BRN Discussion Ongoing

[Bravo]

Howdy All,

I was reading this article published 17 December (see below) about Eutelsat having ordered 100 broadband satellites from Airbus Defence and Space to replenishing its OneWeb low Earth orbit (LEO) constellation. The first batches of satellites are due to be completed before the end of 2026.

This caused me to ruminate on the possibility of our technology being incorporated into these new satellites. And it led me to wonder whether the work EDGX are involved in could also possibly be linked in some way to this????

Certainly some interesting partnerships in the low-Earth orbit (LEO) satellite space, some of which overlap.

• In August BrainChip signed two agreements totalling €190,000 with Frontgrade Gaisler and Airbus Defence and Space. These partnerships aim to leverage Akida technology in space applications, in line with a European Space Agency initiative to develop ultra-low-power AI for future space missions.
• There is an existing partnership between the European Space Agency (ESA) and Eutelsat ,who have merged with OneWeb.
• A few months ago Nandan Nyampally liked a Linkedin Post from Relativity Space announcing a Launch Services Agreement between OneWeb and Relativity Space. Relativity Space also work closely with NASA.
• Softbank and OneWeb signed a partnership agreement in 2023 to work toward deploying a low-Earth orbit (LEO) satellite network and competing with rivals like SpaceX's Starlink and Amazon's Project Kuiper.
• EDGX signed its first contract with the European Space Agency in January 2024. They are also working on an onboard neuromorphic AI data processing unit designed for smallsats through the European Space Agency (ESA)
• In October 2023 BrainChip and EDGX announced a collaborative agreement to develop data processing units for very demanding environments.

Eutelsat orders 100 LEO satellites to replenish OneWeb constellation

Eutelsat has ordered 100 broadband satellites from Airbus Defence and Space to start replenishing its OneWeb low Earth orbit constellation in a few years.

spacenews.com

 

[Food4 thought]

On my way to the long white cloud
Waiheke island infact
I too am feeling incredibly grateful that I am
1 a long time holder
2 have a good accumulation of shares
And with the amount of positive information that is being published publicly everyone should know that they are in safe hands.

Ps.
I would like to personally thank
For giving me double green on my portfolio for Christmas what a present
Well done BrainChip

 

[Fullmoonfever]

Did a quick search on Sounak and couldn't find this particular post but maybe I missed it.

Anyway, all pretty positive and some good comments to the post too.

Sometimes back, our work focus was on the use case of "In-car Driver Drowsiness Detection". | Sounak Dey

Sometimes back, our work focus was on the use case of "In-car Driver Drowsiness Detection". We did the usual tasks of designing a Spiking Neural Network (SNN) and then deploying it on BrainChip Akida 1000 neuromorphic board. The model has achieved upto 92% accuracy on four classes. Energy...

www.linkedin.com

Sounak Dey
Senior Scientist, Neuromorphic Computing and Spiking Neural Network, TCS Research
3w Edited

Sometimes back, our work focus was on the use case of "In-car Driver Drowsiness Detection". We did the usual tasks of designing a Spiking Neural Network (SNN) and then deploying it on BrainChip Akida 1000 neuromorphic board. The model has achieved upto 92% accuracy on four classes. Energy consumption is only 16.2 mJ/frame and latency is 46.5 ms/frame. So far, "Nice to have" stuffs, right? Results meet the expectation from Neuromorphic computing systems...and they perfectly align with requirements of battery driven cars. Then came the twist : We thought to train the model with three classes, deploy it and make it learn one new class on-chip. Found that the accuracy drops by 7% only. Wow !! Let's go harder. Trained the model with two classes only and let it learn rest two classes on-chip. This time, accuracy drops by max 21%. Obviously this needs to be investigated further and across different types of data. We are on it. However, this essentially opens up avenues for post deployment customization and personalization of SNN models at edge which is kinda great. And excellent to have for end customers. Meanwhile the work done so far has somehow been accepted at #ICPR 2024, Kolkata and my colleague Ajoy Dey will be presenting this there. Special thanks to Chetan Kadway for his significant contributions in shaping this work. Arijit Mukherjee Arpan Pal Alf Kuchenbuch Gilles Bézard Tata Consultancy Services - Research

• 

Kaustab Pal
Sr. Engineer @ Qualcomm
3w

Very interesting. Will on-chip learning for longer duration have any significant effect on accuracy?
Like
Reply
1 Reaction

