BRN Discussion Ongoing

[Space Cadet]

As in politics, religion, society and all human constructs, corruption is inherent in the system.
Doesn't mean that its either all bad or that it cannot be improved, but big change is unlikely in my view.
The system is well protected by those, behind those, whose hands are on the levers, who gain directly the most, from the current settings.
Bit like the state of our cockeyed housing market.
Any government trying to initiate any meaningful change will be crucified at the polls by the majority of voters who want their primary asset to keep appreciating as opposed to the minority currently struggling for a foothold in the market.
The mistake was designating housing not as necessary and required shelter but as wealth enhancement, and then designing and entrenching systems which perverted the tax function towards that end.

Like or loath Jacqui Lambie, she came up with one of the best opinions I've seen. She was talking about negative gearing in real estate and was being shouted down from other pollies. She pointed put that the pollie doing the most winging had about 20 houses negative geared. She said fair enough to have 1 or 2 houses that had tax incentives but 20 was taking the piss, pretty much.

SC

 

[McHale]

https://www.intel.com/content/dam/www/central-libraries/us/en/documents/neuromorphic-computing-loihi-2-brief.pdf

FF shared the above article over on the other site.

Interesting comment in the article: One of the biggest challenges holding back the commercialization of neuromorphic technology is the lack of software maturity and convergence. Since neuromorphic architecture is fundamentally incompatible with standard programming models, including today’s machine-learning and AI frameworks in wide use, neuromorphic software and application development is often fragmented across research teams, with different groups taking different approaches and often reinventing common functionality. Yet to emerge is a single, common software framework for neuromorphic computing that supports the full range of approaches pursued by the research community that presents compelling and productive abstractions to application developers.

Dio do you think our software assists in bridging the gap they are referring to?

Hi @JB49, as I understand it BRN are using several programming models to support the various iterations of Akida, but have wondered myself why a single format could not be implemented/devised.

But the main reason I'm replying is to say that the video you put up Tuesday (post #90,294) regarding Intel and Loihi is quite definitive, with Mike Davies saying that his view is that: Loihi "Is quite some time away" with an emphasis on the "quite" - which made it very clear that it will be no time soon.

Your post came after @Bravo had posted (#90,274) regarding Mercedes working on a project with Intel, being the Naomi4 project which also included working with Univ of Waterloo Canada and Prof Chris Eliasmith, I was initially disappointed, but on reflection realized this is a current research project.

The important fact is that BRN have been partnered with Mercedes for several years and Akida is the only neuromorphic chip that is ready to go. However after watching the video you put up, I don't think anyone will be putting Loihi into a product any time soon, regardless there will be a lot of businesses who will be looking at Loihi because Intel have such a pre-eminent position in chip making.

Which of course goes back to the challenge facing BRN with regard to introducing a disruptive new technology in a business environment dominated by titans like Intel et al.

 

[itsol4605]

I suspect that Intel benefits from German economic development.
An Intel factory should be built in Germany, funded by the German Ministry of Economic Affairs.

Of course, Mercedes will also benefit from this and very much welcome the financial support.
Unfortunately, Brainchip does not have this advantage of the German subsidy.

As is often the case, it's not about better technology, but about many other factors.

 

[Deadpool]

Link does not work anymore

I just tried and it works for me, maybe your cache needs a tidy up?

 

[itsol4605]

I just tried and it works for me, maybe your cache needs a tidy up?

Now it is working again. I have tried it with several browsers when the page was not available and with not one browser it worked.
Maybe the server had a problem.

 

[cosors]

off topic
During my regular rounds for TLG through Swedish press I found this anecdote. A friend of mine had only thrown away 8 at the time.)

"Oops – he happened to throw bitcoin worth billions at the dump​

Just over ten years ago, a hard drive with thousands of bitcoins ended up in landfills. Now a British man is waging a desperate battle against his hometown to try to get them back.​

16 October 2024
Some decisions can be regretted for the rest of your life. Just ask James Howells, a computer engineer who lives in the Welsh city of Newport in the UK.
Just over ten years ago, Howells and his then-partner decided to throw away various broken gadgets from his home office. One of the things that went into a black garbage bag was one of several hard drives that collected dust in a desk drawer.

