Behind door 2 of our 𝐚𝐝𝐯𝐞𝐧𝐭 𝐜𝐚𝐥𝐞𝐧𝐝𝐚𝐫 𝐨𝐟 𝐢𝐧𝐧𝐨𝐯𝐚𝐭𝐢𝐨𝐧 is a topic that could unlock many future functionalities… | 24 comments on LinkedIn
www.linkedin.com
Yes Markus…… ! Ready when you are bud !
View attachment 73838
"
Together with different research partners, my team is breaking new ground in our research into artificial neural networks that mimic the workings of the human brain. The conventional von Neumann architecture based on processing binary signals supported by a separate memory bank is replaced by spiking neurons and synapses that use event-based processing and local memory. The former only sends signals when something actually happens (events), while the latter eliminates the bottleneck between processing and memory."
Mercedes has set up an AI research project with the University of Waterloo, so that would be one of the research partners.
Interesting choice of language:
"T
he conventional von Neumann architecture based on processing binary signals supported by a separate memory bank is replaced by spiking neurons and synapses that use event-based processing and local memory."
Analog neurons are sometimes described as providing
in-memory compute.
Akida has sometimes been descibed as providing
near-memory compute.
Local memory is broad enough to encompass both.
We know Mercedes has been playing away with Nvidia and Qualcomm.
Qualcomm is developing hybrid digital/analog NNs:
US2023297335A1 Hybrid Compute-in-Memory 20220315
I haven't looked to see if Nvidia does compute-in-memory.
Although we've heard nothing since EQXX, we have not seen anything stating that the Brainchip/Mercedes SNN cooperation has ended.
We also know that Mercedes is implementing the SDV concept
We also know Brainchip introduced an algorithm product line this year.
Does the Mercedes SDV archtecture jigsaw have a TENNs-shaped hole in it?