Neuromorphic Videos to Binge-Watch During Lockdown

Article By : Sunny Bains

A list of neuromorphic video clips curated to introduce you to some key neuromorphic computing approaches, platforms, and players...

[Editor’s Note:  In this installment of her column covering intelligent machines, Sunny Bains presents a list of neuromorphic video lectures curated to introduce you to some key approaches, platforms, and players.]

Though parts of the world have succeeded in suppressing the coronavirus and are now opening up, it will be some time before we can start traveling to conferences again. I was supposed to attend two meetings this spring and then the Telluride Neuromorphic Engineering Workshop this summer. I enjoy poring through the literature, but was looking forward to hearing from the researchers themselves. So I decided to console myself by putting together a list of (mostly) recent technical neuromorphic video talks available online and have shared these with the neuromorphic community (and now with you). So far, I’m up to 180+.

I find conference presentations a much better way into new subject matter than papers: you get a context, explanation, and overview without being bogged down with technical details. However, a list this long is not much use if you’re new to the field, so I thought I’d create an introductory conference session for those looking for a quick immersion. Note that this is not my perfect line up: there are speakers who I would have liked to include who didn’t happen to participate in a recorded conference or seminar over the last year or so, and others who were recorded but the video is too poor to be useful. But it’s close enough.

Chips
The first three neuromorphic video talks I’m going to recommend come from the designers of some of the main neuromorphic hardware platforms currently in use. Although I recommend reading my introductory feature on neuromorphic computing before you start, the advantage of these talks is that they each introduce the topic on their own, as well as getting into the researcher’s own work. By the time you have watched all of them, you should have a clear picture of what neuromorphic computing is, what its concerns are, and how the platforms differ from each other.

Although you can watch these in absolutely any order, I’m going to suggest you start with a talk that Kwabena Boahen from Stanford University, CA, gave on Spike-Based Neuromorphic Computing (30 min) at the 2019 Stanford Computer Forum. Boahen starts off explaining why conventional deep learning techniques (like back propagation) constrain the way engineers design systems, and goes on to motivate analog neuromorphic computing in general before talking about his own work, Neurogrid and (more recently) the incredibly energy-efficient Braindrop.

After that, I would get slightly less neuromorphic and listen to Mike Davies from Intel Corp., Portland, OR, discuss the Loihi chips, which have digital computation but spike-based communication. At the Neuro-Inspired Computational Elements Workshop (NICE) 2019, Davies gave a talk on Advancing Neuromorphic Computing from Promise to Competitive Technology (60 min)Like Boahen, he starts with deep learning, explaining how what neuromorphic does is different (but in a different way, so worth hearing). He then discusses how Loihi was designed, why his group is pursuing a digital approach (determinism), and how much of an advantage Intel’s technology has over other processing techniques. He also discusses the company’s approach to by developing applications by supporting a large community of users.

A talk by Mike Davies from Intel

SpiNNaker is another hardware platform that has been widely used for developing neural applications around the world. It’s not really neuromorphic from an energy perspective, but it’s spike-based and is used to test computational techniques that can eventually be run on more-efficient neuromorphic hardware. Steve Furber of the University of Manchester, UK, discussed SpiNNaker Applications at NICE 2019As well as explaining how his partners have use the technology for everything from theoretical neuroscience to robotics, his neuromorphic video talk covers the history of the project over two decades, and the work leading up to the SpiNNaker2 chip. Unfortunately, the talk is from last April, before the chip was finished, but it should provide a good introduction to new work from the group.

A talk by Steve Furber of the University of Manchester

Cognition
At this point, it might be useful to switch away from hardware to software: to think about how to program brain-like systems. For this I suggest a talk that Chris Eliasmith gave earlier this year at home at the University of Waterloo, Canada, on Spiking Neural Networks for More-Efficient AI Algorithms (55 min). Here he describes how he and his colleagues developed the Neural Engineering Framework, the Neural Engineering Object (Nengo) which allows user to program their own networks, and the artificial brain his team built with 2.5 million neurons. The Nengo software, now developed and distributed through Applied Brain Research, can be used with all of the hardware mentioned previously: indeed, this group do important benchmark testing for the whole field.

A talk by a talk that Chris Eliasmith

Speeding up now, I want to move to cognitive sensing: creating meaning out of sensor inputs for robots and other machines. Amos Sironi of Prophesee in Paris, France, gave a nice talk at the 2019 Conference on Computer Vision and Pattern Recognition (CVPR) on Learning from Events: on the Future of Machine Learning for Event-based Cameras (15 min).

Amos Sironi of Prophesee

At the same meeting, Andrew Davison from Imperial College London, UK, talked about Novel Hardware for Spatial AI (25 min): a not very enlightening title for a very interesting presentation about using SLAM (Simultaneous Localisation and Mapping) techniques in robotic vision using the same event-based cameras.

Andrew Davison from Imperial College London

Again at CVPR 19, Yulia Sandamirskaya from the Institute of Neuroinformatics (INI) Zurich, Switzerland, discussed Neuromorphic Computing: towards event-based cognitive sensing and control (25 min). Weirdly, she manages to talk about this in a way that is interesting, mathematical, and accessible.

Yulia Sandamirskaya from the Institute of Neuroinformatics (INI) Zurich

Finally, I would end with another INI Zurich talk, from NICE 2019 again. Giacomo Indiveri gave a presentation called DYNAP-SEL: An ultra-low power mixed signal Dynamic Neuromorphic Asynchronous Processor with Self Learning abilities (35 min). This talk is important partly because it represents the fourth of the main neuromorphic platforms currently in use for research. However, what I particularly appreciate is that he covers some important ground not covered by the others: this includes the use of neuromorphic devices in healthcare, 3D silicon (discussed in my columns in April and particularly May), and memristors.

 A presentation by Giacomo Indiveri

— Sunny Bains teaches at University College London, is author of Explaining the Future: How to Research, Analyze, and Report on Emerging Technologiesand is currently writing a book on neuromorphic engineering.

Explore more:

Accelerating Your Machine Learning Compute for the IoT and Embedded Market with Arm

Take advantage of wide-ranging AI opportunities. GigaOm report shows you how to devise, define, and deploy the right AI for job.

 

Leave a comment