Had 7,500 bitcoin in crypto wallet​

The next day, the partner went to the city's landfill. James Howells didn't think any more about it – but that would change.
As a technology enthusiast, Howells had taken an early interest in the cryptocurrency bitcoin. In a couple of years, his holdings had grown to 7,500 bitcoin. He stored these in a crypto wallet in the form of a hard drive that was in his office.
In the fall of 2013, his digital treasure cache was worth the equivalent of just over $14.6 million. At that time, James Howells began to suspect trouble. What had actually happened to the storage unit that contained all the bitcoins?
To his dismay, Howells discovered that he had accidentally mixed up his hard drives during the deep cleaning. The one with all the bitcoins was no longer at home – but had ended up in the dump, writes The New Yorker .

Must go through 100 tons of rubbish ​

Since then, James Howells has waged an increasingly desperate battle, trying to convince Newport City Council to dig up the landfill to snoop on the hard drive. According to his own statement, Howells has offered the city to put together a team that can dig through 100 tons of garbage.
He has also promised a juicy ransom: If he gets his way and the storage device is found, the city will pay ten percent of what the crypto coins are worth.
Newport has so far ignored James Howell's repeated applications. Therefore, he has now decided to sue the city for just over SEK 6 billion – a sum that corresponded to the value of SEK 7,500 when the application was submitted.

"The municipality is sitting on my stolen property"​

According to Newport's municipal council, the work would cost at least SEK 135 million and is estimated to take between 18 and 36 months to complete. After the excavation, it would also require at least another year of work to restore the soil on and around the soptation.
"The municipal executive board has made it clear to James Howells several times that the excavation is impossible to carry out. It would also cause a major environmental impact on surrounding areas," says a spokesperson from the city of Newsport in a comment.
But Howells doesn't give up. If the hard drive were to be found, he estimates that there is still an 80 percent chance that the digital coins can be retrieved from it.
"I see it as the municipality has been sitting on my stolen property for ten years," James Howells told The Sun ."

Hoppsan – han råkade slänga bitcoin värda miljarder på tippen

För drygt tio år sedan hamnade en hårddisk med tusentals bitcoin på soptippen. Nu för en brittisk man en desperat kamp mot sin hemstad för att försöka få dem tillbaka.

www.nyteknik.se

Ooops - $503,928,825 thrown away

 

[DK6161]

off topic
During my regular rounds for TLG through Swedish press, I found this anecdote. A friend of mine had only thrown away 8 at the time.)

"Oops – he happened to throw bitcoin worth billions at the dump​

Just over ten years ago, a hard drive with thousands of bitcoins ended up in landfills. Now a British man is waging a desperate battle against his hometown to try to get them back.​

16 October 2024
Some decisions can be regretted for the rest of your life. Just ask James Howells, a computer engineer who lives in the Welsh city of Newport in the UK.
Just over ten years ago, Howells and his then-partner decided to throw away various broken gadgets from his home office. One of the things that went into a black garbage bag was one of several hard drives that collected dust in a desk drawer.

Had 7,500 bitcoin in crypto wallet​

The next day, the partner went to the city's landfill. James Howells didn't think any more about it – but that would change.
As a technology enthusiast, Howells had taken an early interest in the cryptocurrency bitcoin. In a couple of years, his holdings had grown to 7,500 bitcoin. He stored these in a crypto wallet in the form of a hard drive that was in his office.
In the fall of 2013, his digital treasure cache was worth the equivalent of just over $14.6 million. At that time, James Howells began to suspect trouble. What had actually happened to the storage unit that contained all the bitcoins?
To his dismay, Howells discovered that he had accidentally mixed up his hard drives during the deep cleaning. The one with all the bitcoins was no longer at home – but had ended up in the dump, writes The New Yorker .

Must go through 100 tons of rubbish ​

Since then, James Howells has waged an increasingly desperate battle, trying to convince Newport City Council to dig up the landfill to snoop on the hard drive. According to his own statement, Howells has offered the city to put together a team that can dig through 100 tons of garbage.
He has also promised a juicy ransom: If he gets his way and the storage device is found, the city will pay ten percent of what the crypto coins are worth.
Newport has so far ignored James Howell's repeated applications. Therefore, he has now decided to sue the city for just over SEK 6 billion – a sum that corresponded to the value of SEK 7,500 when the application was submitted.

