BRN Discussion Ongoing

[Frangipani]

View attachment 71033

Neuromorphic computing has considerable potential for us across many areas.

By mimicking the functionality of the human brain, it can make next-generation AI computation considerably faster and more energy efficient.

One current research project is NAOMI4Radar funded by the German Federal Ministry for Economic Affairs and Climate Action. As consortium leader, we are working with partners to assess how neuromorphic computing can be used to optimise the processing chain for radar data in automated driving systems.

Current Mercedes-Benz models use frontal radar to see 200 metres in front of them. For instance, our DRIVE PILOT system uses radar data as one of its many sources for enabling conditionally automated driving.

The aim of the NAOMI4Radar project is to demonstrate that neuromorphic computing can bring fundamental benefits to future generations of automated and autonomous driving systems.

But as I said, this is just one current research. More on that soon.

View attachment 71034
Loihi 2 - smoke screen or smoked



Over 12 months ago Mercedes served this up fingers crossed our time has arrived



View attachment 71035


My new “In the Loop” series kicks off with #Neuromorphic Computing – the clear winner of my poll a few weeks ago.

For those unfamiliar, this highly significant field of computing strives to emulate the multi-tasking of the human brain. Traditional microprocessors function sequentially. However, as the complexity and scale of calculations sores, this way of doing things is rapidly running out of road.

The idea is not new, but trying to “put a brain on a chip” is a mammoth task. To put it into figures: the human brain has 86-100 billion neurons operating on around 20 watts. Current neural chips from leading developers such as BrainChip and Intel Corporation contain around 1 million neurons and consume roughly 1 watt of power.

So, you see, despite impressive advances, there is still a very long way to go. Neuromorphic computing goes well beyond chip design and includes a specific kind of artificial neural network called #spikingneuralnetworks (SNN). They consume far less energy because the neurons are silent most of the time, only firing (or spiking) when needed for events.

Together with intense parallel execution on neuromorphic chips, the new processing principles require us to go beyond the application of existing #AI frameworks to neuromorphic chips. We have to fundamentally rethink the algorithms that ultimately enable future AI functions in our cars, gathering joint inspiration from machine learning, chip design and neuroscience. Our experts are working closely with our partners to examine their potential in new applications.

The thing is, even a tiny fraction of the thinking capacity of the human brain can go a long way in several fields that are extremely relevant to automotive applications. Examples include advanced driving assistance systems #ADAS as well as the on-board analysis of speech and video data, which can unlock major advances in how we communicate with our cars.

We already made some interesting findings here with our VISION EQXX, where we applied neuromorphic principles to the “Hey Mercedes” hot-word detection. That alone made it five to ten times more energy efficient than conventional voice control. As AI and machine learning take on an increasingly important role in the software-defined vehicle, the energy this consumes is likely to become a critical factor.

I’ll touch on our latest findings in an upcoming “In the Loop” and tell you my thoughts on where this is taking us.

In the meantime, for those of you interested in reading up on neuromorphic computing, check out the slider for my recommended sources. I’ve graded them to ensure there’s something for everyone, from absolute beginner to true geeks.

View attachment 71038

“The NAOMI4Radar project uses the “Loihi 2” Intel chip with up to 1 million neurons and up to 120 million synapses.”

 

[Tothemoon24]

“The NAOMI4Radar project uses the “Loihi 2” Intel chip with up to 1 million neurons and up to 120 million synapses.”

View attachment 71040

Correct ! As I stated in my post

 

[7für7]

View attachment 71033

Neuromorphic computing has considerable potential for us across many areas.

By mimicking the functionality of the human brain, it can make next-generation AI computation considerably faster and more energy efficient.

One current research project is NAOMI4Radar funded by the German Federal Ministry for Economic Affairs and Climate Action. As consortium leader, we are working with partners to assess how neuromorphic computing can be used to optimise the processing chain for radar data in automated driving systems.

Current Mercedes-Benz models use frontal radar to see 200 metres in front of them. For instance, our DRIVE PILOT system uses radar data as one of its many sources for enabling conditionally automated driving.

The aim of the NAOMI4Radar project is to demonstrate that neuromorphic computing can bring fundamental benefits to future generations of automated and autonomous driving systems.

But as I said, this is just one current research. More on that soon.

View attachment 71034
Loihi 2 - smoke screen or smoked



Over 12 months ago Mercedes served this up fingers crossed our time has arrived



View attachment 71035


My new “In the Loop” series kicks off with #Neuromorphic Computing – the clear winner of my poll a few weeks ago.

For those unfamiliar, this highly significant field of computing strives to emulate the multi-tasking of the human brain. Traditional microprocessors function sequentially. However, as the complexity and scale of calculations sores, this way of doing things is rapidly running out of road.

The idea is not new, but trying to “put a brain on a chip” is a mammoth task. To put it into figures: the human brain has 86-100 billion neurons operating on around 20 watts. Current neural chips from leading developers such as BrainChip and Intel Corporation contain around 1 million neurons and consume roughly 1 watt of power.

So, you see, despite impressive advances, there is still a very long way to go. Neuromorphic computing goes well beyond chip design and includes a specific kind of artificial neural network called #spikingneuralnetworks (SNN). They consume far less energy because the neurons are silent most of the time, only firing (or spiking) when needed for events.

Together with intense parallel execution on neuromorphic chips, the new processing principles require us to go beyond the application of existing #AI frameworks to neuromorphic chips. We have to fundamentally rethink the algorithms that ultimately enable future AI functions in our cars, gathering joint inspiration from machine learning, chip design and neuroscience. Our experts are working closely with our partners to examine their potential in new applications.

The thing is, even a tiny fraction of the thinking capacity of the human brain can go a long way in several fields that are extremely relevant to automotive applications. Examples include advanced driving assistance systems #ADAS as well as the on-board analysis of speech and video data, which can unlock major advances in how we communicate with our cars.

