BRN Discussion Ongoing

[Fullmoonfever]

Pretty sure this Benosman article from early last year has been posted before and whilst doesn't discuss Akida, I picked an excerpt reminder below about his thoughts on a right processor & sensor.

https://www.eetimes.com/a-shift-in-computer-vision-is-coming/

Why post it again though?

Cause I just saw a reference to it from a week ago that was from this gentleman and we know who they like to work with and who that company has liked to work with.....


Lyes Khacef
Senior Research Scientist at Sony R&D | Neuromorphic Computing
1w

General–purpose neuromorphic processors are lagging behind their DVS camera counterparts. Efforts from some of the industry’s biggest players (IBM Truenorth, Intel Loihi) are still a work in progress. Benosman said that the right processor with the right sensor would be an unbeatable combination.

“[Today’s DVS] sensors are extremely fast, super low bandwidth, and have a high dynamic range so you can see indoors and outdoors,” Benosman said. “It’s the future. Will it take off? Absolutely!”

“Whoever can put the processor out there and offer the full stack will win, because it’ll be unbeatable,” he added.

Lyes Khacef on LinkedIn: In case you missed it, here is a brilliant article from R.B. Benosman on…

In case you missed it, here is a brilliant article from R.B. Benosman on event-based sensing and neuromorphic computing.

tz.linkedin.com

 

[Sirod69]

Continental
Frank Petznick (left), head of Autonomous Mobility Business Area-Continental, and Fermi Wang, president and CEO of Ambarella, ink partnership.
INDUSTRY NEWS

AI Gets Boost in Continental’s Latest Partnership​

Continental is connecting with Silicon Valley-based Ambarella combining Ambarella’s computer vision know-how, system-on-chip and software modules with Continental’s software and hardware expertise and broad portfolio of automotive system solutions.
Joseph Szczesny | Jan 18, 2023

Continental, one of the automotive industry’s top suppliers, is launching a strategic partnership with Ambarella, an edge AI semiconductor company.

The alliance underscores the growing dependence of automakers and their Tier 1 partners such as Continental on the suppliers of sophisticated semiconductors. Semiconductor shortages, which have disrupted production, are prompting automakers and key suppliers to develop closer ties to chip makers.

.....

In addition to the development of camera-based perception solutions for driver assistance or ADAS systems, Continental and Ambarella are focusing on “full-stack systems” for Level 2+ up to highly automated vehicles.

AI Gets Boost in Continental’s Latest Partnership

Continental is connecting with Silicon Valley-based Ambarella.

www.wardsauto.com

 

[Fullmoonfever]

Just on neuromorphic vision.....appears Collins (Raytheon) moving in that direction as well.

Principal Research Engineer, Perception Systems
• Deadline: May 1, 2023
• Career levels: Assistant Professor, Associate Professor, Company Researcher, Lecturer, PostDoc, Professor, Senior Researcher
• Keywords: Artificial intelligence, Computer vision, Machine learning, Neuromorphic computing, Robotics

Collins Aerospace, a Raytheon Technologies company, is a leader in technologically advanced and intelligent solutions for the global aerospace and defense industry

• Experience with OpenCV, Python coding, and image processing are considered essential. Experience with 3D cameras and event cameras are considered highly beneficial
• Experience with deep learning image processing, and experience with Keras and Tensorflow are considered essential.
• Knowledge of neuromorphic computer vision is considered a key differentiator between candidates.
• Passion in building prototypes integrating software/hardware components.

https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&cad=rja&uact=8&ved=2ahUKEwj_nMbOk9H8AhVXwjgGHXTFBWM4ChAWegQIHxAB&url=https%3A%2F%2Fwww.hipeac.net%2Fjobs%2F13833.pdf&usg=AOvVaw3vbGDqdxi-se9G71qcLKKe

 

[Diogenese]

https://www.valeo.com/en/valeo-scala-lidar/

Valeo’s third-generation laser LiDAR technology, which is scheduled to hit the market in 2024, will take autonomous driving even further, making it possible to delegate driving to the vehicle in many situations, including at speeds of up to 130 km/h on the highway. Even at high speeds on the highway, autonomous vehicles equipped with this system are able to manage emergency situation autonomously.
...

