sayan mitra

Great conversation today on @21stShow (@ILPublicMedia) on @Waymo's Illinois expansion — covering how AV perception and decision-making work, what safety data shows, & what deployment in cities like Chicago demands. Replay on 📻 7pm, or listen anytime: will.illinois.edu/21stshow/story… #AI

@YiMaTweets Hybrid Systems theory has been around for 30+ years with dedicated conferences and journals #HSCC, #NAHS, with new infusion of ideas from gradient-based methods.

The maximum rate reduction, or information gain, criterion aims to do that, but only at the distribution of samples (trajectories or sequences) level. This is in a similar spirit of approximating nonlinear systems with hybrid linear systems. But individual trajectories are not necessarily flat. I believe the proper level of linearisation is at the distribution (a group of similar samples) level. Linearizing every individual trajectory or sequence might be too much to ask… I believe a good representation is one that the learned code book allows (sparse) linear superposition among the atoms of the dictionary…

At first glance this is a totally reasonable perspective. Training PhD students is a duty! But consider this — *effectively* advising a PhD student over a 5-year period is well over 1,000 hours of work, not to mention bringing in hundreds of thousands of dollars in grants. Professors will do some things for mostly altruistic reasons (peer review) but the time commitment for advising is not something that's reasonable to ask of someone without some form of compensation. So there are two options. One is to make advising a job requirement. Unfortunately this doesn't work, because the *quality* of advising is unobservable and can't be quantified by metrics, leading to a race to the bottom. The other option is the current system — advising helps advance the professor's research agenda because PhD students do most of the work, so they take on students voluntarily. Which means it's important to ask if this subtle alignment of incentives will continue despite advancing AI capabilities. Academia has many such "subtle alignments of incentives" that the system relies on in order to function — rarely articulated, poorly understood, and fragile. Maybe the advisor-advisee relationship in CS will survive the AI transition, as @sayashk predicts, but many processes and structures will surely break. Best to rethink the system now, before it's too late.

In system theory, it is called "linearization"... which has been studied and used for decades. Honestly, folks, there is no need to invent or introduce any new terminology. Remember, there is rarely anything new under the sun...

@roydanroy Congrats Dan!

Big announcement time... Today is my last day as Research Director at the Vector Institute. It has been my incredible privilege over the past 2.5 years to serve the Vector community and help build an institution that supports world-class ML research and real-world impact.

My corner of this website seems to be broken, but in any case:

AI @ECEILLINOIS has a new page #topic=all&group=772" target="_blank" rel="nofollow noopener">ece.illinois.edu/research/cross…

The work is led by PhD students Chenxi Ji, Yangge Li, and Ziangru Zhong, in collaboration with my amazing colleague Huan Zhang in @ECEILLINOIS .

Abstract Rendering lets us formally verify statements such as: – “No stop sign is ever detected by this classifier as the camera moves along a 5 m path,” or – Identify the range of viewing angles over which a neural pose estimator stays within a required error tolerance.

Our #AbstractRendering work appearing as a #NeurIPS2025 Spotlight. Fundamental question: how can to compute all images a scene can produce as the camera moves? We show how these Abstract Images can be computed and used for certifying visual models.

Paper deadline in mid Nov. St. Malo, France is going to be amazing!

Big changes announced for CPS-IoT Week 2026, with ICCPS and HSCC coming together and a new AI-Autonomy track. hscc.acm.org/2026/

Chicago river ready for swimming after an century nytimes.com/card/2025/09/2…

@LivermoreTech @TaraBull808 The wake vortices from one plane creates turbulence for the other. So landing gets technically more complicated for pilots. See NASA's ALAS protocol.

@TaraBull808 I don't know the details but why would this be impossible anyway? In theory you could have 15 runways in parallel.

I'm just going to leave this here

This is a first step toward a theory of indistinguishability for real-world autonomy. It helps us ask: what's unknowable for robots? mitras.ece.illinois.edu/research/2025/…

We show that indistinguishability can be: Automatically checked under mild conditions Viewed as a bisimulation relation Approximated with an iterative algorithm Analyzed via an observability-inspired method that converges in finitely many steps #Robotics #Observability

Einstein’s “most beautiful thought” was about indistinguishability: a falling person can’t perceive their own weight. In CS, this idea underpins impossibility results in distributed systems. What does indistinguishability mean for autonomous robots with real-world sensors?