BRN Discussion Ongoing

[Diogenese]

Following on from the above, there is something HUGE brewing here - South Korean chipmakers Sapeon and Rebellions have agreed to merge!!!!

https://www.reuters.com/technology/artificial-intelligence/south-korean-ai-chip-makers-rebellions-sapeon-agree-merge-2024-08-18/

This now brings together all of the following companies: Aramco + Prophesee + Samsung + Rebellions + Qauclomm + Sapeon.

South Korea has been strategically working towards becoming one of the world's top three countries in terms of AI innovation and this merrger seem to be a by-product of that. What is particularly fascinating about this is that South Korean president Yoon Suk Yeol referred to neuromorphic devices in the speech he gave in April 2024.


View attachment 68404


Sapeons' CEO has also discussed back in November 2022 neuromorphic technologies to improve memory operation processing performance.

Extract.
View attachment 68405

Sapeon "Enhancing computation with next-generation memory artificial brain ‘CIM'"

SK Group artificial intelligence (AI) semiconductor fabless Sapeon Korea introduced the next-generation memory semiconductor Compute-in-Memory (CIM) technology trend with enhanced computing functions.

english.etnews.com

I wonder where they could get their hands on the missing piece of the jigsaw puzzle, this neuromorphic technology?

Hi Bravo,

Compute-in-memory is usually used for analog, eg, where the "memory" stores the amplitude of the summed voltage.

Sapeon use MACs with a little trick, the development of which I feel I may have made a significant contribution.

US2022309127A1 MATRIX MULTIPLICATION METHOD AND DEVICE BASED ON WINOGRAD ALGORITHM 20202529

The present disclosure provides an operation apparatus operating based on the Winograd algorithm for multiplying a first matrix by a second matrix to generate a third matrix

 

[DingoBorat]

AL

Following on from the above, there is something HUGE brewing here - South Korean chipmakers Sapeon and Rebellions have agreed to merge!!!!

https://www.reuters.com/technology/artificial-intelligence/south-korean-ai-chip-makers-rebellions-sapeon-agree-merge-2024-08-18/

This now brings together all of the following companies: Aramco + Prophesee + Samsung + Rebellions + Qauclomm + Sapeon.

South Korea has been strategically working towards becoming one of the world's top three countries in terms of AI innovation and this merrger seem to be a by-product of that. What is particularly fascinating about this is that South Korean president Yoon Suk Yeol referred to neuromorphic devices in the speech he gave in April 2024.


View attachment 68404


Sapeons' CEO has also discussed back in November 2022 neuromorphic technologies to improve memory operation processing performance.

Extract.
View attachment 68405

Sapeon "Enhancing computation with next-generation memory artificial brain ‘CIM'"

SK Group artificial intelligence (AI) semiconductor fabless Sapeon Korea introduced the next-generation memory semiconductor Compute-in-Memory (CIM) technology trend with enhanced computing functions.

english.etnews.com

I wonder where they could get their hands on the missing piece of the jigsaw puzzle, this neuromorphic technology?

Although there's no doubt in my mind, that we've worked with at least Samsung, in the past..

For some reason, I don't think Korea is going to play ball with us, although they have looked at them and had a good sniff..

They have their Kaist Institute, which made some (actually misleading) claims, earlier this year?..

There's our known partnership with Prophesee, but I think this new development, is more likely to weaken that connection, than be a more confident shoe in the door..

 

[Fullmoonfever]

Quite like this Leaderboard on gesture recognition

 

[Fullmoonfever]

Quite like this Leaderboard on gesture recognition

View attachment 68419

Looks like starting the repo on Tenns from a couple of weeks ago.

 

[Fullmoonfever]

Looks like starting the repo on Tenns from a couple of weeks ago.

View attachment 68421 View attachment 68422

Links back to a previous post as in Peabrane.

BRN Discussion Ongoing

Not sure if This was posted already.. but anyway interesting. It was posted from a user in the German forum https://brainchip.com/hardware-pioneers-max24-may-28-29-2024-london/ Thanks for posting this. Every major sensor player should be highly interested in this. Akida being sensor agnostic...

thestockexchange.com.au

 

[Fastback6666]

Did they really want 3mil or was that a token gesture due to the Question at the AGM regarding offering sh some shares at the next raise.

Totally Token gesture for retail.

Then insto market manipulated down below the buy price of .193 to keep away the little retail rascals …us!

They didn’t need the minor 3mil from us anyway.

BRN got their time gap fill of $20mil and move on to the bigger goals.

