A summary of our recent preprint on a holistic assessment of machine learning systems is now being featured by @mtlaiethics. Thanks @atg_abhishek!
https://t.co/thNm2IcEZr
Thanks CSET for the support! We measured a bunch of different performance metrics on over 500 models to see which techniques improve all of them. We found that ensembling and fine-tuning were the clear winners. See the preprint and other insights here: https://t.co/ciwhCorQts
How can we tell if an AI system is reliable—and what if a system is reliable on one metric, but not on another? Read our new blog post by @romeo_valentin_ and @hlntnr, summarizing research with @aicorso, @aiprof_mykel + other Stanford collaborators! ⬇️⬇️⬇️https://t.co/6PwKhx6x6T
⌛Last Call: Enroll by Friday, June 23 to save 20% on our new AI course!
The reliability & safety of #AI is more important than ever. Learn to create #trustworthyai by studying the fundamental concepts to design reliable AI systems. Enroll now: https://t.co/4rKn3w24DK
I'm super excited to announce that this summer I will be teaching a week-long course on building reliable and robust AI systems. You'll learn about the latest techniques being developed in the ML community to build, test and evaluate the most reliable AI systems. Check it out!
Our new 1-week #AI course starts July 24th! @aicorso will teach principles and techniques for building safe and robust machine learning models. Enroll by June 23rd for 20% savings: https://t.co/4rKn3w1wOc
This summer the Stanford Center for AI Safety hosted @hima_lakkaraju who presented a wonderful day-long workshop on ML Explainability. The recordings of that illuminating workshop are now being released in segments on youtube. Check it out! https://t.co/CoWfb84z9w
Thanks so much to @lilyxu0@rolf_comma_e and @MilindTambe_AI for hosting this great workshop! It was a privilege to learn about the amazing work going on at the intersection of AI and conservation.
We just had an absolutely awesome workshop on "AI assisted decision making for conservation" at @Harvard at @HCRCS; will publish our workshop report soon. Big thx to all the amazing participants!! Thx to my co-organizers @lilyxu0 & @rolf_comma_e for their leadership. #aiforgood
How can we design safer perception systems? Check out this short video on Risk-Driven Design of Perception Systems based on a #NeurIPS 2022 paper by SISLers @aicorso and @Sydney_Katz in collaboration with @InfAtEd: https://t.co/7nLV3zUfrR
Not all perception errors are created equal. We found that incorporating the risk of perceptual errors into the training process can improve overall system safety. Check out our #NeurIPS2022 paper “Risk-driven design of perception systems” (https://t.co/dJlPkmYWNT)
Wondering how we can test the limits of the AI algorithms used in autonomous vehicles? Check out this recent @StanfordHAI blog post about SISL's survey of black-box safety validation techniques led by @aicorso: https://t.co/vmrfd4NfQN
New Article: "A Survey of Algorithms for Black-Box Safety Validation of Cyber-Physical Systems" by Corso, Moss, Koren, Lee and Kochenderfer https://t.co/9ihO5vxUUO
With AI, validation of safety-critical autonomous systems is necessary before deployment.
Check out our latest blog post, on a model for black-box safety validation and a survey the literature to uncover the most effective validation techniques:
https://t.co/yxS9mxNx3s