orbifold

Yes, that would be another example.

- Weights are adjusted at the end of the backward in time (adjoint) integration, according to the weight gradients accumulated at pre-spike times.

- We only consider one kind of model system in this paper but this method would work for any kind of hybrid dynamical system, so also other physical substrates (a lot of exciting work to do there).

- We used to sell a neuromorphic hardware system Spikey for ~3000 Euro (basically at cost), we've recently completed a similar project, we also provide access to remote users via the ebrains collaboratory (https://ebrains.eu/service/collaboratory/). There are a number of commercial offers in the works (SynSense, Inatera). You can also buy SpiNNaker boards or access them via ebrains. Loihi and TrueNorth either don't sell or are pretty expensive, but they have "research agreements" in place.

I agree that the choice of task is not ideal. It is something that I struggled quite a bit with, since coming up with a good task can be a lot of work. Unfortunately even some of the "neuromorphic" datasets that are in use can be solved by massive temporal averaging or result in reduced performance of the network relative to "analog" temporal input (e.g. on Google Speech Commands). I'm collaborating with a group that is interested in event-based vision and control, so hopefully this will result in more practical/impressive demonstrations in the future.

I actually think I largely agree with the points that you made in the comment thread above. The name of the game can't be to just map a deep neural network to a SNN, I find it far more interesting to identify more natural mechanisms by which SNNs are able to perform information processing on sparse asynchronous event-based data. Also thank you for the nice words :).

I agree with these points, however the main advantage of the method presented in the paper is precisely that both the forward propagation and backward propagation can be seen as being performed by a network operating on temporally sparse events. We absolutely had event-based sensors and control as a motivation in mind. The fact that you can write down the connectivity of the neurons in terms of a weight matrix, does not mean that it can't be sparse. Since you are actually processing one spike at a time (potentially asynchronously), you don't need to implement any matrix multiplication. Current neuromorphic hardware achieves at least some degree of sparsity in their synaptic crossbars (BrainScales2, Spinnaker) or largely eliminates them like Loihi.

That number is what they reported in their publication, but it turns out they actually used both recurrent neurons (compared to feedforward as they state) and 512 instead of 100 neurons (see here https://zenkelab.org/publications/errata_zenke_vogels_2021/). We will adjust those numbers in the final publication.

My aim is to release the method as part of Norse https://github.com/norse/norse. There is some subtlety involved in implementing it for a given integration scheme, though. The event based simulator underlying the paper will also be released in due time.

One of the authors here, happy to answer any questions you might have.

It's also extremely expensive to train for any of these sports, in figure skating in the US you pay hundreds of euros every month for coaching and access to an ice rink only to a have a tiny tiny chance to be selected as one of ~5 competitors max. That doesn't include the mandatory very expensive costumes etc.

Well Poland had the brilliant idea of attacking https://en.wikipedia.org/wiki/Polish–Soviet_War, right after the end of WW1, so I am sure there was no love between them.

Sure because Russia has a long history of attacking Western Europe unprovoked. Napoleon, WW1, UK US intervention in the Revolution, Polish Invasion and WW2 clearly had Russia as an Aggressor.

There are 4.6 bitcoin transactions per second, how many dollar transaction do you think there are per second and what kind of volume of dollars are traded. Also there are different kinds of dollar in circulation, for example M2 is much higher (by a factor of 10) https://fred.stlouisfed.org/series/M2SL.

There is some path dependency here, trains came earlier than cars and early industrialised nations build rail networks that were far more extensive than they are now. There were also freight trams in cities like Berlin, which delivered cargo to smaller industrial sites in the city.

There are a few famous instances where mathematicians ran seminars for close to a decade and developed extensive theories on that way. Von Neumann appears to have been one, Grothendieck is a famous other example. The lectures are basically then meant to be taken by people that dedicate almost all their time to understand them.

In other words they are lecturing as they are developing the new theory. Clearly such things probably happen much less often now a days, because the overall research landscape is such, that this is not really rewarded. On a smaller scale research seminars in mathematics still are like that though, the results discussed there are often only circulated among the participants, which then make them available to the public piece by piece. Peter Scholzes breakthroughs were widely discussed in Bonn long before they were accessible to the overall public and even then most of what was taught in the initial seminars in Bonn wasn't circulated.

This is just factually untrue the GDP of the British empire was much larger than that of the US and in terms of scientific innovation Germany was widely considered to be at the forefront with France and UK as equals. The US was largely a scientific backwater.

Probably he is referring to a situation in China.

Regardless of the physics content, if the constructions he proposes turn out to be well defined they might be a basis for some nice mathematics.

You can get a dump of all your messages with someone pretty easily.

Yes in Germany you need a master of education to be allowed to teach even elementary school. Teachers are paid pretty well overall and the equivalent of Highschool teachers have done 5 years of university study.