We already made some interesting findings here with our VISION EQXX, where we applied neuromorphic principles to the “Hey Mercedes” hot-word detection. That alone made it five to ten times more energy efficient than conventional voice control. As AI and machine learning take on an increasingly important role in the software-defined vehicle, the energy this consumes is likely to become a critical factor.

I’ll touch on our latest findings in an upcoming “In the Loop” and tell you my thoughts on where this is taking us.

In the meantime, for those of you interested in reading up on neuromorphic computing, check out the slider for my recommended sources. I’ve graded them to ensure there’s something for everyone, from absolute beginner to true geeks.

Lohi ? hmmm

 

[Guzzi62]

“The NAOMI4Radar project uses the “Loihi 2” Intel chip with up to 1 million neurons and up to 120 million synapses.”

View attachment 71040

That's a research chip that Loihi2 from Intel.

The Akida chip used in the M.B. project car was for something else, was it the screen control and power management?

So above is just something they are fiddling around with, so years away from being in a car.

Next-Level Neuromorphic Computing: Intel Lab's Loihi 2 Chip

Intel Lab’s Loihi 2 chip delivers outstanding performance & new features, including an open-source, community-driven neuromorphic computing framework.

www.intel.com

 

[Frangipani]

“The NAOMI4Radar project uses the “Loihi 2” Intel chip with up to 1 million neurons and up to 120 million synapses.”

View attachment 71040

This was uploaded on the website of Uni Lübeck, where MB is collaborating with Sebastian Otte (who leads the Adaptive AI research group at the Institute for Robotics and Cognitive Systems) and his doctoral student Saya Higuchi on the NAOMI4Radar project:

Autonomes Fahren: Intelligente Sensoren durch innovative neuronale Netzwerke: Universität zu Lübeck

Universität zu Lübeck. Im Focus das Leben

www.uni-luebeck.de

Autonomes Fahren: Intelligente Sensoren durch innovative neuronale Netzwerke​

Dynamische Darstellung innovativer Radarsensoren und neuronaler Netze für autonomes Fahren (Bild: Anja Stähle, generiert mit Adobe Firefly)

Prof. Sebastian Otte ist am Projekt NAOMI4Radar beteiligt und entwickelt mit Projektpartnern aus Industrie und Hochschulen energieeffiziente Radarsensoren.


Autonome Fahrzeuge benötigen präzise Sensoren für eine schnelle und zuverlässige Umgebungserfassung. Im Projekt NAOMI4Radar arbeitet ein Forschungsteam der Universität zu Lübeck unter der Leitung von Prof. Sebastian Otte gemeinsam mit der Mercedes-Benz AG, TWT GmbH Science & Innovation, Intel Deutschland GmbH und der Technischen Universität München an einer energieeffizienten Radarsensorik. Durch den Einsatz von Neuromorphic Computing und Spiking Neural Networks (SNNs) soll die Batterielaufzeit optimiert, die Reaktionszeit verkürzt und die Sicherheit erhöht werden. Das Projekt wird vom Bundesministerium für Wirtschaft und Klimaschutz (BMWK) gefördert und durch den Projektträger TÜV Rheinland begleitet.

Autonome Fahrzeuge benötigen präzise arbeitenden Sensoren, um schnell und zuverlässig auf ihre Umgebung reagieren zu können. Aktuelle Forschung zielt darauf ab, die Energieeffizienz der Sensordatenverarbeitung zu verbessern, um beispielsweise die Batterielaufzeit zu maximieren und die CO₂-Emissionen zu verringern. Prof. Sebastian Otte vom Institut für Robotik und kognitive Systeme entwickelt mit seinem Team innovative Lösungen in diesem Bereich. Im Projekt NAOMI4Radar arbeitet das Team der Universität zu Lübeck gemeinsam mit der Mercedes-Benz AG, der TWT GmbH Science & Innovation sowie den assoziierten Partnern Intel Deutschland GmbH und der Technischen Universität München an der Optimierung der Radarsensorik für autonome Fahrzeuge durch den Einsatz von Neuromorphic Computing. Diese innovative Technologie orientiert sich an der Arbeitsweise des menschlichen Gehirns und ermöglicht eine energieeffiziente und schnelle Verarbeitung von Sensordaten. Das Lübecker Forschungsteam erhält dafür eine Fördersumme von rund 166.000 Euro.

Neuronale Netze für Radardatenverarbeitung
Die 2024 mit dem Nobelpreis für Physik ausgezeichneten Entwicklungen im Bereich künstlicher neuronaler Netze bilden auch im Projekt NAOMI4Radar eine Schlüsseltechnologie in der Radardatenverarbeitung. Ziel des Projekts ist es, die Radardatenverarbeitung durch Spiking Neural Networks (SNNs), einer speziellen Form neuronaler Netzwerke, effizienter zu gestalten. Im Vergleich zu herkömmlichen KI-Algorithmen bieten SNNs vereinfacht ausgedrückt den Vorteil, dass einzelne Neuronen nur dann aktiv werden, wenn sie tatsächlich gebraucht werden. Durch Einsatz in neuromorphen Prozessoren, wie es im Loihi 2 von Intel vorgesehen ist, kann dieses Potenzial ausgeschöpft werden. Das senkt nicht nur den Energieverbrauch, sondern ermöglicht prinzipiell auch eine schnellere Reaktionszeit der autonomen Fahrzeuge, und erhöht somit die Sicherheit im Straßenverkehr.

