BRN Discussion Ongoing

[Fullmoonfever]

Have this Fraunhofer student connection who has set up the following on GitHub updated 2 days ago.

Wonder if stand alone work or part of a project there?

A comprehensive pipeline for training, quantizing, and converting neural networks to utilise Brainchip's Akida neuromorphic processor to classify ECG Signals.

GitHub - mer0o/Akida-ECG-Classification: A comprehensive pipeline for training, quantizing, and converting neural networks to utilise Brainchip's Akida neuromorphic processor to classify ECG Signals.

A comprehensive pipeline for training, quantizing, and converting neural networks to utilise Brainchip's Akida neuromorphic processor to classify ECG Signals. - mer0o/Akida-ECG-Classification

github.com

https://www.linkedin.com/in/omar-emadeldeen

Omar Emadeldeen​

Studentische HiWi @ Fraunhofer HHI | Communications Engineering Student @GIU Berlin​

Fraunhofer Heinrich Hertz Institute HHI German International University Berlin​

 

[FJ-215]

Can we at least finish with some emerald green for St. Patrick's day.

A nice contrast to the Kilkenny red ale I'm about to open.

Happy St. Pat's all

 

[Pom down under]

Can we at least finish with some emerald green for St. Patrick's day.

A nice contrast to the Kilkenny red ale I'm about to open.

Happy St. Pat's all

 

[Diogenese]

Just another intern helping with our Akida including TENNs and LLM RAG.

https://www.linkedin.com/in/keith-hoffmeister-18a67b221

Keith Hoffmeister​

UCLA Computer Science and Engineering​

BrainChip University of California, Los Angeles​

Los Angeles, California, United States Contact Info​

​

Machine Learning C++ Intern​

BrainChip​

Jun 2024 - Dec 2024 7 months
Laguna Hills, California, United States
Developed an in-depth understanding and code documentation of the Akida Engine Library's C++ API using Sphinx. Created AI powered anomaly detection demos in C++ and Python to showcase the capabilities of Akida using PyTorch, XTensor, Qt, Gradio, WiringPi, and Adafruit. Created a C++ library for pre and post processing of different data types to be used in Akida machine learning demos. Converted Brainchip’s PyTorch TENN state space model for their LLM RAG demo from Python to C++.

So Keith was adapting Akida for C++ systems:

"Created a C++ library for pre and post processing of different data types to be used in Akida machine learning demos. Converted Brainchip’s PyTorch TENN state space model for their LLM RAG demo from Python to C++."


At first, I thought the obvious target was Prophesee:
https://docs.prophesee.ai/stable/get_started/get_started_cpp.html

"In this Section, we will create a minimal sample to get started with Metavision SDK C++ API. The goal of this sample is to introduce in the simplest possible way some basic concepts of Metavision SDK and create a first running example."

But C++ is a very popular language (dare one say "semi-ubiquitous?"), and LLM RAG isn't really of interest to Prophesee applications, so this could open additional opportunities.

Tony Lewis'pre-Embedded World Linkedin post:

https://www.linkedin.com/feed/update/urn:li:activity:7303922948698947584/
...
2) State-Space Models: A paradigm-shifting metaphor for brain processing, capturing the essences of brain-like processing in a predictable engineering framework. This approach has the potential to redefine how we understand and replicate brain-like intelligence. This represents the next generation of BrainChip products.

At Nuremberg I will be carrying under my arm the what I believe is our most valuable intellectual property to date: an extreme-edge Large Language Model (LLM) that redefines the boundaries of AI deployment. You will be able to see the FPGA embodiment at EW at our booth."

The EW demo of NLP (LLM+RAG?) that I saw did not go into much detail other than the Akida/TENNs ASIC would be 10 times faster than the FPGA,.

Looking forward to the Akida 2/TENNs ASIC.

 

[Fullmoonfever]

So Keith was adapting Akida for C++ systems:

"Created a C++ library for pre and post processing of different data types to be used in Akida machine learning demos. Converted Brainchip’s PyTorch TENN state space model for their LLM RAG demo from Python to C++."


At first, I thought the obvious target was Prophesee:
https://docs.prophesee.ai/stable/get_started/get_started_cpp.html

"In this Section, we will create a minimal sample to get started with Metavision SDK C++ API. The goal of this sample is to introduce in the simplest possible way some basic concepts of Metavision SDK and create a first running example."

But C++ is a very popular language, and LLM RAG isn't really of interese to Prophesee applications, so this could open additional opportunities.