"The municipality is sitting on my stolen property"​

According to Newport's municipal council, the work would cost at least SEK 135 million and is estimated to take between 18 and 36 months to complete. After the excavation, it would also require at least another year of work to restore the soil on and around the soptation.
"The municipal executive board has made it clear to James Howells several times that the excavation is impossible to carry out. It would also cause a major environmental impact on surrounding areas," says a spokesperson from the city of Newsport in a comment.
But Howells doesn't give up. If the hard drive were to be found, he estimates that there is still an 80 percent chance that the digital coins can be retrieved from it.
"I see it as the municipality has been sitting on my stolen property for ten years," James Howells told The Sun ."

Hoppsan – han råkade slänga bitcoin värda miljarder på tippen

För drygt tio år sedan hamnade en hårddisk med tusentals bitcoin på soptippen. Nu för en brittisk man en desperat kamp mot sin hemstad för att försöka få dem tillbaka.

www.nyteknik.se

Ooops - $503,928,825 thrown away

The first think that comes to mind is that there is limited number of Bitcoins issued, whereas there is no limit on the number of shares Brainchip will issue

 

[Diogenese]

Hi @JB49, as I understand it BRN are using several programming models to support the various iterations of Akida, but have wondered myself why a single format could not be implemented/devised.

But the main reason I'm replying is to say that the video you put up Tuesday (post #90,294) regarding Intel and Loihi is quite definitive, with Mike Davies saying that his view is that: Loihi "Is quite some time away" with an emphasis on the "quite" - which made it very clear that it will be no time soon.

Your post came after @Bravo had posted (#90,274) regarding Mercedes working on a project with Intel, being the Naomi4 project which also included working with Univ of Waterloo Canada and Prof Chris Eliasmith, I was initially disappointed, but on reflection realized this is a current research project.

The important fact is that BRN have been partnered with Mercedes for several years and Akida is the only neuromorphic chip that is ready to go. However after watching the video you put up, I don't think anyone will be putting Loihi into a product any time soon, regardless there will be a lot of businesses who will be looking at Loihi because Intel have such a pre-eminent position in chip making.

Which of course goes back to the challenge facing BRN with regard to introducing a disruptive new technology in a business environment dominated by titans like Intel et al.

A lot to think about McH,

Note to self:
A. models
B. Mercedes NAOMI4
C. s/w

A. Models are what the NN has to search through.

I'll confine my thoughts to images and speech, but other sensor inputs are treated on the same principles.

Images: Static (photos, drawings); Moving (Video)


Sound: Key word spotting, NLP; other sounds.

Each of these can be divided into several layers of sub-categories with increasing specificity. In a NN, the larger the model, the more power is consumed in making an inference/classification, because the processor needs to examine each example in th model to see which the sensor input most nearly resembles.

Thus it make sense to have specific models foe spcific tasks. The narrower the task, the smaller the model can be.

For example, with image classification in an ADAS/AV, images of astronomy of scuba diving are irrelevant. So ADAS models are compiled from millions of images captured from vehicle-mounted cameras/videos.

Akida excels at classifing static images, and can do this at many frames per second. However, Akida 1 then relied on the associated CPU running software to process the classified images to determine an object's speed and direction. That's the genius of TENNS - it is capable of performing the speed analysis in silicon or in software far more efficiently than conventional software.

I prefer to talk about images/video because Natural Language processing is something I struggle to comprehend, but apparently TENNS makes this a cakewalk too.

Open AI tries to have everything in its model, but that burns a massive amount of energy for a single inquiry - a bit like biting off more than it can chew.

So now we have RAG, where subject-specific models can be downloaded depending on what the NN procesor is intended to do.

B. NAOMI4 - Yes This is a German government funded research project and will not produce a commercial outcome any time soon.

C. H/W v S/W

Valeo does not have an Akida silicon in its SCALA 3. It uses software to process the lidar sensor signals. Because we've been working with them for several years in a JD, I'm hopeful that the software will iclude Akida 2/TENNS simulation software. Sean did mention that we now have an algorithm product line.

The rationale for this was explained in the Derek de Bono/Valeo podcast posted yesterday that software allows for continual upgrading. He also mentioned that provision for some H/W upgrades could also be accommodated. Given TENNS young age, it will have developed significantly in the last couple of years, so it could not be set in silicon at this early stage, although Anil did announce some now deferred preparations for taping out some months ago.

Again, I am hopeful that Akida2/TENNS will be included in the software of both Valeo and Mercedes SDVs (and in other EAP participants' products) because it produces real-time results at a much lower power consumption.

Then there's PICO ... the dormant watchdog ...