Energiesparende Neuronenmodelle
Prof. Otte und sein Team konzentrieren sich dabei insbesondere auf die Weiterentwicklung des Balanced Resonate-and-Fire (BRF) Modells, dessen spezielle Eigenschaften es besonders für die effiziente Verarbeitung von Radardaten interessant macht. Die Effizienz soll durch Verwendung von biologisch inspirierten Sparse Coding Ansätzen noch weiter gesteigert werden. Sparse Coding hat das Ziel, die Robustheit von neuronalen Netzen zu verbessern, um sie beispielsweise fehlertoleranter zu machen. Gleichzeitig wird die Aktivität, also die Menge der Spikes, die im Netzwerk zirkulieren, auf ein Minimum zu reduziert. In Zusammenarbeit mit den Projektpartnern soll eine vollständige Integration von neuromorpher Radardatenverarbeitung realisiert und in einem Prototypenfahrzeug getestet werden.

Das Projekt, das bis August 2025 läuft, wird vom Bundesministerium für Wirtschaft und Klimaschutz (BMWK) gefördert. Die Universität Lübeck bringt ihre Expertise im Bereich Künstlicher Intelligenz und neuromorphe Algorithmen in dieses praxisorientierte Forschungsprojekt ein und leistet damit einen Beitrag für die Entwicklung und Erprobung nachhaltiger KI-Lösungen im industriellen Kontext.

Originalpublikation zum Lübecker BRF-Modell
Higuchi et al. Balanced Resonate-and-Fire Neurons. Proceedings of the 41st International Conference on Machine Learning, Vienna, Austria, 2024.

Kontakt:
Prof. Sebastian Otte
Adaptive AI Forschungsgruppe
Institut für Robotik und Kognitive Systeme
Universität zu Lübeck
Email: sebastian.otte(at)uni-luebeck(dot)de





Autonomous driving: Intelligent sensors through innovative neural networks

Prof. Sebastian Otte is involved in the NAOMI4Radar project and is developing energy-efficient radar sensors with project partners from industry and universities.


Autonomous vehicles require precise sensors for fast and reliable detection of their surroundings. In the NAOMI4Radar project, a research team from the University of Lübeck led by Prof. Sebastian Otte is working together with Mercedes-Benz AG, TWT GmbH Science & Innovation, Intel Deutschland GmbH and the Technical University of Munich on energy-efficient radar sensor technology. The use of neuromorphic computing and spiking neural networks (SNNs) is intended to optimize battery life, shorten reaction times and increase safety. The project is funded by the Federal Ministry for Economic Affairs and Climate Protection (BMWK) and supported by the project sponsor TÜV Rheinland.

Autonomous vehicles require precise sensors in order to react quickly and reliably to their environment. Current research aims to improve the energy efficiency of sensor data processing in order to maximize battery life and reduce CO₂ emissions, for example. Prof. Sebastian Otte from the Institute of Robotics and Cognitive Systems and his team are developing innovative solutions in this area. In the NAOMI4Radar project, the team at the University of Lübeck is working together with Mercedes-Benz AG, TWT GmbH Science & Innovation and associated partners Intel Deutschland GmbH and the Technical University of Munich to optimize radar sensor technology for autonomous vehicles through the use of neuromorphic computing. This innovative technology is based on the way the human brain works and enables energy-efficient and fast processing of sensor data. The Lübeck research team will receive funding of around 166,000 euros.

Neural networks for radar data processing
The developments in the field of artificial neural networks, which were awarded the Nobel Prize in Physics in 2024, are also a key technology in radar data processing in the NAOMI4Radar project. The aim of the project is to make radar data processing more efficient using spiking neural networks (SNNs), a special form of neural network. In comparison to conventional AI algorithms, SNNs offer the advantage that individual neurons only become active when they are actually needed. This potential can be exploited by using them in neuromorphic processors, as envisaged in Intel's Loihi 2. This not only reduces energy consumption, but in principle also enables autonomous vehicles to react more quickly, thereby increasing road safety.

Energy-saving neuron models
Prof. Otte and his team are focusing in particular on the further development of the Balanced Resonate-and-Fire (BRF) model, whose special properties make it particularly interesting for the efficient processing of radar data. Efficiency is to be increased even further by using biologically inspired sparse coding approaches. Sparse coding aims to improve the robustness of neural networks, for example to make them more fault-tolerant.

At the same time, the activity, i.e. the amount of spikes circulating in the network, is reduced to a minimum. In collaboration with the project partners, a complete integration of neuromorphic radar data processing is to be realized and tested in a prototype vehicle.

The project, which will run until August 2025, is funded by the Federal Ministry for Economic Affairs and Climate Protection (BMWK). The University of Lübeck is contributing its expertise in the field of artificial intelligence and neuromorphic algorithms to this practice-oriented research project, thereby contributing to the development and testing of sustainable AI solutions in an industrial context.

Original publication on the Lübeck BRF model
Higuchi et al. Balanced Resonate-and-Fire Neurons. Proceedings of the 41st International Conference on Machine Learning, Vienna, Austria, 2024.


Contact:
Prof. Sebastian Otte
Adaptive AI Research Group
Institute for Robotics and Cognitive Systems
University of Lübeck
Email: sebastian.otte(at)uni-luebeck(dot)de


(Translated by DeepL)

 

[Bravo]

View attachment 71033

Neuromorphic computing has considerable potential for us across many areas.

By mimicking the functionality of the human brain, it can make next-generation AI computation considerably faster and more energy efficient.

One current research project is NAOMI4Radar funded by the German Federal Ministry for Economic Affairs and Climate Action. As consortium leader, we are working with partners to assess how neuromorphic computing can be used to optimise the processing chain for radar data in automated driving systems.

Current Mercedes-Benz models use frontal radar to see 200 metres in front of them. For instance, our DRIVE PILOT system uses radar data as one of its many sources for enabling conditionally automated driving.

The aim of the NAOMI4Radar project is to demonstrate that neuromorphic computing can bring fundamental benefits to future generations of automated and autonomous driving systems.

But as I said, this is just one current research. More on that soon.

View attachment 71034
Loihi 2 - smoke screen or smoked



Over 12 months ago Mercedes served this up fingers crossed our time has arrived



View attachment 71035


My new “In the Loop” series kicks off with #Neuromorphic Computing – the clear winner of my poll a few weeks ago.

For those unfamiliar, this highly significant field of computing strives to emulate the multi-tasking of the human brain. Traditional microprocessors function sequentially. However, as the complexity and scale of calculations sores, this way of doing things is rapidly running out of road.

The idea is not new, but trying to “put a brain on a chip” is a mammoth task. To put it into figures: the human brain has 86-100 billion neurons operating on around 20 watts. Current neural chips from leading developers such as BrainChip and Intel Corporation contain around 1 million neurons and consume roughly 1 watt of power.

So, you see, despite impressive advances, there is still a very long way to go. Neuromorphic computing goes well beyond chip design and includes a specific kind of artificial neural network called #spikingneuralnetworks (SNN). They consume far less energy because the neurons are silent most of the time, only firing (or spiking) when needed for events.

Together with intense parallel execution on neuromorphic chips, the new processing principles require us to go beyond the application of existing #AI frameworks to neuromorphic chips. We have to fundamentally rethink the algorithms that ultimately enable future AI functions in our cars, gathering joint inspiration from machine learning, chip design and neuroscience. Our experts are working closely with our partners to examine their potential in new applications.

The thing is, even a tiny fraction of the thinking capacity of the human brain can go a long way in several fields that are extremely relevant to automotive applications. Examples include advanced driving assistance systems #ADAS as well as the on-board analysis of speech and video data, which can unlock major advances in how we communicate with our cars.

We already made some interesting findings here with our VISION EQXX, where we applied neuromorphic principles to the “Hey Mercedes” hot-word detection. That alone made it five to ten times more energy efficient than conventional voice control. As AI and machine learning take on an increasingly important role in the software-defined vehicle, the energy this consumes is likely to become a critical factor.

I’ll touch on our latest findings in an upcoming “In the Loop” and tell you my thoughts on where this is taking us.

In the meantime, for those of you interested in reading up on neuromorphic computing, check out the slider for my recommended sources. I’ve graded them to ensure there’s something for everyone, from absolute beginner to true geeks.

It'll be very interesting to see how this plays out.

As one person noted in the comments on Markus Shafer's Linkedin post, the first mover will likely gain the advantage, i.e. the greatest market share.

What we do know is that at present Loihi2 is still a research chip and in the Smarter Cars article sited below dated December 2020 Mike Davies stated he thought commercialisation of Loihi2 would be 5 years from the date of that article, which is approximately December 2025.

If we compare the Loihi2 with AKIDA we can see AKIDA has the edge, no pun intended on Loihi, in terms of the higher number of neurons and synapses.

AKIDA: 1.2 million neurons and 10 billion synapses VERSUS
Liohi2: 1 million neurons and 120 million synapses

And obviously BrainChip has a commercially available product which Intel doesn't at this particular point in time.

 

[yogi]

Gerrit Ecke on LinkedIn: #neuromorphic #adas #ad

Mercedes-Benz invests in #Neuromorphic Computing! The technology is key to overcome the issue of increasing compute requirements on a tight energy budget of…

www.linkedin.com

View attachment 70974 View attachment 70975 View attachment 70976

Looks like they went with Loihi on this Markus just posted

 

[Diogenese]

View attachment 71033

Neuromorphic computing has considerable potential for us across many areas.

By mimicking the functionality of the human brain, it can make next-generation AI computation considerably faster and more energy efficient.

One current research project is NAOMI4Radar funded by the German Federal Ministry for Economic Affairs and Climate Action. As consortium leader, we are working with partners to assess how neuromorphic computing can be used to optimise the processing chain for radar data in automated driving systems.

Current Mercedes-Benz models use frontal radar to see 200 metres in front of them. For instance, our DRIVE PILOT system uses radar data as one of its many sources for enabling conditionally automated driving.

The aim of the NAOMI4Radar project is to demonstrate that neuromorphic computing can bring fundamental benefits to future generations of automated and autonomous driving systems.

But as I said, this is just one current research. More on that soon.

View attachment 71034
Loihi 2 - smoke screen or smoked



Over 12 months ago Mercedes served this up fingers crossed our time has arrived



View attachment 71035


My new “In the Loop” series kicks off with #Neuromorphic Computing – the clear winner of my poll a few weeks ago.

For those unfamiliar, this highly significant field of computing strives to emulate the multi-tasking of the human brain. Traditional microprocessors function sequentially. However, as the complexity and scale of calculations sores, this way of doing things is rapidly running out of road.

The idea is not new, but trying to “put a brain on a chip” is a mammoth task. To put it into figures: the human brain has 86-100 billion neurons operating on around 20 watts. Current neural chips from leading developers such as BrainChip and Intel Corporation contain around 1 million neurons and consume roughly 1 watt of power.

So, you see, despite impressive advances, there is still a very long way to go. Neuromorphic computing goes well beyond chip design and includes a specific kind of artificial neural network called #spikingneuralnetworks (SNN). They consume far less energy because the neurons are silent most of the time, only firing (or spiking) when needed for events.

Together with intense parallel execution on neuromorphic chips, the new processing principles require us to go beyond the application of existing #AI frameworks to neuromorphic chips. We have to fundamentally rethink the algorithms that ultimately enable future AI functions in our cars, gathering joint inspiration from machine learning, chip design and neuroscience. Our experts are working closely with our partners to examine their potential in new applications.

The thing is, even a tiny fraction of the thinking capacity of the human brain can go a long way in several fields that are extremely relevant to automotive applications. Examples include advanced driving assistance systems #ADAS as well as the on-board analysis of speech and video data, which can unlock major advances in how we communicate with our cars.

We already made some interesting findings here with our VISION EQXX, where we applied neuromorphic principles to the “Hey Mercedes” hot-word detection. That alone made it five to ten times more energy efficient than conventional voice control. As AI and machine learning take on an increasingly important role in the software-defined vehicle, the energy this consumes is likely to become a critical factor.

