Bravo
If ARM was an arm, BRN would be its biceps💪!
Hi Tech,
Great find!
One exciting thing is the priority date of 20220429, just 18 months ago, indicating a potential ongoing relationship. After all, who does CNN2SNN better than BRN?
... and a couple of the inventors are familiar faces ...
US2023325001A1 ACOUSTIC SYSTEM AND METHOD BASED GESTURE DETECTION USING SPIKING NEURAL NETWORKS 20220429
Inventors: GIGIE ANDREW [IN]; GEORGE ARUN [IN]; KUMAR ACHANNA ANIL [IN]; DEY SOUNAK [IN]; PAL ARPAN [IN]
View attachment 49937
Conventional gesture detection approaches demand large memory and computation power to run efficiently, thus limiting their use in power and memory constrained edge devices. Present application/disclosure provides a Spiking Neural Network based system which is a robust low power edge compatible ultrasound-based gesture detection system. The system uses a plurality of speakers and microphones that mimics a Multi Input Multi Output (MIMO) setup thus providing requisite diversity to effectively address fading. The system also makes use of distinctive Channel Impulse Response (CIR) estimated by imposing sparsity prior for robust gesture detection. A multi-layer Convolutional Neural Network (CNN) has been trained on these distinctive CIR images and the trained CNN model is converted into an equivalent Spiking Neural Network (SNN) via an ANN (Artificial Neural Network)-to-SNN conversion mechanism. The SNN is further configured to detect/classify gestures performed by users.
[0089] The active power consumption of a neuromorphic hardware is mainly contributed by the spiking network's total number of synaptic operations (SOP). Following (7) and the method mentioned in Sorbaro et al. (e.g., refer “Martino Sorbaro, Qian Liu, Massimo Bortone, and Sadique Sheik, “Optimizing the energy consumption of spiking neural networks for neuromorphic applications,” Frontiers in Neuroscience, vol. 14, pp. 662, 2020.”), total number of synaptic operation for the SNN of the system 100 is found to be ˜35M while that for the CNN is ˜95M (considering matrix multiplication only). This converted SNN can be implemented on neuromorphic platforms such as Brainchip Akida (e.g., refer “Brainchip unveils the akidatm development environment,” https://www.brainchipinc.com/news-m...chip-unveils-the-akida-developmentenvironment, 2019″), Intel® Loihi (e.g., refer “Mike Davies. et. al, “Advancing neuromorphic computing with loihi: A survey of results and outlook,” Proceedings of the IEEE, vol. 109, no. 5, pp. 911-934, 2021.”), etc. to achieve further power benefit (˜100×).
One of the inventors is Sunak Dey
I recall he had some great stuff to say
About how Akida is getting on
Being betterer than NVIDIA Jetson
Last edited: