BRN Discussion Ongoing

manny100

Regular
I aked co pilot who has commercialised cloudless AI at the Edge.
The answer was Brainchip, Aarna Networks and Cloudflare.
I looked up Aarna and Cloudflare and both actually still use the cloud. You could easily get the impression that they are cloudless if you were not aware of Brainchip.
Interesting.
 
  • Like
  • Thinking
  • Fire
Reactions: 11 users

Frangipani

Regular
His LinkedIn profile states he’s almost a perpetual student therefore his access to BrainChip’s true potential would be seriously limited. I’d suggest he digs a little deeper than looking at one promotional video from 3 years ago before declaring we are lying about our capabilities. Next!

Pretty sure it's an auto-correct, that was missed and is supposed to say "presented".

I strongly believe this is a typical case of “lost in translation”. What Loïc Cordone really wanted to express (although he seems to have wrongly translated this into English from his native French) was most likely that BrainChip had claimed something, NOT that they had pretended! Big difference!



B7CB8A1A-037A-4704-A48D-97152DCBA82C.jpeg


[While this box contains two spelling mistakes, it doesn’t concern the accuracy of the statement about this pair of false friends. The author uses the accent aigu (é) correctly on the prefix in prétendre and prétend, but at the same time incorrectly, when he/she writes preténdre twice, just in case anyone wondered…]


E11E07FF-5EBD-4D4F-90FB-76C2C439C4B1.jpeg



His LinkedIn profile states he’s almost a perpetual student therefore his access to BrainChip’s true potential would be seriously limited. I’d suggest he digs a little deeper than looking at one promotional video from 3 years ago before declaring we are lying about our capabilities. Next!

Maybe you should dig a little deeper, before making such a statement?

Firstly, you are doing this guy an injustice by alleging he is “almost a perpetual student”. After graduating with a “diplôme d’ingénieur, mathématiques et informatique” (which is equivalent to a Master’s degree) in 2019, he did a PhD on Bio-Inspired AI at LEAT (Laboratoire d’Électronique, Antennes et Télécommuncations), a joint research unit between the University of Nice-Sophia Antipolis (renamed Université Côte d’Azur in 2019) and the CNRS (Centre national de la recherche scientifique), while working as a researcher at the Renault Software Factory (October 2019 - January 2023). Such industrial PhDs are quite common these days.
Since April 2023, he has been employed as an ML researcher in the field of GenAI at Shift Technology.
“Almost a perpetual student”?

A2303CC8-9E7F-41DB-910B-FC9504FA0584.jpeg


Secondly, when Loïc Cordone defended his PhD thesis in mid-December 2022, the BrainChip video he refers to in his thesis (which he had obviously submitted even earlier that year - there is normally a timespan of at least a couple of months between PhD thesis submission and defense) was just over a year old (BrainChip Demonstrates Smart Automotive, Oct. 2021), and he was factually correct when commenting on the lack of peer-reviewed papers by fellow researchers who had used Akida (and thus the lack of benchmarking of neuromorphic hardware that included Akida) at the time.
You, however, unfairly judge him as if he had written his PhD thesis in 2024.


The way I see it, Loïc Cordone’s verdict on Akida is not at all what you think it is. He did not intend to accuse BrainChip of lying about Akida’s capabilities, but instead included Akida in his table on neuromorphic hardware at a time when a lot of other academic researchers wouldn’t have mentioned BrainChip at all in their papers.

He then went on to say “These different neuromorphic hardware represent exciting and promising leads for the implementation of neuromorphic process toolchains…”. Which included Akida.

36B40B17-AC6D-4222-9E2E-8C8AD81AC11F.jpeg



IMO it all makes sense when we simply blame the native French speaking author (and those who possibly proofread his thesis) for not paying enough attention to the false friends prétendre vs to pretend, when writing his thesis in English.
All in all - the way I see it - he was actually positive about Akida, presenting it in a table with other, much better researched neuromorphic hardware at the time (2022).



EB3CD37D-95EE-4255-982B-C4DE586AE94A.jpeg

24247E74-606F-4A0B-8A61-25326E9FE0BD.jpeg
 
Last edited:
  • Like
  • Fire
  • Wow
Reactions: 25 users

itsol4605

Regular
I strongly believe this is a typical case of “lost in translation”. What Loïc Cordone really wanted to express (although he seems to have wrongly translated this into English from his native French) was most likely that BrainChip had claimed something, NOT that they had pretended! Big difference!