Tony Lesis'pre-Embedded World Linkedin post:

https://www.linkedin.com/feed/update/urn:li:activity:7303922948698947584/
...
2) State-Space Models: A paradigm-shifting metaphor for brain processing, capturing the essences of brain-like processing in a predictable engineering framework. This approach has the potential to redefine how we understand and replicate brain-like intelligence. This represents the next generation of BrainChip products.

At Nuremberg I will be carrying under my arm the what I believe is our most valuable intellectual property to date: an extreme-edge Large Language Model (LLM) that redefines the boundaries of AI deployment. You will be able to see the FPGA embodiment at EW at our booth."

The EW demo of NLP (LLM+RAG?) that I saw did not go into much detail other than the Akida/TENNs ASIC would be 10 times faster than the FPGA,.

Looking forward to the Akida 2/TENNs ASIC.

Cheers. So the language is going to be important to end use obviously and you are saying the C++ may open additional avenues.

Phone processors or home appliaces use what language primarily...you know?

Only cause the LLM demo this guy appeared to have helped with was liked by him on his LinkedIn and was the one which had the fridge etc on the screen.

 

[Iseki]

So Keith was adapting Akida for C++ systems:

"Created a C++ library for pre and post processing of different data types to be used in Akida machine learning demos. Converted Brainchip’s PyTorch TENN state space model for their LLM RAG demo from Python to C++."


At first, I thought the obvious target was Prophesee:
https://docs.prophesee.ai/stable/get_started/get_started_cpp.html

"In this Section, we will create a minimal sample to get started with Metavision SDK C++ API. The goal of this sample is to introduce in the simplest possible way some basic concepts of Metavision SDK and create a first running example."

But C++ is a very popular language (dare one say "semi-ubiquitous?"), and LLM RAG isn't really of interest to Prophesee applications, so this could open additional opportunities.

Tony Lewis'pre-Embedded World Linkedin post:

https://www.linkedin.com/feed/update/urn:li:activity:7303922948698947584/
...
2) State-Space Models: A paradigm-shifting metaphor for brain processing, capturing the essences of brain-like processing in a predictable engineering framework. This approach has the potential to redefine how we understand and replicate brain-like intelligence. This represents the next generation of BrainChip products.

At Nuremberg I will be carrying under my arm the what I believe is our most valuable intellectual property to date: an extreme-edge Large Language Model (LLM) that redefines the boundaries of AI deployment. You will be able to see the FPGA embodiment at EW at our booth."

The EW demo of NLP (LLM+RAG?) that I saw did not go into much detail other than the Akida/TENNs ASIC would be 10 times faster than the FPGA,.

Looking forward to the Akida 2/TENNs ASIC.

"Looking forward to the Akida 2/TENNs ASIC."

Time to call in all favours and make this happen.

 

[Pom down under]

Looking forward to the Akida 2/TENNs ASIC.

yea and some revenue from it in 2030s

 

[Frangipani]

Volkswagen decided to sell their in-house currywurst in supermarkets… and some people here are complaining about BrainChip’s t-shirt era in the beginning…

Volkswagen: Currywurst des Konzerns bald im Supermarkt

Im Kerngeschäft lief es für Volkswagen zuletzt denkbar schlecht. Dafür gibt es den Kantinen-Bestseller des Konzerns bald auch für Zuhause.

www.faz.net

That move by crisis-hit German carmaker VW is far from being an act of desperation, though, like some readers might infer from the above comment by our forum’s top Germany-bashing poster: While VW’s car sales have been declining, its signature sausage is in fact wildly popular, sold internationally and even has its own Volkswagen Original Part number - 199 398 500 A.

The FAZ title is somewhat misleading, though, as the iconic sausage as such is already being sold in select supermarkets as well as online - the soon-to-be-launched novelty is going to be a ready-to-eat-once-heated version, cut up and slathered in VW’s secret recipe sauce (a special ketchup produced exclusively for VW by Develey and listed as Volkswagen Originalteil 199 398 500 B.)

https://de.m.wikipedia.org/wiki/Volkswagen-Currywurst (German)

https://en.m.wikipedia.org/wiki/Volkswagen_currywurst (English)

German carmaker Volkswagen's in-house sausage brand sets sales record

The core car business is weakening, but German auto giant Volkswagen (VW) was able to set a sales record for an unexpected product in 2024: sausages. Volkswagen's line of sausages saw 8,552,000 units sold in 2024 in canteens and supermarkets. That includes VW's packaged currywurst, a much-loved...

www.yahoo.com

“For the flagship VW brand, the sausages are by far the best-selling product. In 2024, the 8.5 million sausages compared to around 5.2 million cars and vans with the VW logo.”