This moves it closer to a "temporal" integration as well. Since the factor on the residual branch can be considered a timestep in an Euler-discretization.
Good chance that Green Party received significant support from outside forces to derail a sensible energy policy. Relying on fossil fuels / gas makes Germany way more dependent on the US and Russia, then otherwise.
Let's see, first of all this is a research project not an attempt to build a commercial product (yet). Indeed it is not currently optimised to do analog matrix multiplication particularly well, but what prevents it from performing better is well understood and relatively easy to fix. Of course we are aware of commercial efforts, but this is a pretty long running research project, so not all of those necessarily existed, when it was set in motion. As an example wafer-scale integration was accomplished on BrainScaleS way before Cerebras pushed for it, clearly they have executed far better on building a commercial product around it.
Our motivation is to build large scale accelerated neuromorphic hardware and to prove that it can be useful. It is not particularly efficient or even feasible to train SNN with GPUs over large timescales, so eventually we will need to use on-chip learning. This paper could be seen as an intermediate step, it's useful to know that the hardware can be optimised in the very least as a base line for further experiments.
Applications are not our primary concern at the moment, for the most part we believe that once we have identified the right algorithm(s) and hardware that is able to support the implementation of these algorithms it will be possible to apply it to many problems. For ANNs backpropagation had been figured out a long time ago, but the recent successes started after 2010. To be clear for SNN inference our chip is far faster and has far better latency than a GPU and is roughly ~10x faster than Intel's Loihi. Application areas for that are admittedly niche, especially as long as we can't scale to far more neurons.
The chip under discussion is able to do both (analog matrix multiplication and SNN operation). Briefly there are a bunch of “nano”-devices which are more amendable to spike based operation. Moreover analog computation is hard to scale up so the layer wise digitalization and weight loading and communication wipe out a lot of the potential benefit in the case of analog matrix accelerators. Part of the advances in recent years make SNN work “well enough” in software, especially considering the relative smaller overall investment in them.
One of the co-authors here. Demonstrating inference on a neuromorphic chip can already be challenging, especially if it contains analog components. This is a way to make this kind of hardware "useful" on a given task. Of course learning on chip is the holy grail, but given that this kind of hardware has potential latency and power-consumption advantages for small (event-based) data, a general way of optimising the weights is pretty cool, I think.
ML in general seems to have a really low standard for citing related and earlier work. In part I think that is because a lot of papers are based on relatively poorly developed foundations or empirical evidence. A prime example are the "Neural ... " papers, which in my view largely took existing research in the optimal control / differential equations literature, prepended the word "Neural" and provided some "interesting" application. This would be fine, if there was some proper context given for the work, but unfortunately that wasn't always the case there.
These conservation laws apply to open systems (which biological platforms are). All ready at the cell level the primary concern is how to "efficiently" do energy cascades which sometimes have unintended (irreversible) side-effects. So cells get "garbage collected", there are other more system level irreversible changes, some intended (memory) and unintended, which I guess are indeed ultimately unsalvageable, which is why reproduction is a thing (pristine copy of genome in new cell).
Yeah, whenever I see others scrolling through content at Facebook, I can't help but wonder who would voluntarily subject themselves to so much garbage information. It is similar with Twitter, which I use for "science communication". But the amount of miscellaneous memes and low quality click bait ads that one has to endure is almost physically painful.
Who would want to buy a Facebook phone and with what operating system? I think “Social OS” could have been a good idea, but the actual Facebook App sucks.
One of my favourite memories from school was translating parts of Cicero's words from latin. He is considered "easy" to read and translate precisely because he took great care of not using difficult or ambiguous language in his writing as well.
It is kind of fun to just not follow national news and then get into conversations with relatives / friends that do. Most conversations are really pre-programmed including the views that they will have. "Telegram is bad because it is used by extremists" for example. The most egregious example being COVID of course. The different national narratives were slightly out of sync at all times, so that people got told the Vaccines were effective at preventing infection, spread and the "way out", while at the same time another country was already running Booster programs because two doses were not enough and weren't even effectively preventing hospitalisations after 6 months.
Pretty sure it is the NSA (Nakamoto SAtoshi), most likely they’ve broken the underlying cryptographic primitives and now use Bitcoins as their slush fund. It also fits perfectly with their mission statement, as they can now precisely track almost all illicit transactions and they should be able to unmask both source and targets pretty easily.
Sure but that doesn't include shutting down Telegram. It is easy enough to infiltrate those groups and keep them under surveillance. You can do very basic police work to counter act them.
I judge them very harshly, they deserve no respect or sympathy. It should never be acceptable to walk with people that minimise the holocaust or talk openly about how they had talks with former army members of armed insurrection in their telegram channels (there are screenshots to prove that). They chose "Reichsprogrom Nacht" of all days to do a candle procession through Heidelberg for instance, with known Neo-Nazis in their ranks.
But yeah they should be free to continue to self-incriminate themselves on Telegram and elsewhere. It is not like they are some kind of sophisticated state level threat.
I wouldn’t have a problem with it if it was just that. Police here deserve to be stress tested and in principle you can spontaneously declare something to be an assembly by naming someone the assembly leader. The issue is their message and the right wing / nazis that they are walking among them.
Arguably in a well run society those groups would have so little power / influence that they wouldn’t be limited by how well they are able to communicate privately. The student protests in the 70s and coocurring terror campaign was not happening in a vacuum but because former Nazis had been part of all aspects of post-war Germany and young people were not having it.
Teenagers will be cruel no matter what. Finding non-shitty friends in Highschool is an important skill, that is pretty much independent of what phone you have.
Wild conspiracy: What if the omicron variant was created by the same process as is used in GoF research more generally. Use humanised mice with human ACE2 receptors and pass it through them over many generations selecting for infectiousness / immune escape wrt the existing vaccines. Maybe they even observed anti-body dependent enhancement (ADE) in mice.
I don't think this is entirely implausible because current research is doing things along those lines: There is an established "mouse" model now and people are actively working on chimeric vaccines, engineering virus variants with different spike proteins (for chimeric vaccines see e.g. https://www.science.org/doi/10.1126/science.abi4506). They engineered several COVID variants in that paper, some which antagonise the innate immune response.
> To confirm that scrambled coronavirus spikes are biologically functional, we also designed and recovered several high-titer recombinant live viruses of RsSHC014/SARS-CoV-2 NTD, RBD, and S2 domain chimeras that included deletions in nonessential, accessory open reading frame 7 (ORF7) and ORF8 that encoded nanoluciferase (fig. S1C). SARS-CoV-2 ORF7 and -8 antagonize innate immune signaling pathways (25, 26), and deletions in these ORFs are associated with attenuated disease in humans (27, 28).
and made sure that the "single" spike vaccine was ineffective vs. the chimeric vaccines, which gave broad protection.
There is bioweapon potential in these combination of technologies beyond simply engineering a virus, I won't go into detail here, since maybe (however unlikely) no-one else has come up with this particular scheme. However recent work by Baric and others leads me to believe that this kind of scheme is precisely their endgame.
It is ok to feel lost and overwhelmed from time to time. Even the most brilliant people often go through hardships and difficult times. Try to find things to do that you enjoy not because other people expect them of you. External validation can be nice but also a burden. Set your own goals and try to check from time to time how much progress you have made to accomplish them. For me at the moment, this is to publish a bunch of good papers at ML conferences and land a permanent research position. But this can literally be any other goal and I don’t think it is helpful to compare yourself to others, too much. Opportunities will come to you if you are doing good work.