Links for 2023-04-30
A completely autonomous model-free reinforcement learning system capable of learning aggressive driving policies in a variety of outdoor and indoor environments from pixels entirely in the real world matching human performance after just 20 minutes of practice.
The system learns the concept of a “racing line,” finding a smooth path through the lap and maximizing its speed through tight corners and chicanes. The robot learns to carry its speed into the apex, then brakes sharply to turn and accelerates out of the corner, to minimize the driving duration. With a low-friction surface, the policy learns to over-steer slightly when turning, drifting into the corner to achieve fast rotation without braking during the turn. In outdoor environments, the learned policy is also able to distinguish ground characteristics, preferring smooth, high-traction areas on and around concrete paths over areas with tall grass that impedes the robot’s motion.
Project page: https://sites.google.com/view/fastrlap
Video: https://youtu.be/eZATlf0ybXk
Simeon Campos: “…long-term memory…learning on the fly and being able to leverage what you learned in the past…I expect this to be solved in 2 years…superhuman coding AI…I expect this to also happen with like 60% chance within…2 years.” https://www.youtube.com/watch?v=LFeiBIDuvP8
Accelerating Chip Design With GPUs, And Adding AI To Push It Further https://www.nextplatform.com/2023/04/20/accelerating-chip-design-with-gpus-and-adding-ai-to-push-it-further/
Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System https://arxiv.org/abs/2304.13343
ChatVideo: A Tracklet-centric Multimodal and Versatile Video Understanding System https://www.wangjunke.info/ChatVideo/
“REMEDIS shows significant improvement in data-efficient generalization across medical imaging tasks and modalities with a 3–100x reduction in site-specific data for adapting models to new clinical contexts and environments...These results suggest that REMEDIS has the potential to significantly accelerate the development of ML systems for medical imaging...” https://ai.googleblog.com/2023/04/robust-and-efficient-medical-imaging.html
Keeping Large Language Models From Running Off The Rails: "CalypsoAI, a five-year-old startup launched by people who worked at such places as NASA, DARPA, and the Department of Defense...unveiled its Moderator product to help organizations get control of the LLMs...Now comes Nvidia, a longtime vocal proponent and enabler of AI in the enterprise that has rolled out NeMo Guardrails" https://www.nextplatform.com/2023/04/26/keeping-large-language-models-from-running-off-the-rails/
“We propose ray conditioning, a lightweight and geometry-free technique for multi-view image generation.” https://ray-cond.github.io/
Efficient Automation of Neural Network Design: A Survey on Differentiable Neural Architecture Search https://arxiv.org/abs/2304.05405
200 Concrete Problems In AI Interpretability https://docs.google.com/spreadsheets/d/1oOdrQ80jDK-aGn-EVdDt3dg65GhmzrvBWzJ6MUZB8n4/edit#gid=0
COOL-MC: A Comprehensive Tool for Reinforcement Learning and Model Checking https://arxiv.org/abs/2209.07133
Thread of what could happen over the next 5 years as AI makes it practically impossible for ordinary folks to distinguish truth from fiction. https://twitter.com/dschwarz26/status/1652373688698945536
A Rolls-Royce nuclear Micro-Reactor for the Moon https://www.rolls-royce.com/media/our-stories/discover/2023/uk-space-agency-backs-rolls-royce-nuclear-power-for-moon-exploration.aspx
Germany's Rheinmetall introduces a UAV drone-carrier concept - a drone mothership carrying up to 8 small kamikaze drones. https://www.youtube.com/watch?v=Cfi7sdGnJPY
Elites Publicly Condemn What They Privately Support https://www.psychologytoday.com/us/blog/sex-and-civilization/202304/elites-publicly-condemn-what-they-privately-support