This comment is for the algorithm. As an AI enthusiast, I rarely see someone this close to the bleeding edge of this technology. Great content. I had been curious how we were going to get faster when chips are already at the molecular level - and BAM a heroine comes to the rescue. Hooray! I'd subscribe to anyone for this, but it makes me especially happy to see a young lady delivering the information. Pretty and smart CAN go hand in hand! I hope Anastasi becomes far more famous than the Kardashians of the world. Girls need more role models such as this.
Explain neuromorphic chip. ChatGPT Response: A neuromorphic chip is a type of computer chip that is designed to mimic the structure and function of the human brain. It is inspired by the way neurons and synapses in the brain work and is intended to enable more efficient and flexible computation. Neuromorphic chips are typically made up of a large number of artificial neurons and synapses that are connected in a way that allows them to process information in a similar way to the human brain. They can be used to perform a wide range of tasks, such as image recognition, natural language processing, and decision-making. One of the key advantages of neuromorphic chips is that they are highly energy efficient, as they are able to perform complex tasks using much less power than traditional computer chips. They are also able to operate in real-time and can learn and adapt to new information without the need for explicit programming. There are a number of different approaches to designing neuromorphic chips, including using traditional semiconductor materials, such as silicon, and more novel materials, such as memristors. Research in this area is ongoing, and it is hoped that neuromorphic chips will eventually be able to perform tasks that are currently beyond the capabilities of traditional computer chips.
This is awesome. A neuromorphic approach to AI is something that I imagined even before joining engineering because it felt so intuitive. Here I am today with more questions after watching this video. This is the kind of research that excites me and makes me light up like a neuron lol.
As a long time software developer who has not yet worked with AI, this type of revolutionary news can be a little scary for me =) but it is also exciting and motivational to finally start learning AI so that I won't be obsolete in the near future.
The format of this video is FANTASTIC. Way better than just the raw interview and better than just the pure "report". If your channel were a stock, I'd be buying. Thank you for investing so much effort in this video. You are bringing the AI, journalism, and media world in a beautiful direction. I hope the future is full of videos like this.
Two points you made that I found very interesting. First is the low power requirements. The second is how neuromorphic chips are for targeting the distributed market as opposed to big machine learning data centers. It seems to me that between these two factors this could make a huge impact through a wide variety of mobile devices. Maybe even stop people looking at their phone from stepping in to traffic. Now wouldn't that be a wonder. :)
There are quite a few other companies working on spiking neural network chips (BTW, the idea dates back to 1952!). Prophesee in France is using their SNN for event-based vision. Sony even released a couple of image sensors with Prophesee's technology built in. It's going to be interesting to watch this over the next few years. Thanks for the video.
I've been following neuromorphic since IBM's project over a decade ago. I feel a bit sad they were 'ahead of their time' but they did validate the concept in an at scale "commercial" way (not lab with wires all over) and inspire the industry. It's wonderful to see this tech now on verge of coming of age, and now way more AI engineers are out there and commercial demand for AI in edge devices is high. Excellent video.
Well, most of the tech arises ahead of it's time, and fades away out of either no use or being flat out discarded And then gets brought up again when it becomes actually practical and finds a demand
I wonder... it was about 20 years ago I re programmed an assembly line that had some problems with data sets, and was slow. I set a comparator, to see the data in, vs the data retained in 8 bit chunks, and if nothing changed, I would skip re sending the information, until there was data that needed to update. It was massively faster, and more efficient. It almost sounds like I set a primative neural net without even knowing it. BTW... it worked fantastic.
Thanks for these videos I'm starting to think this will be the history defining video series once AI takes over this series will show future civilization how AI take over started 🤣 great work as always thank you for always giving us a glimpse into future.
This is one of your most intriguing videos to date. I’ve been following neural network architectures for a long time. My first digital circuit designs were asynchronous, to gain the speed advantage of propagation over clocking. This new architecture seems to have a natural advantage over conventional deep learning systems. I look forward to hearing more about its future.
Pretty nice video!! Well organized and clear. From someone researching on SNNs and spiking architectures just some extra info: The loihi chip is not asynchronous per se. All the logic is clock based since it is digital. It is asynchronous on a message protocol level between neurons. Although it gives amazing improvements in power consumption, there is still room for SNNs. The akida chip is more focused on feedfoward SNNs converted directly from CNNs for simple tasks as classification. Also, only LIF neurons are supported. At the end, events are packed in frames and the temp information of the spike is lost. However, they consume less than GPUs. I'm glad that you mentioned memristors and RRAM architectures, not a lot of people know about it. Just to mention there are already some teams researching on photonics architectures, Check Ipronics and the UPV in Spain. Glad to see someone talking about neuromorphics in RU-vid! Keep going!!