Laurent Hili

3w

Indeed very promising avenues for edge training. Looking forward for further posts on the same topic
Like
Reply
4 Reactions

Soubhagya Paul
Student at Heritage Institute of Technology (MCA)
2w

I explored it a little bit it's totally worth it .
Like
Reply

Chetan Kadway
Data Scientist | ML Researcher | Computer Vision | Automotive | SpaceTech | Edge AI | Neuromorphic Computing | Azure ML | ML-Ops
3w

Thanks Sounak Dey & BrainChip team. It was a fun quick hypothesis to explore. I highly appreciate Ajoy Dey for conducting all the experiments. He learned & contributed very quickly.

 

[Space Cadet]

https://bascomhunter.com/products-services/bh-tech/digital-products/neuromorphic-processors/?msclkid=9b8eb1fa672c1b330741d69db107251e&utm_source=bing&utm_medium=cpc&utm_campaign=**LP DSA - All Pages&utm_term=bascomhunter&utm_content=All Pages

Bascom Hunter

SC

Hi Dio,

If this is in your wheelhouse, could you give us a run-down on how we could possibly fit into these photonic neural networks? Previously when I saw mention of photonics I usually disregarded and moved on. Just wondering I'd we ould be a co processor to this?
SC

 

[Diogenese]

Hi Dio,

If this is in your wheelhouse, could you give us a run-down on how we could possibly fit into these photonic neural networks? Previously when I saw mention of photonics I usually disregarded and moved on. Just wondering I'd we ould be a co processor to this?
SC

Hi SC,

BH's photonic processor uses analog. I guess that means that at some stage they add a lot of light signals to produce a proprtional electrical signal, so they will need to overcome the manufacturing variations. I don't know about manufacturing variations (MV) in the optical realm (getting each LED to produce the same amount of light output for the same number of photons in the input), but the analog electrical section will still suffer from the MVs.

BH do use MACs, but in the optical realm.

https://bascomhunter.com/products-services/bh-tech/digital-products/neuromorphic-processors/?msclkid=9b8eb1fa672c1b330741d69db107251e&utm_source=bing&utm_medium=cpc&utm_campaign=**LP DSA - All Pages&utm_term=bascomhunter&utm_content=All Pages

Our nation’s most demanding applications require performance and accuracy at the speed of light. Missions like Automatic Target Recognition (ATR), Hypersonic Vehicle Detection, and Autonomous Navigation all require real-time processing and inferencing on countless high bandwidth multi-phenomena sensors in parallel, rendering modern Machine Learning (ML) processors like CPUs, TPUs, and GPUs inert. Bascom Hunter specializes in leveraging Silicon Photonics to offload the most compute-intensive tasks in ML, like Matrix Vector Multiplication (MVM). By leveraging analog optical processing, our processors avoid the classic electronic bottlenecks to achieve classification latencies as low as 200 picoseconds; translating to 0.16 Terra Multiply-Accumulate operations per second.


Bascomb Hunter Photonic NN

US2021406668A1 Integrated Neuromorphic Computing System 20200629

A hybrid neuromorphic computing device is provided, in which artificial neurons include light-emitting devices that provide weighted sums of inputs as light output. The output is detected by a photodetector and converted to an electrical output. Each neuron may receive output from one or more other neurons as initial input. Interconnects between neurons may be optical, electrical, or a combination thereof. The neurons also may provide imaging sensor and/or display capabilities.


BH have been working with photonics for over 14 years:

US2012033966A1 OPTICAL LEAKY INTEGRATE-AND-FIRE NEURON 20100916

An optical system includes an optical integrator, a readout mechanism, and an optical thresholder. The optical integrator is configured to perform temporal integration of an optical input signal having a first wavelength received at an input. The readout mechanism is coupled to the optical integrator and provides optical signals having a second wavelength to the optical integrator for measuring a state of the optical integrator. The optical thresholder is coupled to an output of the optical integrator and is configured to receive a signal representing a temporal integration of the optical input signal from the optical integrator and produce an optical signal identifying if an amplitude of the signal representing the temporal integration of the optical input signal is above or below a threshold value.


In my view, , using photonics for Akida would involve a ground-up redesign.