I’ll touch on our latest findings in an upcoming “In the Loop” and tell you my thoughts on where this is taking us.

In the meantime, for those of you interested in reading up on neuromorphic computing, check out the slider for my recommended sources. I’ve graded them to ensure there’s something for everyone, from absolute beginner to true geeks.

Great find TTM,

Intel continue to refer to Loihi 2 as a research chip:

Intel Labs’ new Loihi 2 research chip outperforms its predecessor by up to 10x and comes with an open-source, community-driven neuromorphic computing framework.

https://download.intel.com/newsroom/2021/new-technologies/neuromorphic-computing-loihi-2-brief.pdf

Wayback shows the latest revision as 20240924.

I haven't found any information indicating Loihi 2 is in commercial production.

Obviously Intel will also have simulation software for Loihi 2, so we can't definitively rule it out, but they don't have TENNS - advantage BRN.

This is not an in-house MB project. It is government funded. I'm sure MB can carry two thoughts at the same time.

 

[Frangipani]

A video going along with that paper was uploaded to YouTube yesterday:



Both paper and video relate to another paper and video published by the same Uni Tübingen authors earlier this year. At a cursory glance, at least the videos (posted about six months apart) appear to be VERY similar:

https://thestockexchange.com.au/threads/brn-discussion-ongoing.1/post-416900


View attachment 70372
View attachment 70373


Now compare the slides to those in the video uploaded October 3:

View attachment 70368


View attachment 70369

View attachment 70370

In fact, when I just tried to cursorily compare the new paper to the March 15 paper that @Fullmoonfever had linked at the time (https://thestockexchange.com.au/threads/brn-discussion-ongoing.1/post-416313), I discovered that the link he had posted then now connects directly to this new paper, published on September 16, so it seems to be an updated version of the previous paper.

I did notice the addition of another co-author, though: Sebastian Otte, who used to be a PhD student and postdoc at Uni Tübingen (2013-2023) and became Professor at Uni Lübeck’s Institute for Robotics and Cognitive Systems just over a year ago, where he heads the Adaptive AI research group.

To put the results that our competitors’ neuromorphic offerings fared worse in the benchmarking tests alongside Akida somewhat into perspective:
In all fairness, it should be highlighted that Akida’s superiority was at least partly due to the fact that AKD1000 is available as a PCIe Board, whereas SynSense’s DynapCNN was connected to the PC via USB and - as the excerpt Gazzafish already posted shows - the researchers did not have direct access to a Loihi 2 edge device, but merely through a virtual machine provided by Intel via their Neuromorphic Research Cloud. The benchmarking would obviously yield better comparable results if the actual hardware used were of a similar form factor:

“Our results show that the better a neuromorphic edge device is connected to the main compute unit, e.g., as a PCIe card, the better the overall run-time.”


Anyway, Akida undoubtedly impressed the researchers, and as a result they are considering further experiments: “(…) future work could involve evaluating the system with an additional Akida PCIe card.”


View attachment 70374

---

In an earlier post (https://thestockexchange.com.au/threads/brn-discussion-ongoing.1/post-426404), I had already mentioned that the paper’s first author, Andreas Ziegler, who is doing a PhD in robotics and computer vision at Uni Tübingen, has meanwhile completed his internship at Sony AI in Switzerland (that - as we know - partially funded the paper’s research):

View attachment 70375


Fun fact: One of his co-authors, Karl Vetter, however, is no longer with Uni Tübingen’s Cognitive Systems Lab, but has since moved to France, where he has been working as a research engineer for…

Neurobus for the past three months!
It’s a small world, isn’t it?!

View attachment 70376
View attachment 70377

I had posted about Sebastian Otte being a newly added co-author in the updated Uni Tübingen table tennis robot paper just ten days ago.

To those constantly dissing our competition: Believe it or not, there are actually neuromorphic researchers who appreciate both Akida and Loihi:

Those that feel the need to comment on LinkedIn, please make sure you use the correct spelling of our company’s name (Brianchip?!) and of technical terms (neurophonic?!), as this kind of publicity surely doesn’t help us in the public eye.

 

[Tothemoon24]

It'll be very interesting to see how this plays out.

As one person noted in the comments on Markus Shafer's Linkedin post, the first mover will likely gain the advantage, i.e. the greatest market share.

What we do know is that at present Loihi2 is still a research chip and in the Smarter Cars article sited below dated December 2020 Mike Davies stated he thought commercialisation of Loihi2 would be 5 years from the date of that article, which is approximately December 2025.

If we compare the Loihi2 with AKIDA we can see AKIDA has the edge, no pun intended on Loihi, in terms of the higher number of neurons and synapses.

And obviously BrainChip has a commercially available product which Intel doesn't at this particular point in time.


View attachment 71045

View attachment 71046
View attachment 71047
View attachment 71048

Hi Bravo yes very interesting & I see the post is causing much interest, from my point of view there’s a comment amongst it that may of been missed by some

 

[Rach2512]

This was uploaded on the website of Uni Lübeck, where MB is collaborating with Sebastian Otte (who leads the Adaptive AI research group at the Institute for Robotics and Cognitive Systems) and his doctoral student Saya Higuchi on the NAOMI4Radar project:

Autonomes Fahren: Intelligente Sensoren durch innovative neuronale Netzwerke: Universität zu Lübeck

Universität zu Lübeck. Im Focus das Leben

www.uni-luebeck.de

Autonomes Fahren: Intelligente Sensoren durch innovative neuronale Netzwerke​

View attachment 71041

Dynamische Darstellung innovativer Radarsensoren und neuronaler Netze für autonomes Fahren (Bild: Anja Stähle, generiert mit Adobe Firefly)

Prof. Sebastian Otte ist am Projekt NAOMI4Radar beteiligt und entwickelt mit Projektpartnern aus Industrie und Hochschulen energieeffiziente Radarsensoren.


