alone, apparently checking for intruders before emerging to drag the cricket inside, leaving it close to her eggs to provide food for them when they hatch. This extremely complex sequence of behaviour, whilst seeming to show signs of planning and forethought, is actually a completely automated routine, a fixed action path which gives the illusion of intelligence. If an experimenter moves the cricket very slightly away from the burrow whilst she is inside, she will re-emerge, drag the cricket back to the threshold, and re-enter the burrow once more. This cycle of behaviour has been induced dozens of times, illustrating that the cricket is merely following a strict procedure in response to an initial stimulus, in this case placing the cricket in front of the burrow.

When one of Miesenböeck's graduate students, Susana Lima, engineered a fruit fly so that just two of its brain cells expressed a light-activated pore causing the fly to invariably take off in response to a flash of light, she had clearly found the cells which were responsible for initiating one of the fly's fixed action patterns. She was able to initiate the reflex using an internal neuronal stimulus, rather than the external releaser that is so readily identifiable in the Sphex wasp example.

Day 017 - Optogenetic control of the brain

Submitted by Sam on 7 June, 2011 - 00:14

So far we have seen projects committed to making computers work like brains, encoding the immense complexity of the biological world into the procedural logic of the digital. Waynflete Professor of Physiology at Oxford Gero Miesenböeck has approached the brain from the reverse perspective, developing genetic strategies to control the brain itself like a computer.

Miesenböeck is the founder of the emerging field of optogenetics, which exploits genetic engineering to create targeted cells in living animals which can be switched from one state to another by a flash of light. Optogenetic remote control of the brain is the subject of his TED talk, below.

Miesenböeck formulates the familiar mantra of projects like The Human Brain Project as "if we could record the activity of our neurons, we would understand the brain", but modulates the recipe through his own work as the more practical “if we could control the activity of some neurons, we would learn much about the brain”. Rather than taking apart the brain and rebuilding it to see how it works, Miesenböeck's research focuses on controlling the brain in order to understand it.

In order to control the electrical impulses of the brain, which in turn control both behaviour (as we have known since Galvani's experiments on frogs in the 18th century) and thought, Miesenböeck re-engineered selected neural elements to become responsive to light, a non-invasive, diffuse signal. Through his optogenetic engineering he turned selected nerve cells into receivers that allowed him to control their function through a flash of light onto the brain. Using this technique, Miesenböeck successfully simulated an unpleasant memory in a fruit fly, causing it to 'remember' to walk away from a certain odour every time it encountered it. Through repeated experimentation with differentially activated nerve cells, Miesenböeck was able to narrow this behaviour down to twelve specific neurons in the fly's brain, which he identifies as the brain's 'critic', defining policy for the fly's 'actors', such as the circuits which control leg movement.

Miesenböeck's optogenetical control of behaviour provides arresting experimental evidence for the physical, mechanistic understanding of the mind that connectionist models are attempting to simulate in silicon.

Day 016 - Neuroinformatics

Submitted by Sam on 6 June, 2011 - 00:10

Projects like The Blue Brain Project and The HBP require a consolidated and consistent body of neuroscientific data against which to calibrate their virtual brain models, and so standardizing and aggregating the findings from international, interdisciplinary research is consequently vital to the success of computational neuroscience. The International Neuroinformatics Coordinating Facility, or INCF, was established in 2005 in response to this growing need, aiming to develop and steward a highly structured international informatics network of standardized neuroscientific databases and models. Neuroinformatics is therefore a discipline which encompasses all levels and scales of neuroscience (from genes to behaviour), facilitating the sharing and accessibility of data, tools and techniques amongst the neuroscientific community.

The task of collecting and integrating the field's data has become increasingly difficult as it has gained in popularity and diversity, with many thousands of publications representing many different countries, disciplines and methodologies released every year. The principle goal of the INCF is to systematize this immense wealth of material into databases which can be efficiently analyzed and modelled by researchers worldwide. The INCF also promotes standards and common frameworks in order to address infrastructural issues within the international neuroscientific community itself, which arise from the inherent difficulties of different labs working with multiple experimental techniques. By developing programmes for data-sharing, data-reuse, modelling, reporting and recording, the INCF benefits not only neuroscience as a whole, but also specifically supports efforts to build digital simulations of the brain by providing an addressable body of knowledge against which to test and verify virtualized behaviour.

Day 015 - The Human Brain Project: another brain simulation

Submitted by Sam on 5 June, 2011 - 00:51

The Human Brain Project, or HBP, is described as the successor to the Blue Brain Project. The HBP does what it says on the tin, explicitly aiming to simulate the complete human brain with as much biological detail as is technically possible. The HBP will pursue this goal by building a phased sequence of biophysical, phenomenological and abstract models at various resolutions and scales (from model brain regions, mesocircuits, to whole brain systems, or macrocircuits), starting with animals like rats and mice, progressing to cats and monkeys and finally to humans.