 

[Frangipani]

On August 20, four researchers from Nara Institute of Science and Technology’s NAIST Robot Learning Lab (one of them has meanwhile graduated with an M.E. and landed a job as an engineer at NTT Docomo, Japan’s largest mobile provider owned by NTT - Nippon Telegraph and Telephone) and a former Nara Institute of Science and Technology M.E. student, now working for Mega Chips, published a paper titled Robust iterative value conversion: Deep reinforcement learning for neuro-chip-driven edge robots.
They also credit another NAIST Master student, Alonso Ramos Fernandez, for his experimental assistance.

The paper’s first author, Yuki Kadokawa, reports financial support was provided by MegaChips Corporation.

While I can’t access the full version of their paper, the video they uploaded to YouTube on August 21 is clear evidence they utilised AKD1000 for their research.

https://www.sciencedirect.com/science/article/abs/pii/S0921889024001660

 

[DingoBorat]

View attachment 68394
Hi FMF,

Great find!

For a moment, I thought G06N3/49 was a new class created for BRN.

I'm convinced that the SNN Valeo are using is Akida, and the fact that it has been classified in G06N3/49 is strongly indicative of TeNNs.

The priority date is 20230202, so they would have been working with Akida 2 for some time.

Basically they are comparing successive images, much like I described in my Prophesee-based analogy of TeNNs - morphic resonance.

Hey Diogenese, this patent, would obviously be for a Future potential product, or an upgrade to a current one?..

Do you still think there's a possibility, that AKIDA 2.0 (which includes TeNNs) is a part of the software stack of Scala 3?
Wouldn't there be a patent attached to Scala 3, for the G06N3/49 class, if there was?..

Or is that because, this is a more specific application of TeNNs and the details of Scala 3 patents, can be more ambiguous?

I'm making the assumption, that this recent patent, is definitely us.

 

[DingoBorat]

On August 20, four researchers from Nara Institute of Science and Technology’s NAIST Robot Learning Lab (one of them has meanwhile graduated with an M.E. and landed a job as an engineer at NTT Docomo, Japan’s largest mobile provider owned by NTT - Nippon Telegraph and Telephone) and a former Nara Institute of Science and Technology M.E. student, now working for Mega Chips, published a paper titled Robust iterative value conversion: Deep reinforcement learning for neuro-chip-driven edge robots.
They also credit another NAIST Master student, Alonso Ramos Fernandez, for his experimental assistance.

The paper’s first author, Yuki Kadokawa, reports financial support was provided by MegaChips Corporation.

While I can’t access the full version of their paper, the video they uploaded to YouTube on August 21 is clear evidence they utilised AKD1000 for their research.

https://www.sciencedirect.com/science/article/abs/pii/S0921889024001660

View attachment 68411

View attachment 68420

View attachment 68412

View attachment 68423



View attachment 68424





View attachment 68414

View attachment 68415

View attachment 68416

View attachment 68417

View attachment 68418

That's all very interesting but...

 

[TECH]

On August 20, four researchers from Nara Institute of Science and Technology’s NAIST Robot Learning Lab (one of them has meanwhile graduated with an M.E. and landed a job as an engineer at NTT Docomo, Japan’s largest mobile provider owned by NTT - Nippon Telegraph and Telephone) and a former Nara Institute of Science and Technology M.E. student, now working for Mega Chips, published a paper titled Robust iterative value conversion: Deep reinforcement learning for neuro-chip-driven edge robots.
They also credit another NAIST Master student, Alonso Ramos Fernandez, for his experimental assistance.

The paper’s first author, Yuki Kadokawa, reports financial support was provided by MegaChips Corporation.

While I can’t access the full version of their paper, the video they uploaded to YouTube on August 21 is clear evidence they utilised AKD1000 for their research.

https://www.sciencedirect.com/science/article/abs/pii/S0921889024001660

View attachment 68411

View attachment 68420

View attachment 68412

View attachment 68423



View attachment 68424





View attachment 68414

View attachment 68426

View attachment 68416

View attachment 68417

View attachment 68418

Hi Frangi....you are correct I believe....I noticed something in one of the photos you posted, so I got my magnifying glass and
to me it looks like one of our AKD 1000 engineering samples with a number (278) written in black marker like all the NSoC were.

Anyway, nice work.

Tech.

 

[Frangipani]

That's all very interesting but...

View attachment 68425

Here you go…

 

[Frangipani]

Hi Frangi....you are correct I believe....I noticed something in one of the photos you posted, so I got my magnifying glass and
to me it looks like one of our AKD 1000 engineering samples with a number (278) written in black marker like all the NSoC were.

Anyway, nice work.

Tech.

Thank’s, Tech!

Looks more like an upside down 325 or 525 to me, but I don’t have a magnifying glass at hand, so could be wrong.

 

[Diogenese]

Hey Diogenese, this patent, would obviously be for a Future potential product, or an upgrade to a current one?..