Yes, I can totally believe that, the software itself is actually distributed in a fairly convenient format, given that some of it is only compatible with some obscure RedHat version, so installation amounts to copying files from an iso image and pointing it to a license server. It probably is also easy to patch out the License server code. I would be fairly surprised if those places also had access to current gen PDKs and the required memory compilers. We have to basically order memory blocks by mail and don't get access to the tools to generate them ourselves.

The US is heavily investing in that, there are recent DARPA programs (POSH, IDEA, SHIELD, SAHARA, ...) addressing this and continuous government investment in semiconductor technologies.

The professional tools are licensed at 100k-200k/year per seat each, there is no way around them if you want to actually tape out the chip. An experienced backend engineer needs 3-4 months to take a design from Verilog to a file that can be send to a Foundry. You also need access to a PDK from a manufacturer like TSMC and probably will have to license some of the digital-analog components (PLL, IO-Pads) from someone. Manufacturing itself is fairly expensive as well, we taped out via europractice ( https://europractice-ic.com/schedules-prices/), you pay about 12k per square millimetre in TSMC 65nm. Even older processes are more affordable, but you have to keep in mind that each of them requires at least some customisation and access to blocks specific to the process (memories for instance).

On the design and verification side the story is slightly better, verilator (https://www.veripool.org/wiki/verilator) for instance is an excellent tool that lets you simulate system-verilog code with high performance. It has seen wide industry adoption in the last years and now a couple of companies like Google, SiFive, etc. are investing in open source tooling for hardware. You can go a long way to building your design without having to pay for expensive licenses. The downside is that most of the interface type components (PCI-Express, DDR4, etc.) are prohibitively complicated to build yourself, so you will need to rely on external IP at some point.

This is what China doing nothing looks like https://twitter.com/globaltimesnews/status/10020822477368852....

I was under the impression that this more or less has happened both in Syria and Armenia.

The constant shilling for Julia gets old after a while. When it comes to numerical mathematics the implementation language is just a way to express ideas that are more or less independent of whatever language you are using. Last time I checked Julia had close to no professional grade libraries for PDE solving , while there are multiple deal.ii, FenICs, petsc high quality libraries for C++. Don’t get me wrong Julia is great, but what really matters is the quality of the available libraries. The differential equations work in Julia is nice and obviously the resulting code is performant, but a ton of work in numerical mathematic is spent on adapting methods to specific problems and at that point C++ feels like a much nicer choice to me in many cases.

Signal is a US company and „do not roll your own crypto“ is an NSA meme. It translates to „only use crypto we probably know how to compromise“. That doesn’t mean the algorithm itself must be wrong, it could just be that the implementation has subtle bugs.

Telegram hires excellent developers and was started by someone that had successfully run a company before. The clients for the different platforms in the case of telegram were developed by very small teams (1-2) of exceptional programmers.

- macOS client: https://github.com/overtake/TelegramSwift/graphs/contributor.... Written almost single handedly by one developer.

- Telegram Desktop: https://github.com/telegramdesktop/tdesktop/graphs/contribut.... Written almost single handedly by one developer, who also wrote almost all of the supporting Qt code.

When I joined twitter a few month ago to promote my scientific publications + open source software projects, I set myself a number of clear rules:

1. Don't engage with or comment on political / social issues.

2. Don't install the app, set a time limit on twitter for all devices.

3. Follow other scientists with few exceptions.

4. Flag posts as irrelevant to me that express anger or controversial political opinions (I especially don't need to know anyones take on the Corona situation).

So far it has worked pretty well, I've got invited to give talks based on initial visibility I generated on Twitter. The open source software project I'm working on came to the attention / got engagement / retweets from relevant people. I reconnected with people that I met a few years ago and it even generated a few leads for job opportunities.

It is also really useful to keep track of scientific progress in a field and seminars / workshops I might want to attend.

The algorithm basically works against you though, even if you follow a few people that post / retweet a lot, they can come to dominate your newsfeed. I also came very close to engage with social / political content, simply because popular content that gets shown to you is often the most controversial. I ended up banning a few mainstream politicians, whose tweets kept getting retweeted by people I wanted to follow and that has solved the problem for the most part.

Overall so far twitter seems to create more opportunities than issues for me, but I notice that I spend way more time there than I initially planned.

There are lots of non-gaming related apps that would be near impossible to use as a visually impaired person. Among them image/video editors most DAW (digital audio workstations) etc.

I think that is more a matter of being young (a kid or young adult) as opposed to related to any specific technology. Think of all the books that you could potentially read if you wandered into a library. Decades of knowledge waiting to be discovered.

I agree that calcium seems like a natural candidate and I suggest as much in my thesis. Coming from physics, I didn't know about reverse replay in the hippocampus for a long time, but I also have this association now. I would be glad to talk more, is there a way to reach you?