View attachment 62603

[While this box contains two spelling mistakes, it doesn’t concern the accuracy of the statement about this pair of false friends. The author uses the accent aigu (é) correctly on the prefix in prétendre and prétend, but at the same time incorrectly, when he/she writes preténdre twice, just in case anyone wondered…]


View attachment 62604




Maybe you should dig a little deeper, before making such a statement?

Firstly, you are doing this guy an injustice by alleging he is “almost a perpetual student”. After graduating with a “diplôme d’ingénieur, mathématiques et informatique” (which is equivalent to a Master’s degree) in 2019, he did a PhD on Bio-Inspired AI at LEAT (Laboratoire d’Électronique, Antennes et Télécommuncations), a joint research unit between the University of Nice-Sophia Antipolis (renamed Université Côte d’Azur in 2019) and the CNRS (Centre national de la recherche scientifique), while working as a researcher at the Renault Software Factory (October 2019 - January 2023). Such industrial PhDs are quite common these days.
Since April 2023, he has been employed as an ML researcher in the field of GenAI at Shift Technology.
“Almost a perpetual student”?

View attachment 62612

Secondly, when Loïc Cordone defended his PhD thesis in mid-December 2022, the BrainChip video he refers to in his thesis (which he had obviously submitted even earlier that year - there is normally a timespan of at least a couple of months between PhD thesis submission and defense) was just over a year old (BrainChip Demonstrates Smart Automotive, Oct. 2021), and he was factually correct when commenting on the lack of peer-reviewed papers by fellow researchers who had used Akida (and thus the lack of benchmarking of neuromorphic hardware that included Akida) at the time.
You, however, unfairly judge him as if he had written his PhD thesis in 2024.


The way I see it, Loïc Cordone’s verdict on Akida is not at all what you think it is. He did not intend to accuse BrainChip of lying about Akida’s capabilities, but instead included Akida in his table on neuromorphic hardware at a time when a lot of other academic researchers wouldn’t have mentioned BrainChip at all in their papers.

He then went on to say “These different neuromorphic hardware represent exciting and promising leads for the implementation of neuromorphic process toolchains…”. Which included Akida.

View attachment 62610


IMO it all makes sense when we simply blame the native French speaking author (and those who possibly proofread his thesis) for not paying enough attention to the false friends prétendre vs to pretend, when writing his thesis in English.
All in all - the way I see it - he was actually positive about Akida, presenting it in a table with other, much better researched neuromorphic hardware at the time (2022).



View attachment 62608
View attachment 62609
Interesting stuff !!
Brainchip Akida and Akida2.0 seems to be the best!

Why doesn't anyone use Akida??

Why are there so few or no IP license agreements?
 
  • Like
  • Fire
Reactions: 3 users

Diogenese

Top 20
Well, talking about peer reviews:

7.2.2
"... While SNNs operate on binary spikes, they still rely on floating point weights and potentials, and as we’ve seen in Chapter 6, the number of bits on which they are encoded has a direct influence on the energy consumption of the network, but also on the latency and on the size of the network considered. While several post-quantization schemes for SNNs are already available in the literature, we think that developing new Quantization-Aware Training (QAT) methods for SNNs will bring better performance when considering the encoding of floating point values on a low number of bits. This will ensure that the performance achieved during training will be met in real-life conditions."

The paper is loaded with useful background information and analysis, but is deficient in that it only mentions Akida superficially and does not mention N-of-M coding, feedforward ML, asynchronous digital, and assumes Floating Point weights ... and now there's TeNNs.

Interesting to see Masqualier (ex Spikenet) and Benosman (ex Prophesee) as supervisors.
 
  • Like
  • Love
  • Fire
Reactions: 21 users

IloveLamp

Top 20
 
  • Like
  • Fire
  • Wow
Reactions: 9 users

IloveLamp

Top 20
Last edited:
  • Like
  • Fire
  • Love
Reactions: 48 users

IloveLamp

Top 20
  • Like
  • Wow
Reactions: 3 users

Tothemoon24

Top 20
IMG_8911.jpeg






Last week, I had the incredible opportunity to visit the Netherlands (with my colleague Sounak Dey) on a business trip that turned out to be much more than I expected.

