Neuromorphic Computing: Kapoho Point Development Board and Lava Software Framework
This demo introduces Intel’s latest neuromorphic innovation, Kapoho Point. This stackable system features eight Loihi chips in a compact 4x4-inch form factor with an Ethernet interface. We will showcase the new development board and demonstrate a basic workload on a functional neuromorphic system with a Lava software framework.
Transcript
[MUSIC PLAYING]
So today we are showcasing Kapoho Point. This is our new hardware system that has eight Loihi 2 neuromorphic chips. We really foresee these being useful for a variety of commercial applications where you need AI computing at the edge, essentially very low-power computing.
In particular anything where you need to learn on the fly, Loihi 2's learning engine is actually capable of reprogramming synapses in real time. And so this is kind of a fundamentally different AI computing paradigm where you can learn from very few examples. So today, we have a sample deep-learning workload. This is called the PilotNet task. Essentially, the goal is to predict steering angle from dashboard images while you're driving.
This particular network was trained using PyTorch on a CPU. But then the network configuration were transferred to our Loihi neurocores via the open-source LAVA software framework. And we're running that network in inference mode very efficiently, low energy on this chip.