Autonome Fahrzeuge benötigen präzise Sensoren für eine schnelle und zuverlässige Umgebungserfassung. Im Projekt NAOMI4Radar arbeitet ein Forschungsteam der Universität zu Lübeck unter der Leitung von Prof. Sebastian Otte gemeinsam mit der Mercedes-Benz AG, TWT GmbH Science & Innovation, Intel Deutschland GmbH und der Technischen Universität München an einer energieeffizienten Radarsensorik. Durch den Einsatz von Neuromorphic Computing und Spiking Neural Networks (SNNs) soll die Batterielaufzeit optimiert, die Reaktionszeit verkürzt und die Sicherheit erhöht werden. Das Projekt wird vom Bundesministerium für Wirtschaft und Klimaschutz (BMWK) gefördert und durch den Projektträger TÜV Rheinland begleitet.

Autonome Fahrzeuge benötigen präzise arbeitenden Sensoren, um schnell und zuverlässig auf ihre Umgebung reagieren zu können. Aktuelle Forschung zielt darauf ab, die Energieeffizienz der Sensordatenverarbeitung zu verbessern, um beispielsweise die Batterielaufzeit zu maximieren und die CO₂-Emissionen zu verringern. Prof. Sebastian Otte vom Institut für Robotik und kognitive Systeme entwickelt mit seinem Team innovative Lösungen in diesem Bereich. Im Projekt NAOMI4Radar arbeitet das Team der Universität zu Lübeck gemeinsam mit der Mercedes-Benz AG, der TWT GmbH Science & Innovation sowie den assoziierten Partnern Intel Deutschland GmbH und der Technischen Universität München an der Optimierung der Radarsensorik für autonome Fahrzeuge durch den Einsatz von Neuromorphic Computing. Diese innovative Technologie orientiert sich an der Arbeitsweise des menschlichen Gehirns und ermöglicht eine energieeffiziente und schnelle Verarbeitung von Sensordaten. Das Lübecker Forschungsteam erhält dafür eine Fördersumme von rund 166.000 Euro.

Neuronale Netze für Radardatenverarbeitung
Die 2024 mit dem Nobelpreis für Physik ausgezeichneten Entwicklungen im Bereich künstlicher neuronaler Netze bilden auch im Projekt NAOMI4Radar eine Schlüsseltechnologie in der Radardatenverarbeitung. Ziel des Projekts ist es, die Radardatenverarbeitung durch Spiking Neural Networks (SNNs), einer speziellen Form neuronaler Netzwerke, effizienter zu gestalten. Im Vergleich zu herkömmlichen KI-Algorithmen bieten SNNs vereinfacht ausgedrückt den Vorteil, dass einzelne Neuronen nur dann aktiv werden, wenn sie tatsächlich gebraucht werden. Durch Einsatz in neuromorphen Prozessoren, wie es im Loihi 2 von Intel vorgesehen ist, kann dieses Potenzial ausgeschöpft werden. Das senkt nicht nur den Energieverbrauch, sondern ermöglicht prinzipiell auch eine schnellere Reaktionszeit der autonomen Fahrzeuge, und erhöht somit die Sicherheit im Straßenverkehr.

Energiesparende Neuronenmodelle
Prof. Otte und sein Team konzentrieren sich dabei insbesondere auf die Weiterentwicklung des Balanced Resonate-and-Fire (BRF) Modells, dessen spezielle Eigenschaften es besonders für die effiziente Verarbeitung von Radardaten interessant macht. Die Effizienz soll durch Verwendung von biologisch inspirierten Sparse Coding Ansätzen noch weiter gesteigert werden. Sparse Coding hat das Ziel, die Robustheit von neuronalen Netzen zu verbessern, um sie beispielsweise fehlertoleranter zu machen. Gleichzeitig wird die Aktivität, also die Menge der Spikes, die im Netzwerk zirkulieren, auf ein Minimum zu reduziert. In Zusammenarbeit mit den Projektpartnern soll eine vollständige Integration von neuromorpher Radardatenverarbeitung realisiert und in einem Prototypenfahrzeug getestet werden.

Das Projekt, das bis August 2025 läuft, wird vom Bundesministerium für Wirtschaft und Klimaschutz (BMWK) gefördert. Die Universität Lübeck bringt ihre Expertise im Bereich Künstlicher Intelligenz und neuromorphe Algorithmen in dieses praxisorientierte Forschungsprojekt ein und leistet damit einen Beitrag für die Entwicklung und Erprobung nachhaltiger KI-Lösungen im industriellen Kontext.

Originalpublikation zum Lübecker BRF-Modell
Higuchi et al. Balanced Resonate-and-Fire Neurons. Proceedings of the 41st International Conference on Machine Learning, Vienna, Austria, 2024.

Kontakt:
Prof. Sebastian Otte
Adaptive AI Forschungsgruppe
Institut für Robotik und Kognitive Systeme
Universität zu Lübeck
Email: sebastian.otte(at)uni-luebeck(dot)de





Autonomous driving: Intelligent sensors through innovative neural networks

View attachment 71041

Prof. Sebastian Otte is involved in the NAOMI4Radar project and is developing energy-efficient radar sensors with project partners from industry and universities.


Autonomous vehicles require precise sensors for fast and reliable detection of their surroundings. In the NAOMI4Radar project, a research team from the University of Lübeck led by Prof. Sebastian Otte is working together with Mercedes-Benz AG, TWT GmbH Science & Innovation, Intel Deutschland GmbH and the Technical University of Munich on energy-efficient radar sensor technology. The use of neuromorphic computing and spiking neural networks (SNNs) is intended to optimize battery life, shorten reaction times and increase safety. The project is funded by the Federal Ministry for Economic Affairs and Climate Protection (BMWK) and supported by the project sponsor TÜV Rheinland.