Primary purpose of the visit was to present our research work (as invited speakers) at Morpheus Edge AI & Neuromorphic Computing workshop (https://lnkd.in/g-z-dAxJ) organised by Laurent Hili at ESA/ESTEC (European Space Research & Technology Centre). It was well rounded workshop where we got to meet SpaceTech stakeholders that a innnovating towards embedding intelligence onboard satellites (startups, soc manufacturers, european university researchers, ESA project leaders, etc). Excellent work by workshop organiser and the presenting participants.

Unexpectedly, I also met Sir Ananth Krishnan (Tata Consultancy Services Ex-CTO) on his post retirement holiday. And the first thing that came to my mind was his retirement speech address at our IIT-KGP Research Park office: He motivated us to make products & services that are 1] Usable (should pass Grandma test) 2] Trustworthy/Reliable 3] Frugal (Space & Time resource efficient, think of bits & bytes). His advice still guides us in our research work.

The research work we presented was done in colllboration with our friends at BrainChip (Gilles Bézard and Alf Kuchenbuch). Brief introduction to our work that got deployed on Brainchip AKIDA neuromorphic processor:

Currently, there is a delay of many hours or even days to draw actionable insights from satellite imagery (due to mismatch between data volume acquisition & limited comms bandwidth). We observed that end-users either need RAW images or Analytics-ready meta-data as soon as possible. Therefore, embedding intelligence (Edge AI) onboard satellites can result in quicker business decision making across business verticals that rely on geo-spatial data.

To address this, guided by the foresight of Dr. Arpan Pal we built a bundle of tech capabilities that helps send RAW data & Analytics-ready meta-data as soon as possible to ground station. These tech capabilites include:
1) Cloud Cover Detection Model (high-accuracy, low-latency, low-power).
2) DL based Lossless Compression (around 45% compression ratio).
3) RL based Neural Architecture Search Algorithm (quickly search data+task+hardware specific optimal DL models).

We also had a chance to visit TCS Paceport in Amsterdam, hoping to showcase our research prototype there soon. Looking forward to more future collaborations with Edge AI/Neuromorphic hardware accelelator designers & space-grade SoC manufacturers.

Would like to thank Tata Consultancy Services - Research for such great opportunity to build future tech for future demand. Would also like to thank our Edge & Neuromorphic team: Arijit Mukherjee, Sounak Dey, Swarnava Dey, Abhishek Roy Choudhury, Shalini M., Syed Mujibul Islam, Sayan Kahali and our academic research advisor Dr. Manan Suri.
 
  • Like
  • Fire
  • Love
Reactions: 41 users
Interesting stuff !!
Brainchip Akida and Akida2.0 seems to be the best!

Why doesn't anyone use Akida??

Why are there so few or no IP license agreements?
1715457905351.gif
 
  • Haha
  • Love
Reactions: 9 users

Rskiff

Regular
See how far ahead Tesla is compared to Mercedes despite just using vision. Very interesting.
 
  • Wow
  • Like
Reactions: 5 users

Tothemoon24

Top 20
IMG_8913.jpeg



Partnership with BrainChip Allows Cornell Tech Students Exposure to Neuromorphic Computing

Cornell Tech has partnered with BrainChip, the world’s first commercial producer of neuromorphic artificial intelligence, to introduce a new course in neuromorphic computing to its graduate students by joining the company’s University AI Accelerator Program.
The Cornell Tech course on neuromorphic technology – computing that mimics the neural behavior of the human brain – was introduced to students in the electrical and computer engineering program in the spring 2024 semester.

BrainChip’s University AI Accelerator Program provides platforms, and guidance to students at higher education institutions with AI engineering programs training.
Students participating in the program will have access to real-world, event-based technologies offering unparalleled performance and efficiency to advance their learning through graduation and beyond.

The course at Cornell Tech is currently being taught by Jae-sun Seo, associate professor of electrical and computer engineering.
Seo joined Cornell Tech in 2023 and his research centers on hardware design of machine learning and neuromorphic algorithms as well as hardware-efficient AI algorithm design.

“Our goal at Cornell Tech is to develop leaders for the AI era who are capable of applying technical advancements emerging in industry to make a positive impact on society,” said Seo.
“One of the best ways to do this is to partner with those in both the private and public sectors to advance practical technology solutions that solve real-world challenges.
Working with BrainChip has allowed students to obtain the resources and learning experiences they need to succeed in neuromorphic computing.”