Every now and then I stop by. I think it's because there are so few content creators for engineering and hardware on the deeper topics. The associations made with the chemistry field is interesting too. On spiking systems, I've noticed the similarity with how humans make snap decisions, a situation I'm familiar with. Maybe it was luck, but mine were usually the right ones. My point is, using these systems with others (e.g., GPT), along with expanded hardware speed and integration, could solve more complex problems like this, ones that appear unpredictable. QC? Anyway, thanks a bunch and have a great holiday.
This text was written in German and translated into English using chatGPT. Anastasi, thanks again for such a factual, competent and (can I say?) charming contribution. I feel validated by science. I am a software developer, administrator and AI enthusiast and now I finally have a name for where AI needs to go. You are right, classical approaches (and chatGPT probably also belongs to this category) follow a von-Neumann architecture, while the neuromorphic approach is more promising; I am a fan of decentralized structures and peer-to-peer anyway. I hope the name catches on because it hits the mark. If the project leaves the lab, Intel will probably have to make concessions to classical needs of chips and boards simply to lower costs for the end customer with third-party compatibility - if it ever gets to the end customer or prosumer! I would definitely be in if someone gave me a corresponding PCI-based card to play with :-) or even better - a USB or Lan box that I could connect to my Raspberry cluster,... my goodness, imagine if every household had an AI box, what that would change in the world,... Anyway, thanks again for the contribution, I always enjoy watching you 🙂
Yea, classical computers aren't doing well for AI, because they weren't designed for it to begin with They were just machines that should execute commands sequentially and consistently in sync with the entire system
I really love your videos! 🥰🥰🥰please keep up the hard work. I wish I could at least do coding. But I fail or am unable to problem solve it. My education is limited and my back ground did not set me up for success. I wish I was as capable as you. Truly enjoy your work , your field of research, and your energy for this world of tech/engineering!
Another amazing episode, thank you. Question: We are steadily moving ahead with advances in computer technology in every aspect, but it appears it is only a matter of time where the newest technologies come together and achieves improvements significantly beyond expectations. Based on your knowledge, awareness, education, intuition, gut instincts, educated guesses, etcetera, when do you predict the next major advance in home computers and business computers to hit the mainstream market? Any thoughts on what the top-of-the-line home and business computers will be like in five years? I hope you have a Very Merry Christmas and Happy Holy Days.
This guy has it right. This follows what I came up when around 10 years old, about how to emulate intelligence. Lots of progress over the decades. But my model had greatly expanded, but he hasn't talked about any of the additional stuff needed for intelligence! This is still complicated lego blocks!
Spiking Neural Nets in silicon form are not new at all. IBM was at this years ago in 2015 on the 28nm process with their TrueNorth chip. It had 4,096 cores each with 256 'neurons' that each connected to the other for 65,536 synapses per core. So 1,048,576 neurons and 268,435,456 synapses in a single chip all pulling just 70 milliwatts. Also it was using SRAM which isn't the most area efficient memory so there was a huge room for improvement in a design predicated on a MRAM or ReRAM based compute in memory architecture.
Because timing, tempo, and cadence are inherent to neuromorphic AI, I imagine it will be better at composing original music than traditional AI, which basically just repackages the music it was trained on that was created by human composers. Neuromorphic AI will have an intrinsic feel for music, as opposed to traditional AI that just shuffles around preexisting musical phrases.
One of the main topics of interest for me is organic computing. I expect that in the future, we can use DNA to encrypt and store digital information, proteins to build logic gates and neural networks, and whole cells as computing nodes. There are a couple of very interesting scientific papers about this, but are incipient and expensive
That can also help us live for thousands of years! I want to live to see fusion fully take place and traveling outside of the solar system! I'm 15 so i hope that happens in my lifetime!
@@malindrome9055 The way things are going, you would get to see the world decline into tribal ghet-toe stupidity for thousands of years trapped on this mudball drowing in human feces. Sounds dandy.