• 2018: Valeo was the first company in the world to run an autonomous vehicle in central Paris, in 100% autonomous driving mode, with Valeo Drive4U, equipped exclusively with series-produced sensors
• 2021: The Honda Legend and the Mercedes-Benz S-Class are the first cars to have reached level 3 automation in the market. Both models are fitted with Valeo’s LiDAR technology.

I guess they will be doing some testing of Scala 3 before it hits the market. Admittedly, hitting the market in 2024 could mean December 2024, but who releases a new car at Christmas?

We haven't seen anything to prove Scala 3 has Akida, but that's where the smart money is.

Well Christmas has come early for Volvo enthusiasts:

https://www.msn.com/en-au/motoring/...sedgntp&cvid=cb26d0e8478e4b5794d6d72ed8dbd9c4

2024 Volvo EX90 Is a Seven-Seater Electric SUV With 671 Lb-Ft of Torque​

Story by Brian Silvestro • Yesterday 7:45 am

...
The Swedish company revealed the EX90 in downtown Stockholm Wednesday, though it's clear this car was built to please an American audience. Designed to eventually replace its other full-size SUV, the XC90, it's described as a stylish family car with "modern proportions." It can seat seven people in comfort, and comes packed with 111 kWh worth of battery under the floor.

That battery can deliver an estimated range of up to 300 miles, says Volvo. At launch, only all-wheel-drive variants will be available, meaning permanent magnet electric motors for each axle bringing a total of 496 hp and 671 lb-ft of torque. Volvo hasn't released 0-60 or top speed estimates, but it says the battery can recharge from 10 to 80 percent in just 30 minutes... with the right charger, of course.

That's 900Nm of torque - it could climb the Matterhorn.

https://www.caranddriver.com/volvo/ex90

Volvo has said that the EX90 will also showcase the brand’s next-generation driver-assistance technology, including a new driver-monitoring system that can take action to help the driver if it detects distress or distraction.
...
Volvo promises that new driver-assistance features will make their debut on the EX90, including the company’s next-generation adaptive cruise control with lane-centering and active steering assist. Basics such as automated emergency braking are included too, and the EX90 also features a novel in-car radar system that’s intended to help prevent children and pets from getting left behind in a hot car.

Due out in early 2024.


Volvo have been working with Nvidia on Ai for trucks since 2019.

https://www.businessinsider.com/vol... Business Insider's homepage for more stories.

Johannes Hellstrom, Reuters Jun 28, 2019, 3:52 AM

After getting dropped by Tesla, a US chipmaker is partnering with Volvo to develop artificial intelligence for self-driving trucks​

... but this article also referes to Mercedes Benz working with Nvidia.


Nvidia, which has previously announced technology partnerships with automakers including Volkswagen AG, Mercedes-Benz and Toyota, said it was thrilled to team up with Volvo.

"The latest breakthroughs in AI and robotics bring a new level of intelligence and automation to address the transportation challenges we face," Nvidia Chief Executive Jensen Huang told reporters on a conference call.

Huang emphasized that for Nvidia the partnership with Volvo goes beyond supplying chips, and includes work on developing software, on-board computers and sensor systems for autonomous commercial vehicles. "The technology inside the vehicle is much more significant than we would apply" to a consumer-owned car, he said.
...
Nvidia's so-called Drive Constellation chips often power the machine learning used to refine self-driving car software algorithms inside data centers, and the company has also been working to build its Drive chips into cars.

Automotive chips accounted for $641 million of Nvidia's $11.7 billion in revenue in its most recent fiscal year.

Tesla was previously a major customer for Nvidia's automotive chips but last year CEO Elon Musk said the electric-car pioneer was developing its own chip.