Neuromorphic solutions allow for faster systems that consume less power.
BrainChip focuses on machines that consume less power by drawing on a system of “neurons” in order to do more with less.

BrainChip’s neural processor, Akida™ IP, is an event-based technology that is inherently lower power when compared to conventional neural network accelerators.

Lower power affords greater scalability and lower operational costs.

Among the markets that BrainChip’s technology will impact are the next generation of smart cars, smart homes of today and tomorrow, and industrial IoT.

BrainChip University has implemented similar AI Accelerator Programs at a number of universities including Arizona State University, Carnegie Mellon University, Rochester Institute of Technology, the University of Oklahoma, and the University of Virginia.

 
  • Like
  • Fire
  • Love
Reactions: 36 users

Tothemoon24

Top 20
IMG_8914.jpeg
 
  • Like
  • Fire
  • Love
Reactions: 83 users

Terroni2105

Founding Member

“As performance is an important criterion in the selection of hardware for space missions, the European Space Agency has published an open source benchmark suite called OBPMark [2]. It is a set of benchmarks based on typical space applications and designed to measure system-level performance. However, there is currently no standard tool for evaluating the performance of distributed on-board computers.”

https://indico.esa.int/event/445/contributions/8447/attachments/5819/9710/Poster_hoch_A0_EN_EDHPC_Towards%20a%20Parallel%20Benchmark%20for%20Space%20Applications%20Distributing%20OBPMark's%20Image%20Processing.pdf
 
  • Like
  • Love
  • Fire
Reactions: 19 users

7für7

Top 20
  • Like
  • Fire
  • Haha
Reactions: 6 users

IloveLamp

Top 20
  • Like
  • Love
  • Fire
Reactions: 16 users

Getupthere

Regular

SoftBank Group's (9984.T) Arm Holdings (O9Ty.F), plans to develop artificial-intelligence chips, seeking to launch the first products in 2025, Nikkei Asia reported on Sunday.

UK-based Arm will set up an AI chip division and aim to build a prototype by spring 2025, the report added. Mass production will be handled by contract manufacturers and is expected to start in the autumn of 2025, Nikkei Asia said.

Arm will pay for initial development costs, which may go up hundreds of billions of yen, with SoftBank also contributing, the report said.

Once a mass-production system is established, the AI chip business could be spun off and placed under SoftBank, the newspaper said, adding that SoftBank is already negotiating with Taiwan Semiconductor Manufacturing Corp (2330.TW) and others over manufacturing, looking to secure production capacity.

Arm declined to comment, while SoftBank and TSMC did not immediately respond to requests for comment.

The UK chip designer, which licenses its chip designs and earns funds through royalties, has been expanding into the data center market where operators are looking to build their own chips to power new AI models and reduce their reliance on dominant supplier Nvidia (NVDA.O).
 
  • Like
  • Thinking
  • Fire
Reactions: 14 users

Diogenese

Top 20
View attachment 62668
BRN did a co-presentation with onsemi at CES 2024 on time of flight (ToF).

onsemi use ToF in ultrasonic applications for automotive assisted parking:

Ultrasonic Sensor​

The sensor measures the time of flight to an object and converts to the distance for park assist application.
View Parametric Table

https://www.onsemi.com/products/sensors

onsemi is a big deal. They were spun out of Motorola's semiconductor business 25 years ago and recently purchased Fairchild along with many other IC companies along the way.

https://www.onsemi.com/company/about-onsemi/our-history



 
  • Like
  • Fire
  • Love
Reactions: 55 users

IloveLamp

Top 20
BRN did a co-presentation with onsemi at CES 2024 on time of flight (ToF).

onsemi use ToF in ultrasonic applications for automotive assisted parking:

Ultrasonic Sensor​

The sensor measures the time of flight to an object and converts to the distance for park assist application.
View Parametric Table

https://www.onsemi.com/products/sensors

onsemi is a big deal. They were spun out of Motorola's semiconductor business 25 years ago and recently purchased Fairchild along with many other IC companies along the way.

https://www.onsemi.com/company/about-onsemi/our-history



2950.gif
 
  • Like
  • Haha
  • Wow
Reactions: 14 users
Top Bottom