 

[manny100]

ASX ann dated 6th March 2023 - GEN 2 introduced.

https://announcements.asx.com.au/asxpdf/20230306/pdf/45mc0vvb61xnn9.pdf

No doubt select clients had access to GEN 2 from March'23. Hopefully engagements from around this time should be getting close to conclusion within the next 3 to 6 months tops??
General availability of GEN 2 ANN OCT'23

BrainChip Makes Second-Generation Akida Platform Available to Advance State of Edge AI Solutions

Discover the revolutionary 2nd generation Akida platform by BrainChip, powering energy-efficient Edge AI devices with advanced neural network models.

brainchip.com

" “This is a significant step in BrainChip’s vision to bring unprecedented AI processing power to Edge devices, untethered from the cloud,” said Sean Hehir, CEO, BrainChip. “With Akida’s 2nd generation in advanced engagements with target customers, and MetaTF enabling early evaluation for a broader market, we are excited to accelerate the market towards the promise of Edge AI”.
My bold above.

 

[Frangipani]

I haven’t watched it, yet, but here is a recent interview (focusing on Akida Pico) with our CMO Steve Brightfield by William Wong, Senior Content Director of Electronic Design:

 

[Dallas]

Daily Ai Bytes on LinkedIn: #tesla #autopilot #safetyfirst #dailyaibytes #ai

𝗧𝗲𝘀𝗹𝗮 𝗔𝘂𝘁𝗼𝗽𝗶𝗹𝗼𝘁 𝘀𝗮𝘃𝗲𝘀 𝘁𝗵𝗲 𝗱𝗮𝘆! 🚗🤖 In a close call in Romania, a Tesla Autopilot system made a split-second decision to avoid a…

www.linkedin.com

 

[Draed]

I was reading an article about one of the problems AI is facing (sorry, I dont have the link). It was about how most of the data available on the internet has been reviewed and classified and incorporated into models already. The challenge moving forward is generating and capturing new data to continuously improve these models.

My question to the group is. Could we foresee our akida products being used in combination with other more sophisitcated but power hungry chips or software.

For example with mercades. Our chip or software at the edge and at the sensor. Quick, and power efficient inference done in realtime on input like voice activation, learning someones accent, vocal manerisms etc. But then that more refined data is processed in larger and more power hungry systems in the car or cloud for model improvements.

I see pico being such a device. Not intended to run independently. But more of a extreme edge device at the sensor, that just does data collection and refinement, but doesn't make any decisions on anything...

I might be way off. But that's my impression.

Cheers

Draed

 

[7für7]

As in politics, religion, society and all human constructs, corruption is inherent in the system.
Doesn't mean that its either all bad or that it cannot be improved, but big change is unlikely in my view.
The system is well protected by those, behind those, whose hands are on the levers, who gain directly the most, from the current settings.
Bit like the state of our cockeyed housing market.
Any government trying to initiate any meaningful change will be crucified at the polls by the majority of voters who want their primary asset to keep appreciating as opposed to the minority currently struggling for a foothold in the market.
The mistake was designating housing not as necessary and required shelter but as wealth enhancement, and then designing and entrenching systems which perverted the tax function towards that end.

The big player will never see a curt from the inside! If one Projekt fails, they start a new one. Building up a system with new little fishes. In our eyes maybe big player but for the real bosses just marionettes… just imagine what we are in their eyes…

 

[Diogenese]

TSMC says 'insane' AI demand is 'real' and a boon for chip giant​

TAIPEI -- Taiwan Semiconductor Manufacturing Co. on Thursday said "extremely robust" AI demand is set to continue for years and deliver another record-breaking quarter for the world's largest contract chipmaker.

TSMC said net profit in the July-September quarter surged 54.2% on the year to 325.26 billion New Taiwan dollars ($10 billion). Revenue, meanwhile, reached a record NT$759.69 billion, up 39%.

I was reading an article about one of the problems AI is facing (sorry, I dont have the link). It was about how most of the data available on the internet has been reviewed and classified and incorporated into models already. The challenge moving forward is generating and capturing new data to continuously improve these models.

My question to the group is. Could we foresee our akida products being used in combination with other more sophisitcated but power hungry chips or software.

For example with mercades. Our chip or software at the edge and at the sensor. Quick, and power efficient inference done in realtime on input like voice activation, learning someones accent, vocal manerisms etc. But then that more refined data is processed in larger and more power hungry systems in the car or cloud for model improvements.

I see pico being such a device. Not intended to run independently. But more of a extreme edge device at the sensor, that just does data collection and refinement, but doesn't make any decisions on anything...

