Tim Vieira

Our awesome table at the @black_in_ai dinner! @ojakintande @MathYouF @trojrobert @yuehgoh @irenekameni08pic.twitter.com/mBIbeqa0iI

Great suggestion. Interestingly, it doesn't help much (2-3% faster). Δ-updates have the same # FLOPS w/ ½ as many memops (no sibling lookups). However, in heap structure, siblings are adjacent in memory so the lookup is very cheap.pic.twitter.com/R2vKnFEEFp

Machine Learning Drake says, "Don't be a square, write squared error like a pro!"pic.twitter.com/GNLILN3ogT

Flattered, but I am lazy to my core. https://imagenet-roulette.paglen.com/ pic.twitter.com/5yNHXUvwSQ

I have released a fast implementation of heap-based sampling. Implementation in Cython Super-fast sampling from an evolving set of weights >8x faster than numpy's sampling without replacement method About the method http://timvieira.github.io/blog/post/2016/11/21/heaps-for-incremental-computation/ … Code https://github.com/timvieira/arsenal/blob/master/arsenal/maths/sumheap.pyx …pic.twitter.com/ayt6UrNzbi

I've released a Jupyter notebook w/ "test cases" illustrating the math behind one of my favorite articles √-𝗯𝗶𝗮𝘀𝗲𝗱 𝘀𝗮𝗺𝗽𝗹𝗶𝗻𝗴 Most peoples intuitions lead them to the wrong answer on this one! http://timvieira.github.io/blog/post/2016/06/28/sqrt-biased-sampling/ …pic.twitter.com/7UCZPcSiiy

New post! Ever wonder why restarting stuff, possibly multiple times, seems to fix problems? The restart acceleration trick A cure for the heavy tail of wasted time Applications to speeding up randomized algorithms Applied survival analysis https://timvieira.github.io/blog/post/2019/09/06/the-restart-acceleration-trick-a-cure-for-the-heavy-tail-of-wasted-time/ …pic.twitter.com/HxOhU2jTeq

You need a new blood boy!pic.twitter.com/uBDxJ3K5ud

Release #3 from my implementation trove: 𝗥𝗲𝗶𝗻𝗳𝗼𝗿𝗰𝗲𝗺𝗲𝗻𝘁 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗯𝗮𝘀𝗶𝗰𝘀 https://github.com/timvieira/rl  Plenty more to come in this little library so stay tuned.pic.twitter.com/yu2awmufAV

Release #2 from my trove to careful implementations A reference implementation of algorithms for distributions over spanning trees https://github.com/timvieira/spanning_tree … - the matrix-tree theorem in action - tested against enumeration methods - a new log-det-exp trick for numerical stabilitypic.twitter.com/JdppxP8U6Q

Today, from my trove of carefully worked out implementations of machine-learning concepts, I liberate hypergraphs: A reference implementation of basic & advanced hypergraph algorithms, emphasizing clarity & generality, over efficiency. https://github.com/timvieira/hypergraphs … Enjoy!pic.twitter.com/N4doyqJf8H

Amazing (free!) book by a fantastic teacher! I've been referencing jeffe's notes since I was an undergraduate. http://jeffe.cs.illinois.edu/teaching/algorithms/ …pic.twitter.com/OloYelpiE7

Now that Python 3 makes it easy to have non-ASCII character as identifiers, what do folks think of using Greek letter in code so it looks more like the math?pic.twitter.com/6jk2Fz9fgS

The exponential jumps algorithm: Faster sampling by waiting Reduces the num of random variates from O(n) to O(log n) Very clever: sample waiting times between mins Closely related to the Gumbel-max trick & sampling w/o replacement http://timvieira.github.io/blog/post/2019/06/11/faster-reservoir-sampling-by-waiting/ …pic.twitter.com/CiYpG6RpVR

I like to file away "solutions peer review" alongside "proofs that P≠NP" and "proofs of the Riemann hypothesis."pic.twitter.com/syvpN5BnBb

I love that Randløv & Alstrøm found that it was easier for the reinforcement learner to learn to ride (sorry "drive") a bicycle with no hands! pic.twitter.com/RUkGpz7bHI

Answer! Figure 4 of Randløv & Alstrøm (1998) "Learning to Drive a Bicycle using Reinforcement Learning and Shaping" http://robotics.usc.edu/~aatrash/COMP526/docs/randlv98learning.pdf …pic.twitter.com/8nSOzDflbn

Pop quiz Does anyone recognize the cover art from Sutton & Barto 2.0? http://incompleteideas.net/book/RLbook2018.pdf … Threadpic.twitter.com/9Cl0J84FFx

Not one second later: Google thinks I need help finding a burger joint... Now, I want a burger.pic.twitter.com/S1MVtv7L1U

Berkson's Paradox https://en.wikipedia.org/wiki/Berkson%27s_paradox … The burger example got me pic.twitter.com/U2xAdwtKRS