Last weeks there has been a lot of interesting developments in the machine learning community that I think are relevant for Tesla and their future neural architecture.
The first is a paper by facebook that does end-end object detection:
The second is GPT-3 by OpenAI that does few shot multi task learning, improving SOTA on multiple tasks while not even being tuned on them. It does this by being absolute massive:
Any thoughts on how Tesla will implement this into their neural networks? Should we expect another rewrite of their NN architecture in another year once their 3D labelling Birds Eye View temporal architecture is out:
Edit: ignore title, should be object detection... =)
The first is a paper by facebook that does end-end object detection:
The second is GPT-3 by OpenAI that does few shot multi task learning, improving SOTA on multiple tasks while not even being tuned on them. It does this by being absolute massive:
Any thoughts on how Tesla will implement this into their neural networks? Should we expect another rewrite of their NN architecture in another year once their 3D labelling Birds Eye View temporal architecture is out:
Edit: ignore title, should be object detection... =)