Autonomous vehicles require precise sensors in order to react quickly and reliably to their environment. Current research aims to improve the energy efficiency of sensor data processing in order to maximize battery life and reduce CO₂ emissions, for example. Prof. Sebastian Otte from the Institute of Robotics and Cognitive Systems and his team are developing innovative solutions in this area. In the NAOMI4Radar project, the team at the University of Lübeck is working together with Mercedes-Benz AG, TWT GmbH Science & Innovation and associated partners Intel Deutschland GmbH and the Technical University of Munich to optimize radar sensor technology for autonomous vehicles through the use of neuromorphic computing. This innovative technology is based on the way the human brain works and enables energy-efficient and fast processing of sensor data. The Lübeck research team will receive funding of around 166,000 euros.

Neural networks for radar data processing
The developments in the field of artificial neural networks, which were awarded the Nobel Prize in Physics in 2024, are also a key technology in radar data processing in the NAOMI4Radar project. The aim of the project is to make radar data processing more efficient using spiking neural networks (SNNs), a special form of neural network. In comparison to conventional AI algorithms, SNNs offer the advantage that individual neurons only become active when they are actually needed. This potential can be exploited by using them in neuromorphic processors, as envisaged in Intel's Loihi 2. This not only reduces energy consumption, but in principle also enables autonomous vehicles to react more quickly, thereby increasing road safety.

Energy-saving neuron models
Prof. Otte and his team are focusing in particular on the further development of the Balanced Resonate-and-Fire (BRF) model, whose special properties make it particularly interesting for the efficient processing of radar data. Efficiency is to be increased even further by using biologically inspired sparse coding approaches. Sparse coding aims to improve the robustness of neural networks, for example to make them more fault-tolerant.

At the same time, the activity, i.e. the amount of spikes circulating in the network, is reduced to a minimum. In collaboration with the project partners, a complete integration of neuromorphic radar data processing is to be realized and tested in a prototype vehicle.

The project, which will run until August 2025, is funded by the Federal Ministry for Economic Affairs and Climate Protection (BMWK). The University of Lübeck is contributing its expertise in the field of artificial intelligence and neuromorphic algorithms to this practice-oriented research project, thereby contributing to the development and testing of sustainable AI solutions in an industrial context.

Original publication on the Lübeck BRF model
Higuchi et al. Balanced Resonate-and-Fire Neurons. Proceedings of the 41st International Conference on Machine Learning, Vienna, Austria, 2024.


Contact:
Prof. Sebastian Otte
Adaptive AI Research Group
Institute for Robotics and Cognitive Systems
University of Lübeck
Email: sebastian.otte(at)uni-luebeck(dot)de


(Translated by DeepL)



View attachment 71043

Bravo, didn't you make a connection between the Nobel Prize winner and the use of Akida? Sorry if someone has already pointed this out. I'm playing catchup.

 

[7für7]

Hi Bravo yes very interesting & I see the post is causing much interest, from my point of view there’s a comment amongst it that may of been missed by some

View attachment 71049

Everything is a research regarding our technology no products launched. So nothing wrong about it to post. Thanks @Bravo

 

[Perhaps]

Speaking of EDGX:
I am somewhat surprised no one has yet commented on the fact that EDGX no longer seems to be in an exclusive relationship with us as their neuromorphic partner:

View attachment 70844

Some posters will want to make you believe that as soon as a company / research institution / consultancy has discovered us, they will only have eyes for us, and that the competition can basically pack up and go home. It is a romantic notion for sure, but alas it is not the reality. The companies and institutions truly convinced of the benefits of neuromorphic technology will often be taking their time to explore different solutions and may end up doing business with / recommending (in the case of a consultancy) either
a) us
b) us and someone else or
c) someone else [as unimaginable that may seem to certain posters here].


While Accenture did praise Akida earlier this year, they continue to research Loihi (
https://thestockexchange.com.au/threads/brn-discussion-ongoing.1/post-428774) and have also been evaluating SynSense’s ultra-low power offerings:

View attachment 70909

Or take ESA, for example: Laurent Hili didn’t restrict himself to visiting the BrainChip booth at the AI Hardware & Edge AI Summit in September: He and his colleague Luis Mansilla Garcia (who were both guests on Episode 31 of the BrainChip This is Our Mission podcast in March) also dropped by other AI chip companies’ booths such as that of Intel (-> Gaudi 3) and SpinnCloud Systems ( -> SpiNNaker 2), as evidenced by these recent screenshots I took of photos he posted resp. reposted on LinkedIn:

View attachment 70846

View attachment 70971

Another example:
We know the neuromorphic researchers from TCS to be BrainChip fans.
Yet, a month ago, in the comment section underneath one of his own posts, Sounak Dey from TCS expressed his regret of having missed the chance to meet up with Petrut Antoniu Bogdan from Innatera at Semicon India 2024 (Sept 11-13). No surprise, really, given that in recent months Sounak Dey has liked numerous posts by both BrainChip and Innatera.

View attachment 70848


Of course our competitors are in the same situation, with BrainChip showing up in unexpected places - so standing still is not an option, all those companies need to continually innovate, and BrainChip is doing just that. Having chosen to go the path of an IP company may pay out in the long run, but of course means leaving part of the addressable market to our competitors.

I’d be very cautious to quantify any lead in months or even years, like some posters have done and still do, despite having no insight whatsoever into the negotiations between any of the companies offering neuromorphic technology and their potential customers - in my opinion, such posts lull us into a false sense of security, which in turn could lead to further disappointment among already disappointed shareholders and provide more fodder for the downrampers should one of our competitors land a juicy contract first, especially in case it concerned one that BrainChip had also been vying for.

And in case you were wondering: No, I don’t have any insider information. I am just a keen observer (such as taking note of LinkedIn posts like the ones above or below), and prefer to draw my own conclusions rather than rely on contributions by anonymous shareholders wearing rose-coloured glasses or deliberately cherry-picking info or even twisting the truth to suit their narrative (be it negative or positive - this happens on both ends of the spectrum). And I encourage everyone to do the same (which admittedly is hard to do for many with very limited time to spare.)


