
Computer Scientist. Professor @Caltech. Views are my own.
Pasadena, CA
Joined September 2020
- Tweets130
- Following545
- Followers1.3K
- Likes230
Adam Wierman retweeted
Excited to share that "Online Policy Optimization in Unknown Nonlinear Systems" is accepted to #COLT2024!
Paper: arxiv.org/abs/2404.13009
This work introduces a new bridge between online learning and adaptive control. An important finding in adaptive control is that one often doesn't need to converge to the correct model in order to perform well. In contrast, online learning for control generally requires knowing or converging to the correct model.
This work develops a meta-framework that can bridge this divide. The resulting algorithms enjoy regret guarantees that do not depend on converging to the correct model, and are also very practical to implement (we have follow-up work that uses the approach for real-time agile robotic control). We hope it's a significant step towards unifying the foundations of adaptive control and online learning.
This work was led by Yiheng Lin (yihenglin97.github.io/) and James Preiss (jamespreiss.com/), with also contributions from Fengze Xie, Emile Anand, @SoonJoChung, myself, and @AdamWierman.
The COLT 2024 proceedings are out. Check out some great papers! proceedings.mlr.press/v247/
Excited to be a part of this team!
I am thrilled to announce Carbon Connect, a $12M NSF Expedition in Computing. By rethinking computing infrastructure, from semiconductors to datacenters, will will lay new foundations for environmentally sustainable systems for artificial intelligence. blog.seas.upenn.edu/penn-eng…
Adam Wierman retweeted
🧵 Thrilled to announce the #ICML RL workshop 'Aligning RL Experimentalists and Theorists'! We will have several talks and a panel delivered by a super lineup of speakers: @white_martha, @ShamKakade6, @yayitsamyzhang, Dylan Foster, Niao He, @svlevine, and @MengdiWang10.
1/3
Adam Wierman retweeted
Graduate and undergraduate students (even w/o a paper at the conference) wishing to attend ACM SIGMETRICS / IFIP Performance 2024 (June 10-14, 2024 in Venice, Italy) can apply for a student grant until April 29, 2024 at: sigmetrics.org/sigmetrics202… #sigmetrics #performance #students
Adam Wierman retweeted
Registration is open for ACM SIGMETRICS / IFIP Performance 2024 (June 10-14, 2024 in Venice, Italy) at sigmetrics.org/sigmetrics202…
The early registration deadline is May 15. #sigmetrics #performance #analysis #measurement
Adam Wierman retweeted
One teaser @umutsimsekli gave, see also the Neurips talk of @AdamWierman in link below, is the following: suppose A,B are normally distributed and you want to maximize E[Ax^2+Bx] with stochastic gradient descent, unit batch size 1/4
After lunch, Umut Simsekli di.ens.fr/umut.simsekli/
gave 2 great talks on the board, bringing mathematicians up to speed on ML concepts.
Actually, Umut co-organized a Neurips workshop on heavy tails in '23, talks are on
neurips.cc/virtual/2023/work…
Our meeting continues tomorrow
So excited to finally be able to share a project I have been helping to launch over the past few years... It's a fantastic team tackling an ambitious and important goal techcrunch.com/2024/03/11/si…
Adam Wierman retweeted
Replying to @Caltech
@Caltech invites faculty applications in all areas related to Information, Systems and Computing (including AI and AI-adjacent fields).
applications.caltech.edu/job…
Deadline for full consideration is December 15, 2023
Adam Wierman retweeted
Data centers should schedule their energy use, looking beyond batch jobs and bitcoin mining, which offer flexibility but not at scales needed to align large datacenter demand with intermittent carbon-free supply.
podcasts.apple.com/us/podcas…
Adam Wierman retweeted
[SNAPP Seminar] SNAPP Seminar Series Fall 2023 schedule:
Isaac Grosof, October 9;
Emina Soljanin, October 23;
Lei Ying, November 6;
Shane Henderson, November 20;
Hongseok Namkoong, December 4;
Lightning Talks, December 11;
For more details, please visit: sites.google.com/view/snapps…
Adam Wierman retweeted
The ACM SIGMETRICS / IFIP Performance 2024 conference will be held June 10-14, 2024 in Venice, Italy. Fall submission deadlines are Oct 4 (abstract) and Oct 11 (paper): bit.ly/3NP9k7N
#sigmetrics #performance #analysis #measurement
Congrats @SunghoShinSS
This Post is from an account that no longer exists. Learn more
Adam Wierman retweeted
First AIM workshop in our new location, and the spectacular views seem to be working some kind of magic. We put out the happy hour wine at 5pm as usual, but closing in on 6pm this group was still going...
Adam Wierman retweeted
You’ve heard of AI’s algorithmic inequity. But what about its environmental inequity?
@renshaolei explains how AI’s environmental costs are not evenly distributed and can be disproportionately higher in certain regions than in others. themarkup.org/hello-world/20…
Really excited about this new line of work with Pengfei Li, Jianyi Yang, and @renshaolei
This quoted post is unavailable.
Adam Wierman retweeted
ACM SIGMETRICS 2023: the conference starts on June 19 in Orlando, please check out the exciting program this year: sigmetrics.org/sigmetrics202…
Adam Wierman retweeted
Call for Nominations - ACM SIGMETRICS Doctoral Dissertation Award. Submission deadline: 9/15/2023. All nomination materials must be submitted electronically by the supervisor to the 2023 committee chair, Urtzi Ayesta. For more details, please check sigmetrics.org/awards.shtml#…
Program is now finalized for our workshop on Learning Augmented Algorithms at Sigmetrics/FCRC this year. Its should be a really interesting morning! learning-augmented-algorithm…
Adam Wierman retweeted
🚨New episode out🚨 Join us on a new control-theoretic adventure with John Doyle (@Caltech) Hear about his exploits as a world-class athlete🏋🏽his adventures in the Panamanian jungle🌴and his heroic battles against the toughest control problems in the golden age of robustness!🚀
Adam Wierman retweeted
RL Theory Seminars are back! First talk,
Policy learning "without'' overlap: Pessimism and generalized empirical Bernstein's inequality by Ying Jin! sites.google.com/view/rltheo…


