AB Volvo's and Nvidia's collaboration will be built on Nvidia's full software package for sensor processing, perception, map localization and path planning.

Mercedes was working with Nvidia before 2022.

Volvo must have seen the Mercedes/BrainChip announcement ... have they just been sitting on their hands for the last year?

... and what about Nvidia?

 

[cosors]

MS apparently wants to use GPT for Bing. I guessed it immediately...
Keep thinking of the little search box at the bottom of the Win taskbar, is my other thought. So far mostly meaningless Bing results

Microsoft and OpenAI Working on ChatGPT-Powered Bing in Challenge to Google

 

[cosors]

Dr Roba Abbas was also referenced by Professor Katina Michael as a collaborator in the ASU/BrainChip University AI Accelerator video
She is at the University of Wollongong


View attachment 26557

https://scholars.uow.edu.au/display/roba_abbas

Katina Michael and => BrainChip Inc: AI Accelerator Program - Introduction to Neuromorphic Computing
Januar 8, 2023

It came to my attention through my search in India.

I wondered what this one had to do with BRN. And came across Katina.
(...he mentions an interesting term that is new to me - istead robots cobots)

 

[cosors]

Katina Michael and => BrainChip Inc: AI Accelerator Program - Introduction to Neuromorphic Computing
Januar 8, 2023

It came to my attention through my search in India.
View attachment 27432
I wondered what this one had to do with BRN. And came across Katina.
(...he mentions an interesting term that is new to me - istead robots cobots)

Wow, I'm thrilled and blown away at the same time. I have to make time for this! First step I extract an mp3 and make an audio ~book out of it

___
I clearly recommend watching or listening to it, and above all taking the time to do so. If I have understood correctly three more lessons will follow. I think the statement about Loihi and TrueNorth alone is very good. My comment in my post above about the reading recommended by MB makes me feel a bit ashamed 🫣

 

[stuart888]

BRCHF: A First Mover in a New Market

scr.zacks.com

Sorry if already posted

This is a fantastic Brainchip document, which I doubt many people saw in whole. It summarizes a lot of technical detail as you dig into the details. Brainchip staff obviously provided lots of facts and images here.

https://s27.q4cdn.com/906368049/files/News/2022/Zacks_SCR_Research_11012022_BRCHF_Lantier.pdf

I wanted to post all the text/images, but my browser does not let me post the entire pdf, so people can just read it, rather than having to download it.

On page 9 of 26, they show this image of the Akida Neural Fabric. This goes back to some of the Data-in-flow into Akida that had been discussed. It shows "AXI 4.0" and the "System Interface DMA Data & Configuration".

This stuff is deep, but it relates to the serialized data stream. Probably, not many want this detail, but here ya go.

AXI 4.0 - The AXI4-Stream protocol is used as a standard interface to connect components that wish to exchange data. The interface can be used to connect a single master, that generates data, to a single slave, that receives data. The protocol can also be used when connecting larger numbers of master and slave components.

The protocol supports multiple data streams using the same set of shared wires, allowing a generic interconnect to be constructed that can perform upsizing, downsizing and routing operations.

The AXI4-Stream interface also supports a wide variety of different stream types. The stream protocol defines the association between Transfers and Packets.

The DMA interface acts as a bus master on the system interconnect. Two types of data are transferred on the interface: data descriptors and actual data packets. The interface is very efficient in transferring full duplex Ethernet packet traffic. Read and write data transfers from different DMA channels can be performed simultaneously on this port, except for transmit descriptor reads and write-backs, which cannot happen simultaneously.

Wake back up!

 

[stuart888]

Timothee Masquelier, big SNN Spike Lover, works with Simon Thorpe (JAST). Timothee Masquelier might be the "T" in JAST if memory serves correctly.