View attachment 70893

Reading between the lines: We are also exploring other companies’ offerings and won’t make any promises.


View attachment 70894

Reading between the lines: We are also exploring other companies’ offerings and won’t make any promises.

View attachment 70943
View attachment 70944

View attachment 70896

No reading between the lines is necessary here, I’d say...
They just don’t spell it out with the words: “You’re in good company” or “Trusted by…”, but to me this is essentially saying the same thing, even though the folks at Innatera cannot pride themselves to already have had their tech publicly validated in an MB concept car.

Best post in this forum for months.

 

[IloveLamp]

Looks like they went with Loihi on this Markus just posted

True but Loihi is a research chip and NOT AVAILABLE commercially.....

Mercedes have been posting about neuromorphic A LOT lately. Which other neuromorphic player might they also be engaged with?

I don't mind if they post about loihi in this instance, the more about neuromorphic the better imo, as long as we win in the end, and if "loihi" is our "competition " then that's ok with me.

 

[Pom down under]

“The NAOMI4Radar project uses the “Loihi 2” Intel chip with up to 1 million neurons and up to 120 million synapses.”

View attachment 71040

 

[7für7]

Best post in this forum for months.

Interesting… the whole time there are so many amazing news and progress with our company… no reaction from your side. And suddenly, you find this is one of the best postings for months ? congrats

 

[Diogenese]

Great find TTM,

Intel continue to refer to Loihi 2 as a research chip:

Intel Labs’ new Loihi 2 research chip outperforms its predecessor by up to 10x and comes with an open-source, community-driven neuromorphic computing framework.

https://download.intel.com/newsroom/2021/new-technologies/neuromorphic-computing-loihi-2-brief.pdf

Wayback shows the latest revision as 20240924.

I haven't found any information indicating Loihi 2 is in commercial production.

Obviously Intel will also have simulation software for Loihi 2, so we can't definitively rule it out, but they don't have TENNS - advantage BRN.

This is not an in-house MB project. It is government funded. I'm sure MB can carry two thoughts at the same time.

I think we can adapt the old adage "No one ever got fired for buying IBM" to this project.

 

[Bravo]

I had posted about Sebastian Otte being a newly added co-author in the updated Uni Tübingen table tennis robot paper just ten days ago.

To those constantly dissing our competition: Believe it or not, there are actually neuromorphic researchers who appreciate both Akida and Loihi:

View attachment 71044

Those that feel the need to comment on LinkedIn, please make sure you use the correct spelling of our company’s name (Brianchip?!) and of technical terms (neurophonic?!), as this kind of publicity surely doesn’t help us in the public eye.

Hi @Frangipani,

"BOTH ARE EXCELLENT TECHNOLOGIES!"

Our Aussie battler, BrainChip being quite favourably compared to the technological giant that is Intel.

 

[Humble Genius]

True but Loihi is a research chip and NOT AVAILABLE commercially.....

Mercedes have been posting about neuromorphic A LOT lately. Which other neuromorphic player might they also be engaged with?

I don't mind if they post about loihi in this instance, the more about neuromorphic the better imo, as long as we win in the end, and if "loihi" is our "competition " then that's ok with me.

True but Loihi is a research chip and NOT AVAILABLE commercially.....

I hope I'm wrong, but herein lies the problem with all of our thinking. The Loihi chip has been a "research" chip for quite some time. What differentiates a research chip from a production chip? Whats stopping intel from calling their chip "production ready"? I don't think it would be all that hard for intel to go into mass production with Loihi chip at a moments notice, considering all the resources they have. They've probably just made a decision to continue calling it a research chip as a decoy? Who knows? I hope I'm wrong.

 

[Diogenese]

Can't get into the new Espacenet web site so reverted to old version.

This is a recent Valeo patent which uses a NN algorithm to monitor moving objects TENNS can do this.

WO2024160994 (A1) - METHOD OF RECONSTRUCTING A PROPERTY OF AN IMAGE AND COMPUTING DEVICE 20230202



The present application relates to a Method of reconstructing a property of an image in a sequence of images with a spiking neural network. The method comprises: obtaining raw data of the image (RDt), calculating a similarity measure using raw data of the image (RDt) and raw data of a previous image (RDt-1) preceding the image in the sequence, if the similarity measure exceeds a predefined threshold: - generating a spike train (ST) for the spiking neural network, the spike train (ST) comprising data encoding differential raw data, wherein the differential raw data (DRD) has been calculated using raw data of the im-age and raw data of the previous image (RDt, RDt-1), - inputting the spike train (ST) to the spiking neural network, - reconstructing the property of the image with the spiking neural network, - outputting the property of the image. The application further relates to a computing device (34), a lidar system (30) comprising the computing device (34) and a vehicle (100) comprising the lidar system (30).
...

The spikes train ST is input to the spiking neural network in step 22 (fig. 1). in step 24, a machine learning algorithm based on the spiking neural networks is executed to reconstruct the depth image RIt based on the quantized differential depth image raw data DRD between image and previous image.

The advantage of embodiments of the method is that depth reconstruction may be performed based on the difference between depth image raw data at two moments in time.

...
The method allows to redesign the processing in camera devices, in particular lidar systems, to use neuromorphic processors and spiking neural networks to create a camera device with event based processing. The method is particularly applicable to direct or indirect time of flight systems of flash type.
...
The control, monitoring and/or evaluation tasks performed may for example be implemented as software executing instructions of an algorithm.

The bit about redesigning the camera is interesting as Valeo released their VSS 360 camera with integrated processor earlier this year.
https://www.valeo.com/en/smart3-equipped-with-valeo-smart-safety-360-receives-5-stars-at-euroncap/

Anyone for TENNiS?