@@malindrome9055 i hope so! I'm 26yr and also I want to see the new arrival to the moon and Mars ... But certainly, this are bad times, in special for certain countries/peoples. I highly recommend this short film about the changing world :/ ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-xguam0TKMw8.html
Neuromorphic computing is a type of computing that is inspired by the way the human brain works. It involves the use of hardware and software systems that mimic the structure and function of neurons and synapses in the brain. The goal of neuromorphic computing is to create more efficient and powerful computing systems that can perform complex tasks, such as pattern recognition, decision making, and learning, in a more energy-efficient and robust way than traditional computers. One of the key features of neuromorphic computing systems is that they use parallel processing, which allows them to perform many tasks at the same time. This is in contrast to traditional computers, which typically use sequential processing and can only perform one task at a time. Neuromorphic computing systems also tend to be more energy efficient than traditional computers because they use less power and are able to perform tasks using fewer transistors. This makes them well suited for use in devices that need to be small and portable, such as smartphones and wearable devices. Overall, neuromorphic computing has the potential to revolutionize the way we think about computing and could lead to the development of new technologies and applications that were previously unimaginable.
Another excellent video on a fascinating topic. I think you combine well the excerpts from interviews with your own commentary. Mr. Davies is eloquent in his explanations but i think i would get overloaded if there were long uninterrupted monologs from him. This way the information is quite palatable for lay persons.
Marry Christmas. To All ; In real time energy efficient, high adaption, and optimization solutions ring like jingle bells to me. And it sounds great. Keep doing excellence.😎
You have become the number one source for the highest quality info and current reporting on developments in one of the most important sectors of development in the world. Micro Chip technology is the new nuclear weapon. It is behind drones which will be the future of war far but more importantly we will see that a counties sovereignty is tied to their economic health and ultimately using chips for AI, Block Chain Technologies, Transport, drones etc… will prove to be what determines if a country becomes eaten and eats other countries.
We're not far away from when u intro someone on ur ch , as a head of a company or engineer that's actually just something an ai has dreamed up and deepfacked a video and we won't know the different untill u tell us lol , happy Christmas and hopefully 2023 will b an amazing year for this ch
Our brain is primarily a massively parallel pattern matching system. Our physical body sensors provide the input to our brain's big data set. Most data fields are sensory images. Other fields are subsets containing thoughts, feelings, and inference relationship processing. Our brain has a secondary purpose of being a physical body monitoring system with feedback control of body functions. All of this processing by our brain is output at varying levels of awareness to our mind. Our mind is not located within our brain. Our mind is the envelope that wraps around our awareness, our sense of self. It is from our awareness that we relate to other beings and the physical world we share with them. It is within our awareness that we hold an imperfect knowledge of our self and our stumbling relationship to everything outside ourselves. The functioning of our mind is not indicated on any EEG fMRI, PET Scan, or other brain activity tool.
MRI scans clearly show that decisions are made subconsciously first, and then we consciously become aware of the decisiveness -- as we act. A great deal of practice is required to bring decision making to a methodical slow conscious act of listing the merits and relative weights of each input. This is not conducive to survival when snap decisions are required, but it is imperative for engineering.
@@imconsequetau5275 You are simply giving a great example of how lies are built in science. MRI, fMRI, PET scans and other brain activity sensing systems merely reveal brain activity localized to areas in the brain determined to be specific to certain thought activity of the subject. These regions were first determined by invasive wire stimulation and expanded by thousands of hours of studies of brain activity monitoring with external stimulation of photos, sounds, external spoken remarks, or requested subject thought phrases. It is pure vanity for any researcher to declare that the measured activity is related to the brain's formation of a specific thought. Just as it is vanity for me to assert that all the measured activity, cognitive and pre-cognitive, is limited to pattern matching activity around the topic generated or about to be generated in the subject's mind, which is external to the brain.
Super interesting & well presented, but until we see some practical applications it is impossible to know whether these architecture ideas are practical & powerful or just very early stage technology developments that greatly lag conventional architecture in practical performance. Ideally we would need to see direct comparison on tackling several edge computational tasks between the architectures. Maybe such data exists, but as of now I haven’t seen it & that makes me suspicious about the practicality of these approaches. Thank you for sharing!
This guys sounds exactly like a guy Ian Malcolm from Jurassic Park was talking about: “Your scientists were so preoccupied with whether they could, they didn’t stop to think if they should.”
