Summary: Can one use ML to verify ML-based systems? This post claims the answer is mostly “no”: You mainly have to use other system verification methodologies. However, some ML-based techniques may still be quite useful. How does one verify ML-based systems? A previous post in this series claimed that the “right” way is CDV: Essentially, … More Using Machine Learning to verify Machine Learning?
Summary: This post describes in general terms the problem of “using ML for coverage maximization”, explains why it is important for CDV and for fuzzing, and gives some references. My first post in the “ML and verification” series talked about verifying ML-based systems (lots more to say about that). This post talks about the other … More Machine Learning for Coverage Maximization
Summary: This post – part of a series about Machine Learning (ML) and verification – discusses the hard problem of verifying ML-based systems, and how “Explainable AI” might help. I can think of three interesting topics whose title contains both “ML” and “verification”: Verifying ML-based systems Using a verification infrastructure to train ML-based systems Using … More Machine Learning verification and Explainable AI
Summary: This post tries to fit Fault Tree Analysis and CDV-based “normal” bug-finding into the same conceptual framework. FTA and CDV have a challenging, uneasy relationship with each other. In a nutshell (details below): FTA (and similar techniques) is a good, established way to reason about failures and reliability, but (1) it depends on humans … More About faults and bugs
Summary: In this post I discuss the problem (common in autonomous systems verification) of checking things which don’t have “sharp” Boolean rules. I also suggest some (imperfect) solutions. Say we want to create a growing catalog of Autonomous Vehicles scenarios, and use it to test our AV, as I suggested here. We are talking system-level … More Checking probabilistic components
Summary: This post discusses one of the verification implications of the next 20 years of automation on verification: The problem of trying to predict how people will react to change. A previous post (and another one) discussed the next 20 years of automation and their general implications. This follow-up post discusses one of the … More Future of verification: Better ways to predict behavior
Summary: This post talks about some implications of the Tesla autopilot crash, claims that some autonomous vehicle fatalities are unavoidable, and suggests some ways to minimize them. By now you have probably heard about the Tesla fatal autopilot accident, causing the tragic death of a Tesla enthusiast. There is a lot of discussion about what … More The Tesla crash, Tsunamis and spec errors
Whatever you think of my Stuttgart symposium trip report, at least I made the effort to attend it and report back. In this post I sink even lower, reporting from a conference I did not attend – the ICRA workshop on verification of autonomous systems. I did, however, watch the full 4.5 hour video (warning: … More The ICRA 2016 verification workshop
Summary: This post discusses some recent articles about automation and unemployment, and my reactions to them. Following my post about automation and unemployment, I got several comments saying I was too optimistic, and (fewer) saying all this would never happen. I am still trying to “get the future right” (I know this is really impossible). … More Misc stuff: Automation and unemployment
Summary: This trip report describes my impressions from the Second Autonomous Vehicles Test & Design Symposium in Stuttgart. Short version: The design side is moving fast. The verification side has improved, but has a long way to go. Note: In my last post I talked about what automation will do to this planet over the … More Finding bugs in autonomous vehicles: My impressions from the second Stuttgart symposium