 

[Space Cadet]

Hi SC,

BH's photonic processor uses analog. I guess that means that at some stage they add a lot of light signals to produce a proprtional electrical signal, so they will need to overcome the manufacturing variations. I don't know about manufacturing variations (MV) in the optical realm (getting each LED to produce the same amount of light output for the same number of photons in the input), but the analog electrical section will still suffer from the MVs.

BH do use MACs, but in the optical realm.

https://bascomhunter.com/products-services/bh-tech/digital-products/neuromorphic-processors/?msclkid=9b8eb1fa672c1b330741d69db107251e&utm_source=bing&utm_medium=cpc&utm_campaign=**LP DSA - All Pages&utm_term=bascomhunter&utm_content=All Pages

Our nation’s most demanding applications require performance and accuracy at the speed of light. Missions like Automatic Target Recognition (ATR), Hypersonic Vehicle Detection, and Autonomous Navigation all require real-time processing and inferencing on countless high bandwidth multi-phenomena sensors in parallel, rendering modern Machine Learning (ML) processors like CPUs, TPUs, and GPUs inert. Bascom Hunter specializes in leveraging Silicon Photonics to offload the most compute-intensive tasks in ML, like Matrix Vector Multiplication (MVM). By leveraging analog optical processing, our processors avoid the classic electronic bottlenecks to achieve classification latencies as low as 200 picoseconds; translating to 0.16 Terra Multiply-Accumulate operations per second.


Bascomb Hunter Photonic NN

US2021406668A1 Integrated Neuromorphic Computing System 20200629

View attachment 74755


View attachment 74756 A hybrid neuromorphic computing device is provided, in which artificial neurons include light-emitting devices that provide weighted sums of inputs as light output. The output is detected by a photodetector and converted to an electrical output. Each neuron may receive output from one or more other neurons as initial input. Interconnects between neurons may be optical, electrical, or a combination thereof. The neurons also may provide imaging sensor and/or display capabilities.


BH have been working with photonics for over 14 years:

US2012033966A1 OPTICAL LEAKY INTEGRATE-AND-FIRE NEURON 20100916

View attachment 74757


An optical system includes an optical integrator, a readout mechanism, and an optical thresholder. The optical integrator is configured to perform temporal integration of an optical input signal having a first wavelength received at an input. The readout mechanism is coupled to the optical integrator and provides optical signals having a second wavelength to the optical integrator for measuring a state of the optical integrator. The optical thresholder is coupled to an output of the optical integrator and is configured to receive a signal representing a temporal integration of the optical input signal from the optical integrator and produce an optical signal identifying if an amplitude of the signal representing the temporal integration of the optical input signal is above or below a threshold value.


In my view, , using photonics for Akida would involve a ground-up redesign.

Thanks mate. Appreciate your time.

SC

 

[HappyChippy]

Found this interesting.

3U VPX SNAP Card​

3U VPX SNAP Card

bascomhunter.com

 

[Shadow59]

Morningstar have conceded us a few more cents for the end of the year

 

[DingoBorat]

Looks like a very strong opening.

Shut up Larry...

 

[Shadow59]

33.5 and no gap!!! at open! Now it can run!

 

[buena suerte :-)]

Looks like a very strong opening.

Shut up Larry...

Yep very nice ....

WOW!!! Loving it

 

[TheDrooben]

Looks like a very strong opening.

Shut up Larry...

No gap @DingoBorat .........let's go.

 

[Gemmax]

No gap @DingoBorat .........let's go.

Run Bravo!!

 

[7für7]

Hope you all had a nice Christmas time!


and now back to business!
BOOOOOOOOM

 

[DingoBorat]

Very very strong buying.

We still don't know the catalyst..

 

[HopalongPetrovski]

Its low volume, but damn, nothing very much showing in our way upwards.
Another announcement now and we'd explode.
Bring it, BrainChip!

 

[DingoBorat]

Very very strong buying.

We still don't know the catalyst..

Somebody get Dave a fresh pair of underpants..

 

[Shadow59]

Somebody get Dave a fresh pair of underpants..

Poor Dave!

 

[Pom down under]

Just got of the plane at Sinagapore heading to Veitnam and just looked at the SP

 

[Bravo]

Run Bravo!!

Don't worry. I got this!

This is no mean feat after all the pavlova I ate yesterday, trust me.