Do you still think there's a possibility, that AKIDA 2.0 (which includes TeNNs) is a part of the software stack of Scala 3?
Wouldn't there be a patent attached to Scala 3, for the G06N3/49 class, if there was?..

Or is that because, this is a more specific application of TeNNs and the details of Scala 3 patents, can be more ambiguous?

I'm making the assumption, that this recent patent, is definitely us.

As we know, SCALA 3 is just the lidar transceiver. It's point cloud signals are then processed in software in a CPU.

SCALA 3 has 5 times the laser spots compared to SCALA 2. That's like going from 1 laser pulse per half degree of scan to 1 laser pulse per tenth of a degree of scan. So it is possible (highly probable) that Akida 2/TeNNs was in the right place at the right time. Valeo would have been at the top of the queue to get access to MetaTF with Akida 2 simulation software, and we would have collaborated on compiling the models.

I couldn't say whether Valeo are using the full Akida 2 with TeNNs simulation software, or only TeNNs, but TeNNs would fit with the increased laser pulse rate.

The really funny thing is that the patent does pretty much what my Prophesee DVS / TeNNs analogy from the other day describes.

The patent used the SNN as a "standard" tool. The "invention" is in the processing software utilizing the SNN classification signals.

One issue would be the time taken in processing the frame signals.

The SNN would do the object classification for 2 adjacent frames, and then identifies the differences between the frames, and changes above a threshold difference are used by the CPU in monitoring the cumulative changes. I think TeNNs software could do the change monitoring more efficiently than other software, and, of course, in hardware it would be more efficient again.

 

[DingoBorat]

Here you go…


View attachment 68427

Already got it, but mines in black..

 

[Diogenese]

Already got it, but mines in black..

Yeah - I've got that one with the black writing ...

 

[Frangipani]

Already got it, but mines in black..

You could check with them whether they’d be able to reprint the letters in green or orange, then? And maybe add a time frame?

 

[DingoBorat]

You could check with them whether they’d be able to reprint the letters in green or orange, then? And maybe add a time frame?

How about AKIDA in "AKIDA blue" down the side next to hope?

 

[Frangipani]

Speaking of colours:

My face is as red as a rhesus monkey's ruckseite.

Oh, so your face turned brownish-grey, then?!

Because that’s the fur colour of a rhesus monkey’s “Rückseite“, even though we would never use that word for animals (or human beings) - we’d use “Rücken“ instead.

“Rückseite“ (rear/reverse/flip/back side etc) is used when referring to objects such as a TV, a house, even the Moon. There are, however, exceptions. Some mountains have “Rücken“ (= ridges). And books have literally both a “Rückseite” (back) and a “Rücken“ (spine).

“Butt that’s not what I meant!”

What a bummer! Unfortunately, the colour was lost in translation, arse backside ≠ Rückseite. Literal translations can be tricky.

To convey the fire engine redness of blushing, Affenpopo, Affenhintern or the vulgar, but awesomely alliterative Affenarsch would have been the correct choice of words (usually associated with baboons in German, though, not macaques).


(P.S.: There are those here on TSE who consider me a smart-arse/ass, but hey, I’d rather be that than a dumb-arse/ass).

 

[Frangipani]

How about AKIDA in "AKIDA blue" down the side next to hope?

Doesn’t work for me, I’m afraid, as I am a synaesthete and perceive the letter A as being red, dominating the colour of the whole word (no kidding, this is true).

Synesthesia - Wikipedia

en.m.wikipedia.org

https://en.m.wikipedia.org/wiki/Grapheme–color_synesthesia

 

[McHale]

Totally Token gesture for retail.

Then insto market manipulated down below the buy price of .193 to keep away the little retail rascals …us!

They didn’t need the minor 3mil from us anyway.

BRN got their time gap fill of $20mil and move on to the bigger goals.

Yes, but the instos got their shares @.193, which from memory was around a 4% discount to the 10 day VWAP,

LDA were getting theirs @ around a 9% discount to market, so why were UCP prepared to underwrite
a $20m CR at such a low discount to market (or vwap).

I own a gold miner BGL which is an already profitable producer with v high grade gold and reserve of 3.2m ozs
(world class deposit and v high grade $$$),BGL had to take a 15% discount to market for a recent raise of $120m,
they had $75m in the bank before CR.

A lot of complaints and speculation here about why do a CR, well for a start it's a much better deal for BRN and for us
because there is much less dilution. Then BRN would definitely be able to get on with their commercialization plans with a solid cash runway, and not have to be concerned with burn rate.

Looks like Sean and maybe Antonio (perhaps others) were working on the CR before the AGM, but why were UCP
prepared to take such a low discount to market, many companies are paying more than cashed up BGL for money, 20% discount is not uncommon at present cash is tight.

Again I ask why would UCP take such a low discount and then put on social media how pleased they were to do
the CR with BRN ???