Thinking/Dreaming Cars…

Cars will soon have massive amounts of compute power to accomplish self driving functionality. They’ll have massive storage and continuous network connectivity. But when they are idle and being charged, cars can use the available compute power to reminisce their own driving and identify faults to improve in the future or update incorrect map data.

While driving, AI models have to perform at high speed and so the models have to be small and efficient. At rest, they can use much larger and hence more accurate models to identify mistakes. When mistakes are identified, the smaller runtime models can be fine tuned with the new data.

In some sense, this would be like humans dreaming at night. 🙂