Like the Blue Brain Project, the HBP will be constantly comparing the properties of its model circuits against existing experimental data, ensuring wherever possible that their virtual brains corroborate with the 60,000 pages of neuroscientific research that are published every year. Unlike the Blue Brain Project, the project's computer engineers will be running multi-level simulations to reduce the overall computing requirements by simulating highly active neurons in great detail whilst modelling less active neurons with lower accuracy.

The scope of the HBP's roadmap reads like a transhumanist manifesto. Here are some of the juicy bits:

  • The project intends to develop its facility, tools and skill-sets to be able to model the brain of any animal, at any stage of its development, in any state of health or with any specific disease.
  • The later stages of the project make provision for computing requirements thousands of times more powerful than any in existence today.
  • The high-level mathematical theories of brain function will be able to combine with the technologies needed to create realistic simulations to create a new class of brain-like hardware devices and computer architectures.
  • These new information technologies will have the brain's capabilities to repair itself, to learn, and to be creative, utilizing neuromorphic circuits derived from the circuitry of the brain.
  • A brain simulation will provide insights into the basic causes of neurological diseases such as Alzheimer's, Parkinson's, autism and depression.
  • A virtual model will present a new platform for testing drugs, facilitating the creation of targeted drugs with fewer side-effects, and reducing our reliance on animal testing.

Day 014 - The Blue Brain Project

Submitted by Sam on 3 June, 2011 - 21:28

In his excellent 2008 interview with Seed Magazine, Henry Markram gave a fascinating tour of the Blue Brain Project, which pursues the humble mission of reverse-engineering the brain. Having looked at the principles of connectionism already, the reasoning behind the project should sound familiar. Here's what Markram had to say: “There is nothing inherently mysterious about the mind or anything it makes...Consciousness is just a massive amount of information being exchanged by trillions of brain cells. If you can precisely model that information, then I don’t know why you wouldn’t be able to generate a conscious mind.”

Since 2005 the Blue Brain Project has systematically constructed the foundations for a complete virtual human brain. The first step the project took was to simulate the neurons in a two-week old rat's cortical column, the smallest functional unit of the neocortex, which is believed to be responsible for high-level functions such as conscious thought and sensory perception. As basic units of the cortex, each cortical column seems to be assigned a discrete function – in a rat, for instance, a column is devoted to each whisker. A rat's cortical column is the size of a pinhead, has 10,000 neurons in 50 different types, joined by 108 synapses. Human cortical columns are very similar, but contain around 60,000 neurons.

The team automated the process of analysing the genetic expression of real rat neurons using a patch-clamp robot, and used the data from their experiments to create a precise map of the ion channels in the rat's neurons. They fed this information back into their Blue Brain simulation, which runs on an IBM BlueGene/L supercomputer. Throughout the process, the researchers were able to continually test their model against real neural activity in a real live rat, fine-tuning their simulation against the performance of the real thing.

Using this data the team were able to assemble a three-dimensional model that precisely simulated the neocortical column of their two-week-old rat. When stimulated with the same sort of electrical stimulation that a newborn rat would actually experience, the model reacted just like a real neural circuit, with clusters of connected neurons firing in close synchrony, spontaneously wiring themselves into meaningful patterns.

The generated model's behaviour corroborates results observed from years of neuroscientific experiments, and will be able to serve as a building block for a full-scale simulation of the human brain. It seems that all that is holding the project back now is computational power. The team estimate that a supercomputer capable of processing 500 petabytes of data would be required to run the full simulation of the human brain. This phenomenal computational requirement is needed because there are 100 billion neurons to model, and each one requires 400 independent simulations (and the power of a laptop computer) to accurately replicate the complex chemical activity of its biological counterpart.

One of the most exciting implications of Blue Brain, and the subject of Markram's TED talk, is its potential ability to allow us to step into another brain's reality. Markram has said that “there’s no reason why you can’t get inside Blue Brain ... Once we can model a brain, we should be able to model what every brain makes. We should be able to experience the experiences of another mind.” In order to project a brain's interior experience into perceptual space, the code that generates the electrical objects of neural activity need to be deciphered – a challenge that Markram insists is not impossible. If Markram's projections are true, it will one day be possible to see the world through someone else's eyes.

Day 013 - Simulating the brain

Submitted by Sam on 3 June, 2011 - 00:00

Henry Markram, like Sebastian Seung, is a neuroscientist working towards a connectionist model of the brain. Like Seung, Markram is convinced that it is fundamentally possible to fully simulate the human brain, and that the only limiting factor in doing so is research funding: “it's not a question of years, it's one of dollars” 1. Like Seung, Markram also gave a TED talk about his work modelling the brain, which is linked below.

In his talk, Markram suggests that our brains project a 'perceptual bubble' around us, created from thousands of decisions and inferences that we subconsciously make: 99% of what we 'see' is what we infer, not what comes through our eyes. The question which motivates Markram's talk and which animates his efforts to model the brain is, “can the brain build such a perception?” - does it have the capability to generate it's own reality?

Markram is director of a supercomputing project, Blue Brain, which aims to answer this question. We'll have a look at his methods and progress tomorrow.