SUBBED: My senior project was a digitally controlled network of Op-Amps. 32 op-amps with 3 digital potentiometers each was a big board (in the ‘80s i used resistor ladders). It could be programmed to perform nearly instantaneous analog calculations/processing, and could feedback to ‘weight’ each node, on the fly. Do you know of anyone pursuing this approach? The new chip stacking can make these nodes as dense as FPGAs. (op amps and digital are different FAB processes) (PS, I’ve had so many patents stolen I don’t care anymore and just went SAILING ! , LoL)
@@marmosetman That's an interesting (worthy) application: a digitally controlled analog array might be used to create the training data across a Monte Carlo of variable analog inputs for AI training extremely quickly
You mean Chuck Moore's asynchronous patent which a lot of companies, including Intel were sugned onto. Which Chuck used to design his low energy recied holding processors and processing arrays. You could simulate this AI network type on that fully.
Timing controls the context in which the data is interpreted by the brain, by sounds of it. That makes sense as well, because we interpret things from multiple I/O devices connected to the cantral ind peripheral nervous systems. Eg. Touch, sight, sound etc. Could you imagine if we did not do the same consciously? when interacting with the world.
Neuromorphic analysis is ok but it can be better when that neuromorphic node can explain what it did somehow to digital computer. So there must be built some top layer that can feed and retrive info from such computer. I built in 20 years ago one very simple node on Xilinx and used numbers sorted to Gray Code to address similarity of things that tree that is built on top of gray code is backed in psychology known term priming - and also explains why such computer must sleep to reorganize and optimize coding. So when you look at that structure then it classifies problems and when there is no place where you can put that thing then it stored to short term memory and will reorganized during sleep. And what is amazing all such computers have different ideas since Gray Code can have many variations and coding is also different. It is like rising child when child understands first thing then it starts to build similar tree what is unique and is not changing as much as computers tree of ideas.
asynchronous, capacitance , AC coupled. a far more direct way than event cameras. Mythic is back , very happy to see. the research will soon be keeping an eye on us.
The easiest way to succeed in any field is to work on it consistantly, and consistancy is the key to success. I experienced my self. By the way, thanks for the uploads , a fellow creator.
Wow! Back in the days we used to call this type of research "biomimicry". So, essentially, we tap into the way our brains function, and design algorithms around the synaptic actions. Gee wizz - nature is still light years ahead in efficiency and efficacy of function. Okay - not light years, maybe, but impressively advanced.
So much advancement! Which companies will bring advanced use chips to market first. Nividia seems to be ahead but IBM, Intel, Apple, Adobe, AMD all seem to forging ahead ...
You cannot say that in our brain “memory is distributed”. Scientists still don’t know, or cannot agree on what memory is; Let alone if it is distributed. Memory and thoughts cannot be confused with brain, and there is not certainty that it’s generated by the brain. Some people even said that coincidence, including memory, exists without the brain, for instance when a person dies and its spirits remains around.
Nonsensical conclusion. *Memory* and *Consciousness* are emergent properties of correctly functioning living brain tissue, and *they do not exist* in the absence of life.
I've watched a bunch of Mike Davies videos. The tech looks promising. My only doubts are that is so focused on catching up in the CPU market that i fear Loihi will never get enough attention from leadership to become any more that a research device.
I'm still of the opinion that we're likely to see less monolithic architecture physical layer philosophies in the future as opposed to more composite systems that exploit the most efficient use of chip/architecture types for the tasks and operations they're best suited. For instance we may see a resurgence in and growth in developmental sophistication of analog systems, and those systems get bolted on to a larger system that's also running Neuromorphic, Quantum, and other architectures. A composite system like this would be even more brain like in that a human brain isn't a solid mass of undifferentiated grey matter. Our brains have many different systems all evolved to perform their tasks efficiently. Motor neurons aren't going to play a huge part reading a book. The neocortex is going to be better at neocortex work than the hippocampus or the optic chiasma, although all of those may play a part in working together. Hind-brain structures will be better and more efficient at hind-brain related input and output than the frontal lobe. In this sense, I suspect we'll something similar in the future with the developmental sophistication of AI. After all, we already do some level of composite architecture in traditional computingas we've got graphics cards dedicated to graphics, networking chips for internetworking, and everything else tied together and working as a whole with the CPU.
Cybernetics goes back to James Clerk Maxwell in 1868, I read about it in 1954 in Norbert Wiener's 1948 book. There is nothing new under the sun. Essentially, you learn from nature and apply that to machines and apply that to nature... Its an endless loop.
After being raised in the wake of the industrial revolutionary dogmatic mindset it's wonderful to see all this idealism being put into testable practice right before our very eyes. Out with the dark bland materialism and hello to the revival of the idealistic age.lol
Neuromorphics r not what u think they r. They r meant to solve specific intuitive tasks like pathfinding/ walking/ keeping balance. It is not meant to replace complex thinking. That is still ML