I might be way off. But that's my impression.

Cheers

Draed

Hi Draed,

ChatGPT is not the Be-all-and-end-all. There are many other applications for AI models which do not need the omniscient entity in the cloud. Of course, Akida could reduce the power consumption and latency, but ... one step at a time .. We are hardly in a position to take on Nvidia head on just yet.

ADAS does not require such a massive universal knowledge base. Its image model is built on collections of vehicle mounted videos. I imagine there will be different models for left hand drive and right hand drive which can be downloaded using RAG* (Retrieval-Augmented Generation) as required.

Our association with Edge Impulse is aimed at automating the labelling of such data for the models, leveraging off pre-existing labelled models.

As for "Hey Mercedes", certainly PICO can be used for key word spotting to wake a larger processor, which could be a full strength Akida 2 with TENNS for natural language processing, Akida 2 being the preeminent NN processor. Akida 1 relieved the von Neumann CNN processor from the heavy lifting of inference/classification. TENNS now alleviates the task of temporal analysis of the classified input signals, leaving von Neumann to get on with whatever it does best (whatever that is).

* AI RAG - Where's Scott Joplin when you need him?

https://au.video.search.yahoo.com/s...326b7fc181a9cf1563532754ec1c904e&action=click

 

[TECH]

No one can deny that Sean hasn't assembled a solid team of (veterans) around him..if you look at our home webpage, as I did the other day, you will notice a well balanced team of experienced professionals..so we have the team, we have the brilliance of mind in the newly formed, refreshed team on the SAB...so, either product design has or is close to being finalized or we are still treading water with clients who just keep looking for reasons not to commit...in my opinion, the pressure must be mounting from within our company from the Board down to the engineering teams, the sales team, surely they must be feeling a level of frustration as we all are..Sean comes across to me as the type of person who likes his "i's" dotted and his "t's" crossed, very deliberate in his decision making, I haven't lost faith in his ability to deliver, but at the same time I am well aware that the spring is pretty tightly wound, so something has to give, one way or the other...as we all know the 4C is due in round 8 business days and I'm STILL not expecting any change...not being negative at all...just calling it as I currently see it...if I was the CEO I'd be feeling it right about now..and I don't believe Sean has put a foot wrong, it's a tough gig as Lou discovered...we have absolutely no control over what individual companies decide, the biggest part of that equation is time, as frustrating as that is.

End of my rambling.

Over and out...Tech.

 

[overpup]

FF Posted an interesting link on the crapfest worth a read!
MegaChips still in play...

https://arxiv.org/pdf/2408.13018

 

[CHIPS]

FF Posted an interesting link on the crapfest worth a read!
MegaChips still in play...

https://arxiv.org/pdf/2408.13018

These are the parts where Akida is mentioned:

 

[BrainShit]

Neuromorphic NPU Sips Power to Handle Edge Machine-Learning Models

BrainChip’s Akido Pico neural processing unit, which leverages spiking neural networks, targets low-power IoT and edge-computing devices.

www.electronicdesign.com

 

[Frangipani]

Another of our Laguna Hills engineers leaving without having another job lined up…

It should be noted that he has since been given glowing references from his former BrainChip colleagues.

Nevertheless, he seems to have been unhappy in his job for quite a while (see his 4 month old LinkedIn comment)… What is going on?

 

[cosors]

View attachment 71333

Neuromorphic NPU Sips Power to Handle Edge Machine-Learning Models

BrainChip’s Akido Pico neural processing unit, which leverages spiking neural networks, targets low-power IoT and edge-computing devices.

www.electronicdesign.com

There it is again, a new combination Akido Pico
Sounds kind of Japanese

____
In your posts I usually read over typos and don't notice them. But with Akida?
It's like Intel's at that time Pentium - Pentia, Pentio, Pentiam, Pentiom - hum hum.
Always a close call.

___
Seriously, this needs to be hammered or ram into the heads of all BRN writers.
How else is someone supposed to create a brand or product?
Kuci
Bolex
Nercedes Denz - yeah
Gola
A-pod
Verrari
A-pat
Akido
Dolls-Boyce
Mc Bonald's
And so on

Yo, Akido Bollisto

___
At the beginning I thought more funny, mistakes are human.
What would their company say if something was posted globally and said:
Dugatti
or
Bugatto

___
It's about building a brand. Painchip?!

I don't know why I'm so upset if the own employees do not take their texting seriously. Just letters.