Yes, another sleepy deep read where Brainchip potential solutions are in highly technical depth!
Sorry, buy you get to learn about Quadratic Spike Penalization Loss.

https://hal.science/hal-03906970/document

What I thought was neat is the math, where in my terms "break out of situation, after 4 spikes exceed the thresholds". It is my understanding, that part of the power reduction techniques, is after the first 4 detections over-the-thresholds, it breaks out of the loop (so to speak, there is no loop).

Yes, after meeting these event results, then Akida ignores or does less processing on the rest of the event data.

This appears to be the Math! I could be way off.

 

[Rise from the ashes]

Timothee Masquelier, big SNN Spike Lover, works with Simon Thorpe (JAST). Timothee Masquelier might be the "T" in JAST if memory serves correctly.

Yes, another sleepy deep read where Brainchip potential solutions are in highly technical depth!
Sorry, buy you get to learn about Quadratic Spike Penalization Loss.

https://hal.science/hal-03906970/document

What I thought was neat is the math, where in my terms "break out of situation, after 4 spikes exceed the thresholds". It is my understanding, that part of the power reduction techniques, is after the first 4 detections over-the-thresholds, it breaks out of the loop (so to speak, there is no loop).

Yes, after meeting these event results, then Akida ignores or does less processing on the rest of the event data.

This appears to be the Math! I could be way off.

View attachment 27437

Ahh yes the T in JAST as our learned poster here @Jefwilto reminded me some time back.

 

[stuart888]

Did I miss this new Brainchip document?

Benchmarking AI Inference at the Edge: Measuring Performance and Efficiency for Real-World Deployments

https://brainchip.com/wp-content/uploads/2023/01/BrainChip_Benchmarking-Edge-AI-Inference-1.pdf

 

[stuart888]

Akida 1.0 two page Product Brief "Final". An excellent sales/marketing tool used.

Final might mean, all development of new features is going to Akida 2.x version next!

https://brainchip.com/wp-content/uploads/2022/06/Akida-1.0-IP-Product-Brief_final.pdf

 

[stuart888]

Not sure if this document made it here. It is a summary PDF by Kristofer Carlson at TinyML a few months back.

PDFs are a pain for me on my tablet, but even on the 32" PC screen, this document needs mega zooming.

Just want to make sure our tech experts here get a chance to at least brief over everything that is publicly available.

https://cms.tinyml.org/wp-content/uploads/talks2022/Carlson-Kristofer-Brainchip.pdf

 

[DollazAndSense]

AFRL Information Directorate Awards $406M InSITE Deal to Northrop Grumman​

The Air Force Research Laboratory Information Directorate has awarded Northrop Grumman a $406 million contract to carry out the Intelligence Systems Infrastructure, Tools and Enhancements program.

InSITE is aimed at modernizing information collection and analysis capabilities through artificial intelligence, potentially enhancing decision-making by warfighters across domains.

Northrop plans to facilitate data exchange activities between various Department of Defense and intelligence community installations by introducing cloud-based applications.

According to Rebecca Torzone, vice president and general manager of combat systems and mission readiness, the contract award continues a 40-year partnership between Northrop and AFRL/RI and supports the achievement of space domain awareness and counter-space intelligence goals, Northrop said.

AFRL/RI’s mission is to develop technologies that ensure that the U.S. has an advantage in the areas of command, control, communications, computers and intelligence. Its specific aims include enabling warfighters to quickly receive critical information, establishing dominance in the cyber domain, facilitating the fusion of large quantities of data for decision-making purposes and developing complex C2 capabilities.

Some of the directorate’s areas of research and development include radio frequency, small unmanned aerial systems and neuromorphic and quantum computing.

Another concern for AFRL/RI is the implementation of reliable cloud connectivity for warfighters in remote areas. Norman Leach, an official within the directorate, called for building redundancy with hotspots and satellite communications to ensure that personnel can continue working on their missions.

