Missed my flight. Missed later flight I was accepted standby for. Stayed in terminal overnight. Got a flight the next morning. Was accepted standby on 1st flight of morning ("fortunately", security line was still long enough that many people missed the flight, so got accepted from standby). Not fun.
That beat my 7 hours 20 minutes at ATL coming back from INFORMS Optimization Conference Sunday night.
Stanford Emeritus Professor of Operations Research, Frederick "Fred" S. Hillier, died last week at the age of 89.
See www.linkedin.com/feed/update/...
ππππ : ππ«ππ’ππ’ππ’ππ₯ πππ¨ππ‘ππ¬ππ’π ππππ¬π¨π§π’π§π ππ§πππ₯π₯π’π ππ§ππ
ASRI is absent from all frontier AI chatbots and doesnβt seem to be on the near-term horizon
ASRI would correctly perform probability calculations in coherent and consistent manner, using Law of Total Probability
www.linkedin.com/feed/update/...
The Operations Research version of "There's no free lunch" , as articulated by "The Mythical Man Month author", Fred Brooks:
"You can only get something for nothing if you have previously gotten nothing for something."
Links to lnkd.in/eAz8n2bA The 3 secrets to AI chatbot prompt generation are constrain, constrain, constrain.
Is it time to recognize stochastic thinking as the new frontier in AI chatbot inference and prompt engineering? I think so. What do you think? www.linkedin.com/posts/markls...
Bonus points for the cool article title.
Doing Operations Research (OR):
Figure out how something does or could work, and make it work better.
My thoughts on using an Operations Research approach to formulate prompts for AI chatbots capable of performing computationally intensive inference.
Hint: Inference should be based on conditional probability, so optimization should be constrained, not unconstrained
www.linkedin.com/feed/update/...
For me, GenAI's like Wiki on steroids. Drill deep & broad, go w/flow (2nd order conditions for nonlinear SDP when stationary point not exact) - interactive learning w/ renaissance teacher who's not always right. Critically think about what GenAI says; call it out when wrong, which it usually admits.
As MIT undergrad, did UROP on Dial-A-Ride in Center for Transportation Studies. Hani sat 5 feet from me. His 1st OR course used H&L. Told me was hardest book/course he ever had. That's how I first heard of OR. I skipped H&L; undergrad Math right to OR PhD courses at MIT (as undegrad), then Stanford.
det(exp(X)) = exp(tr(X)). Remarkably, the RHS does not involve off-diagonal elements of X.
Often overlooked: the price of some paths may be highly uncertain.
Does this technique allow construction and SC proof of a barrier for Symmetrized Quantum Relative Entropy (SQRE), defined as SQRE(A,B) = QRE(A,B)+QRE(B,A) ? This would hopefully allow use of a single symmetric cone instead of 2 asymmetirc cones?
Very nice. It would be interesting to see how many "unsolved" (Erling's challenge) problems on ask.cvxr.com could now be conic reformulated, and to what extent computer algebra type techniques might help in finding such refomulations, along the lines of ask.cvxr.com/t/ph-d-thesi... .
I note in passing that I read a new paper today which had 2 instances of "we remark in passing". That's a writing style I associated with decades past, such as a paper I coauthored in "Operations Research" in 1983 which I note in passing used "we note in passing" (attributable to my older coauthor)
Now fixed (at least for me).
Yes, water will freeze at 27 degrees Fahrenheit because the freezing point of water is 32 degrees Fahrenheit, so any temperature below that will cause water to freeze; 27 degrees is below 32 degrees.
Key point: Water freezes at 32 degrees Fahrenheit.
Can single precision cut the mustard on first order methods, such as PDHG for LP (perhaps if followed, if needed, by some type of repair step or crossover in double precision)?
Sorry. I meant single vs. double precision.
What precision is used on GPU?
Corollary: If if A and B are symmetric d x d matrices and B is positive definite, then AB is diagonalizable.
How does energy (electricity) used to solve the problem compare between GPU and CPU?
SR1 is a fairly well-known Quasi-Newton update. Most frequently used in conjunction with Trust Regions on indefinite Lagrangians.
SR! Quasi-Newton update is same as SR1, except for holding on to the shift key too long after the R. Most frequently used by older people with declining visual acuity.
A crappy simulation is a crappy simulation, whether or not it's called a Digital Twin.
