The origin story of ABR: What is Spaun?

ABR was spun out of Professor Chris Eliasmith's Computational Neuroscience Research Group (CNRG) at the University of Waterloo about two years after the Spaun model was published. Just before that, Peter Suma had joined the CNRG to see if their approach to building neural and cognitive models might have commercial applications.

A few years earlier, the CNRG had started using their Neural Engineering Framework (NEF) and Semantic Pointer Architecture (SPA) to build models targeting neuromorphic hardware. The combination of advanced algorithms, the ability to run them on the latest hardware, and Pete's extensive business background made ABR a no-brainer. Today ABR is pushing neuromorphic computing as a new platform, demonstrating how to achieve state-of-the-art power, speed, and accuracy across a range of AI applications.

What is Spaun?

Spaun is the world’s largest functional brain model. It was created using specialized simulation software for modelling neural networks called Nengo [1].

Who started Spaun? When, where and why?

Spaun was started in August 2011 as a Computational Neuroscience Research Group (CNRG) project within the Centre for Theoretical Neuroscience (CTN) at the University of Waterloo. Chris Eliasmith is the director and team leader of CTN and is the originator of Spaun.

Eliasmith began this project simply because he had a general interest in brains. He never had a dream planned out in his head to create Spaun, rather it was his passion for Neuroscience, Philosophy, and Engineering which led him to begin this project. Eliasmith is jointly appointed in Philosophy and Systems Design Engineering, and is cross-appointed to School of Computer Science [2].

What does Spaun look like? This simulated brain is comprised of three key components: an eye, brain, and arm. All of these components share the same purpose as the same body parts on animals. The appearance of Spaun can be referenced from the image above.

First, the eye is crucial because it provides information for the brain to process. It locates this information on the vertical screen attached to the laptop-like graphic Spaun faces. In the case of the image above, the question mark is information or input. Spaun would be blind without its eye and perform very poorly on the tests used to study it.

Second, the brain works according to a specific model known as the Semantic Pointer Architecture Unified Network. The various “thought bubbles” surrounding the brain show how Spaun processes information from its eye. As well, the colours on the brain symbolize different levels of neuron activity: the cooler the colour (blue and violet), the lower this activity while the warmer the colour (red and yellow), the higher this activity. Lower neuron activity would mean Spaun is not using some or more regions of the brain very much. Higher neuron activity would imply the opposite. For instance, when Spaun is writing out an answer, the region of Spaun’s brain responsible for motor control would be a warmer colour.

Third, the arm writes out Spaun’s answer to a test after its brain processes the input. This is displayed on the horizontal surface of that “laptop” Spaun faces. Additionally, a top-down view of Spaun’s answers or output is seen in the top-right corner of the simulation. What is interesting is that the output is not exactly neat or perfect; rather Spaun has flaws with being concise and does scribble, similar to us. This has to do with how Spaun perceives its output. Output tends to get extra messy if Spaun is attempting to re-write the last elements of a long list.

The following link presents some of the tests where Spaun receives numerical input and writes output: http://nengo.ca/build-a-brain/spaunvideos.

How does Spaun work?

Spaun was programmed in Nengo, a development environment for modelling neural networks. The connection between Spaun and neural networks is that the brain is a neural network, which is a collection of nerve cells that communicate with each other called neurons. Much like a human brain, Spaun operates with sub-networks of neurons exchanging information with each other, and some of these are responsible for perception, action, and cognition. More specifically, these sub-networks model only a fraction of the sections of the human brain, including the prefrontal cortex and basal ganglia. This model of how the brain works is exactly in the name of this incredible brain, Semantic Pointer Architecture Unified Network (Spaun) [3].

Similarities and differences between Spaun and the human brain

Similarities

Both Spaun and the human brain:

1. contain the following regions: motor cortex, striatum, and globus pallidus (C. Eliasmith, personal communication, February 15, 2017).
2. have the ability to control an arm and write thoughts.
3. perform recognition tasks of writing out either what single number it sees or from a list. This includes the number corresponding to a certain position in a list [4].
4. possess approximately 100% chance of recognizing typewritten numbers from 0 to 9 [2].
5. perform the same recognition tasks as mentioned in 3, but writing the digit as identical as possible (called ‘copy drawing’) [4].
6. can count numbers and write them out, starting with a number and increase by a certain number [4].

Differences

1. Spaun is missing regions of the brain, including the hippocampus, amygdala, and cerebellum (C. Eliasmith, personal communication, February 15, 2017)
2. Spaun only deals with numbers 0 to 9 [2]
3. Spaun possesses approximately 94% chance of recognizing handwritten numbers (from 0 to 9), while the human brain is about 98% accurate [2]

In the end, Spaun is incredible. Initially, the project team worked on Spaun by programming the subnetworks that comprise this brain. Then they put these together. Amazingly, it took 6 months. It took 6 months after beginning Spaun in August of 2011 to be where it stands on the Nengo website right now: http://nengo.ca/build-a-brain/spaunvideos. The future of Spaun is expanding on how it learns and upgrading these networks. Every second of Spaun running takes 2 and a half hours of real time to process. I can’t wait to see Spaun in a few years!

Back to Spring 2017 Home

References

[1] “Videos for Spaun simulations”. Nengo. Web. 01 February 2017. <http://nengo.ca/build-a-brain/spaunvideos>.

[2] Eliasmith, Chris. Chris Eliasmith’s Homepage. Web. 01 February 2017. <http://watarts.uwaterloo.ca/~celiasmi/>.

[3] “Chris Eliasmith on Singularity 1 on 1: We Have Not Yet Learned What The Brain Has To Teach Us!”Youtube. Web 02 February 2017. <https://www.youtube.com/watch?v=ZjDp8EnfDvo>.

[4] “Spaun – several tasks”. Nengo. Web. 15 February 2017. <http://nengo.ca/build-a-brain/severaltasks>.

[5] [Spaun]. (n.d.). Retrieved from https://www.extremetech.com/wp-content/uploads/2012/11/simulated-brain-canadian-diagram.jpg

‍