We're the Yue Lab at @Caltech with PI @yisongyue! We work on #MachineLearning for science and engineering applications 🤖 🧠 💻 Tweets by students & postdocs!
Caltech
Joined October 2020
- Tweets272
- Following100
- Followers1.8K
- Likes420
KSI pushes the Pareto frontier of self-improvement agents 🚀
A simple thought experiment: What if we stop trying to cram intelligence into agents, and instead externalize it into a shared knowledge base?
Introducing Knowledge-centric Self-Improvement (KSI).
TL;DR: We show that a remarkably simple knowledge curation mechanism pushes the Pareto frontier of self-improvement agents, outperforming SOTA agent-centric approaches (e.g., harness or prompt optimization) across reasoning, coding, and terminal benchmarks at a fraction of the cost.👇
Yue Lab Caltech retweeted
A simple thought experiment: What if we stop trying to cram intelligence into agents, and instead externalize it into a shared knowledge base?
Introducing Knowledge-centric Self-Improvement (KSI).
TL;DR: We show that a remarkably simple knowledge curation mechanism pushes the Pareto frontier of self-improvement agents, outperforming SOTA agent-centric approaches (e.g., harness or prompt optimization) across reasoning, coding, and terminal benchmarks at a fraction of the cost.👇
Yue Lab Caltech retweeted
Tokenization (aka the root of suffering, iykyk) has gotten a terrible rap this past week😅 but in nascent fields this rap is wack🌶️
I am ecstatic to share➡️TOTEM: TOkenized Time Series EMbeddings for General Time Series Analysis
TOTEM learns discrete time tokens (not patches!!)
Yue Lab Caltech retweeted
Excited to share our work on symbolic music generation: arxiv.org/abs/2402.14285!
We introduce a symbolic music generator with non-differentiable rule guided diffusion models, enabling musicians to effectively use it as a compositional tool.
Website: scg-rule-guided-music.github…. 🧵👇
Yue Lab Caltech retweeted
Congratulations @ganlumomo!!!
I’ll be joining @GeorgiaTech as an Assistant Professor in @GTaerospace in January 2024!! Deep thanks to my PhD advisors Ryan Eustice, Jessy Grizzle and @GhaffariMaani, my postdoc advisors Soon-Jo Chung and @yisongyue, and all my group members at @UMRobotics and @Caltech.
Congrats to our postdoc @ganlumomo 🥳🥳🥳
I’ll be joining @GeorgiaTech as an Assistant Professor in @GTaerospace in January 2024!! Deep thanks to my PhD advisors Ryan Eustice, Jessy Grizzle and @GhaffariMaani, my postdoc advisors Soon-Jo Chung and @yisongyue, and all my group members at @UMRobotics and @Caltech.
Our favorite photo from ICLR 2023!
It's going to be a YueCrew kind of ICLR 2024 😍🧑💻🇦🇹
📸 PC @yeh_im_excited
Today's lab meeting reminder 😂
Coming to you live from Pasadena, it's Friday Night Live with your hosts @SaberaTalukder & @vdorbs. In tonight's top story @yeh_im_excited returns from Rwanda with the latest ICLR tidbits 👀 What ML news hides in plain sight? Tune in to find out!
Yue Lab Caltech retweeted
All animals behave in 3D - we discover 3D poses directly from multi-view videos without requiring annotations.
Essentially videos -> 3D keypoints + connections
We will be @CVPR on June 21!
BKinD-3D Paper: arxiv.org/abs/2212.07401
Co-first-authors Lili Karashchuk & @_AmilDravid
Excited for this!! 🤓
I am honored to serve as Senior PC for #ICLR2024. Looking forward to serving with @_beenkim (General Chair) and a fantastic PC cast (@YizhouSun, @swarat, @EmtiyazKhan, & Katerina Fragkiadaki). Hope to see you all in Vienna!
Yue Lab Caltech retweeted
Really exciting work by @jxbz, @chris_mingard & others on designing hyperparameter-free optimizers for deep neural networks!
We are thrilled to announce "automatic gradient descent"---a neural network optimiser without hyperparameters. AGD trains out-of-the-box and at ImageNet scale.
paper: arxiv.org/abs/2304.05187
PyTorch: github.com/jxbz/agd
1/5
Yue Lab Caltech retweeted
Conformal Generative Modeling on Triangulated Surfaces
A general framework that can reliably train probability distributions on a wide range of complicated manifolds.
1st Author: @vdorbs
Project: vdorbs.github.io/conformal-g…
Yue Lab Caltech retweeted
We are thrilled to announce "automatic gradient descent"---a neural network optimiser without hyperparameters. AGD trains out-of-the-box and at ImageNet scale.
paper: arxiv.org/abs/2304.05187
PyTorch: github.com/jxbz/agd
1/5
Yue Lab Caltech retweeted
phd time @berkeley_ai @jlistgarten @svlevine !! also time to start using twitter…?
super excited for the next several years and genuinely so much appreciation to @yisongyue @ryan_p_adams @YueLabCaltech @denizzokt @klbouman @countability + many others who have built me up :))
Yue Lab Caltech retweeted
GUYS the @CosyneMeeting workshop program just dropped 🥳
AND @EricMTrautmann's + my workshop "Taming Complexity: Discovering Interpretable Latent Spaces in Human Brains & Behaviors" will be on your Mont Tremblant menu🧑🍳
Too early to say #Cosyne2023? 🙊🤫
cosyne.org/workshops-program…
Yue Lab Caltech retweeted
NeurIPS officially ended yesterday 😭 and it was 🔥 both in person and virtually! I enjoyed seeing old faces and meeting so many new ones ♥️
I put together some topics + papers that piqued my interest and maybe they'll pique yours too🤗
saberatalukder.com/neurips_2…
Go see @francescazfl's poster tomorrow at #NeurIPS!!
If you are at #NeurIPS2022 and interested in ML for bio/proteins, check out our poster “Pretrained protein language model transfer learning: is the final layer representation what we want?” @workshopmlsb tomorrow (Room 288 - 289) @avapamini @KevinKaichuang @alexijielu
Yue Crew #Neurips Dinner Underway 🤗
So many embarrassing @yisongyue stories coming to light 😂🤫