Posted on January 18, 2023 by Noah Lowell

AFRL Information Directorate Awards $406M InSITE Deal to Northrop Grumman

The Air Force Research Laboratory Information Directorate has awarded Northrop Grumman a $406 million contract to carry out the Intelligence Systems Infrastructure, Tools and Enhancements program. InSITE is aimed at modernizing information collection and analysis capabilities through artificial...

potomacofficersclub.com

As we all know, AFRL selected Brainchip to utilise it’s neuromorphic technology in early 2022 for a radar solution. Glad to see neuromorphic computing is still a research focus for them in 2023.

 

[Rise from the ashes]

Thanks @sonofkong ... can someone please tell me how to download the document, can't seem to do it on linkedin.

Is this what you are looking for not sure if someone already sorted you out.

 

[FKE]

Anil shared the Brainchip benchmark post on his LinkedIn account.

He got three interesting likes.

https://www.linkedin.com/in/lakshmi-narayana-kethamreddy-86b6838

https://www.linkedin.com/in/rajeev-muralidhar

https://www.linkedin.com/in/eran-briman

 

[Rise from the ashes]

Probably shared here before. Much interest for me as I've been working on projects that tracks people by their wifi signal and also being able to see number of people in a room and where they are positioned.(through walls)

https://arxiv.o rg/abs/2301.00250

Do yourself a favour and turn off WiFi scanning in your improve location accuracy setting. And wifi when not in use (phone)
Lolz randomized Mac addresss

 

[stuart888]

AliveCor is/maybe/apparently testing the Brainchip Akida SNN solution for ECG medical wearables. Sounds like they concluded to move forward and test it on hardware. Notice the "future goal" and the first 2 references.

Conclusion, Limitation & Future works:
In this work, we have proved the efficacy of a new Peak based encoder along with different spiking networks for ECG classification by testing them on five widely varying ECG datasets.

Encoding performance of Peak encoder can be improved by removing the noise peaks in ECG signal by correlating the signal from another auxiliary low energy sensor like an accelerometer.

Our SNNs are not tested on real NC hardware such as Brainchip Akida [2] and Intel Loihi [19].

Our future goal is to benchmark the SNNs and encoders on those hardware.

References
[1] Alivecor kardia. URL https://clinicians.alivecor.com/our-devices/.
[2] Brainchip unveils the akida development environment. https://www.brainchipinc.com/newsme...hip-unveils-the-akida-development-environment, 2019.

https://clinicians.alivecor.com/our-devices/
The most sophisticated AI ever brought to personal ECG
Kardia devices can detect more arrhythmias than any other personal ECG—no wires, patches, or gels required.

https://openreview.net/pdf?id=V8yemRAs00-

 

[Rise from the ashes]

AliveCor is/maybe/apparently testing the Brainchip Akida SNN solution for ECG medical wearables. Sounds like they concluded to move forward and test it on hardware. Notice the "future goal" and the first 2 references.

Conclusion, Limitation & Future works:
In this work, we have proved the efficacy of a new Peak based encoder along with different spiking networks for ECG classification by testing them on five widely varying ECG datasets.

Encoding performance of Peak encoder can be improved by removing the noise peaks in ECG signal by correlating the signal from another auxiliary low energy sensor like an accelerometer.

Our SNNs are not tested on real NC hardware such as Brainchip Akida [2] and Intel Loihi [19].

Our future goal is to benchmark the SNNs and encoders on those hardware.

References
[1] Alivecor kardia. URL https://clinicians.alivecor.com/our-devices/.
[2] Brainchip unveils the akida development environment. https://www.brainchipinc.com/newsme...hip-unveils-the-akida-development-environment, 2019.

https://clinicians.alivecor.com/our-devices/
The most sophisticated AI ever brought to personal ECG
Kardia devices can detect more arrhythmias than any other personal ECG—no wires, patches, or gels required.

View attachment 27452

https://openreview.net/pdf?id=V8yemRAs00-

View attachment 27453

Excellent find Stuart I don't recall seeing this before.

 

[Rise from the ashes]

Easy on the fire @Esq.111
It burns image only for Esq I only expect to see 1 view on the counter
Well well this why you can't trust people online. 70 views