Day 012 - Connectionist theories of mind

Submitted by Sam on 1 June, 2011 - 23:32

Yesterday's video, Untangling the brain ended on an optimistic note for the future of connectomics, projecting that neural maps might perhaps “one day reveal how a tangled mess of billions of cells enables us to see, to dream, and to study the brain as scientists, and how to repair it when it goes wrong.” MIT's Sebastian Seung shares this optimism, believing that we are our connectome, and that one day we will have the technologies to test his theory “I am my connectome”, which he lays out in his 2010 TED talk below.

In essence, Seung and other connectionist neuroscientists theorize that our personality and memories are encoded in full by our neural connections. Unlike our genome, which we possess in its entirety from birth and which changes only by random mutation over our lifetime, our connectome must be constantly changing as experiences and memories are encoded in to the neural pathways of the brain. This part of the theory is substantiated by the decades of evidence for the neuroplasticity of the brain, which shows that the brain is able to change both anatomically and physiologically as the strength of connections between neural units are changed, and synapses are created and deleted. The only real uncertainty in Seung's theory is whether these changes are indeed the basis of memory and the encoding of experience and personality.

If they are, and memories are indeed stored as chains of synaptic connections, as sequences of neural activity which are activated in the brain during memory recall, then it will be possible to represent any mental state through a numeric description of activation values of the neural units in the brain. Memories, thoughts and personalities will be able to be expressed as n-dimensional vectors.

Day 011 - Towards a circuit diagram of the brain

Submitted by Sam on 31 May, 2011 - 22:54

Neuron cells are so small that they cannot be seen clearly with a light microscope, as the finest branches of their tree-like structures are less than a tenth of a micron in diameter and so are smaller than the wavelength of visible light. Neurons are so small and so highly entangled with each other that they can only be seen with an electron microscope. Their tiny size and dense interconnectedness makes the work of mapping a connectome of even the simplest nervous system an exceptionally difficult task.

It took over a decade for scientists to map the 302 cells that make up the nervous system of c. elegans, and despite over a quarter-century of technological progress in imaging and automation techniques the process is still painstakingly manual today. Just as our friend the c. elegans worm was chopped into 50nm slices, photographed and reconstructed by hand from the resultant 8000 prints 1, so scientists in pursuit of a connectome today must combine knife and microscope to generate a three-dimensional structure from a sequence of two-dimensional images. The very recently released video below provides a very nice visualization of how this process was used by researchers at both the Max-Planck Institute for Medical Research and at Harvard, as well as an extremely concise overview of the scope of connectomics as a whole.

As the research of neural wiring is still very much in its infancy, the algorithms that are currently available for automating the tracing of neural pathways are still far from perfect, making classification mistakes which might merge two neurones into one, or split one into two. Humans are currently faced with an incredible amount of manual work just to trace all of the branches in a cubic-millimetre of neural tissue. However, in what will become a motif of this blog, scientists like Sebastian Seung, Professor of Computational Neuroscience at MIT, are optimistic that advances in our technological capabilities will one day allow researchers to overcome these technical problems. Once we have robust pattern-recognition algorithms, Seung believes that we will be ready to find whole connectomes, starting with simple nervous systems and scaling up to larger brains as our technical abilities grow 2.

  • 1. White, J. G., E. Southgate, J. N. Thomson, and S. Brenner. "The Structure of the Nervous System of the Nematode Caenorhabditis Elegans." Philosophical Transactions of the Royal Society B: Biological Sciences 314.1165 (1986)
  • 2. Seung, Sebastian. "Connectomics - Dana Foundation." Brain and Brain Research Information - Dana Foundation. Web. 31 May 2011. http://www.dana.org/news/cerebrum/detail.aspx?id=13758.

Day 010 - Connectomics

Submitted by Sam on 30 May, 2011 - 22:05

A complete map of the totality of a brain's neural connections is called a connectome. The term was coined in 2005 in anticipation of a new field of neuroscience – connectomics – being driven into existence through technological advances that facilitate the tracing of neural wiring in brains. There is an intentional similarity between the words connectome and genome (the entirety of an organism's hereditary genetic material) which indicates the potential significance of this emerging field.

Appropriately, The Human Connectome Project bears a strong resemblances to The Human Genome Project, which completed its mission to produce a complete sequence of human DNA in 2003. With the successful description of our friend c. elegans' connectome as a precedent, The HCP aims to produce a complete structural description of the human brain by using imaging techniques to comprehensively map the circuitry of 1200 healthy adult's brains. It is a multi-million dollar research project of exceptional scale.

The project is consequently faced with many of the criticisms that were levelled at The Human Genome Project, principally concerns about its ability to achieve results in a realistic time frame. Regardless of the viability of the project's deadlines, the potential consequences of successfully mapping a human connectome are manifestly too important not to pursue. A complete schematic of the wiring of our brains would help us understand diseases like autism and schizophrenia, and would give us critical insight into the foundations of our consciousness, our intelligence, our memory and our personality.

Syndicate content
Drupal theme by Kiwi Themes.