@leeley18

李力 Staff ML engineer@Meta ex-ML engineer@YouTube ex-Researcher@GoogleAI Aikido 3rd dan blackbelt Retired Kaggle competition master

 Mountain View, CA
Joined August 2020
  • Tweets148
  • Following206
  • Followers605
14Media
My last work at @GoogleAI "Evolving symbolic density functionals" just appeared @ScienceAdvances !Inspired by recent advances in AutoML and program synthesis, we show that ML can compose symbolic density functionals with simplicity as human-designed ones. science.org/doi/10.1126/scia…

Evolving symbolic density functionals

Machine learning composes density functionals in symbolic forms with similar simplicity as human-designed ones.

science.org
30
3
129
Solving Schrödinger Equation Using Tensor Neural Network arxiv.org/abs/2209.12572 #compchem

Solving Schrödinger Equation Using Tensor Neural Network

In this paper, we introduce a novel approach to solve the many-body Schrodinger equation by the tensor neural network. Based on the tensor product structure, we can do the direct numerical...

arxiv.org
3
19
Very interesting! A bit like Autopilot but for your computer.
This quoted post is unavailable.
Try viewing anyway · View on X
24
116
11
1,165
Learning with Differentiable Algorithms If you’re interested in making algorithms differentiable so that algorithms can be learned from training data, @FHKPetersen’s thesis covers differentiable versions of Sorting and Ranking, Top-k, Rendering, Logic. arxiv.org/abs/2209.00616
2
50
257
Li Li retweeted
Xavi@xamat4 Sep 2022
My friend @jure claims that modern #recsys at most companies are "based on graphical representations" (youtu.be/aBHC6xzx9YI min.8). In my experience that is not the case. Graphs might be *used*, but are never the fundamental construct. Anyone has similar/different experience?
5
5
50
Fei-Fei to me after I showed her my first image captioning (image to text) network around 2015: “very cool, now do it backwards!”. Me: “haha that’s impossible” 🥲. Turns out you just need a few ~B alt-text dataset scrape, transformer, diffusion, and a cluster of ~thousand A100s.
27
161
12
2,657
Yesterday @nnaisense released EvoTorch (evotorch.ai), a state-of-the-art evolutionary algorithm library built on @PyTorch, with GPU-acceleration and easy training on huge compute clusters using @raydistributed. (1/2)

EvoTorch

Next-generation Evolutionary Search, Learning and Planning

evotorch.ai
8
195
6
994
A nice article from Dr. Luciano Abriata featuring our recent work on symbolic regression and density functional theory done at @GoogleAI towardsdatascience.com/googl… #dft #physics #ai4science #ml Link to our paper "Evolving symbolic density functionals" arxiv.org/abs/2203.02540

Google proposes new method to derive analytical expressions for terms in quantum mechanics...

The new method derives accurate functionals (elements of quantum mechanical calculations) in symbolic form, hence human-interpretable...

towardsdatascience.com
1
36
3
195
There's a special place in hell for people who code like this: if condition == True: return True else: return False Don't be one of those people!
33
30
6
276
levels.fyi is your friend
misha smirnov@SaladZombie24 Mar 2022
Talking to a recruiter about a senior data scientist job. I gave her my salary expectation ($150k) and she laughed and told me she’s never seen an offer lower than the 200s. I likely won’t get this job but want to share since no one tells academics these numbers.
1
Julia has been the future of machine learning for 10 years and will stay as the future of machine learning for the next 10 years.
50
68
35
1,142
Can computer construct density functionals in the symbolic form? I would like to share with you our latest work on machine learning applying to density functional theory -- Evolving symbolic density functionals (arxiv.org/abs/2203.02540) #SymbolicRegression #ProgramSynthesis 1/3
3
4
2
27
Inspired by recent advances in AutoML and program synthesis, we proposed a new framework, Symbolic Functional Evolutionary Search (SyFES), that can automatically produce XC functionals with similar simplicity as functionals designed by humans in the past few decades 2/3
1
2
-- symbolic forms with a manageable amount of parameters. This work is machine learning applied to developing human-readable scientific expressions and not just a blackbox prediction. I hope you find it interesting. 3/3
1
3
I'm excited to share that our Probability group at #Meta is open sourcing Bean Machine, a platform for fast, composable probabilistic modeling on PyTorch. Check it out to learn more at: research.facebook.com/blog/2… #beanmachine #ppl #bayesian #inference

Bean Machine: Composable, Fast Probabilistic Inference on PyTorch | Meta Research

Today, we’re excited to announce an early beta release of Bean Machine, a PyTorch-based probabilistic programming system that makes it easy to represent and to learn about uncertainties in the...

research.facebook.com
1
1
8
The Differentiable Programming Workshop at #NeurIPS21 calls for papers!
Differentiable Programming@diffprogramming5 Aug 2021
Do you have new exciting work on differentiable programming or using differentiable programming that you are eager to talk about? The Differentiable Programming Workshop at #NeurIPS21 call for papers is out! ⏰ Submission deadline: 10 September 2021 ⏰ diffprogramming.mit.edu/call…
3
11
Working with Steve is awesome and you should consider this opportunity if you work in ML and drug discovery.
Steven Kearnes@StevenKearnes2 Jun 2021
Join us at @Relay_Tx as we drive AI applications in drug discovery! We're building a world-class team---with Patrick Riley, Brian Goldman, and @wpwalters. Apply here: linkedin.com/jobs/view/25669…

Machine Learning Scientist at Relay Therapeutics — Cambridge, MA | LinkedIn Jobs

Apply for Machine Learning Scientist at Relay Therapeutics in Cambridge, MA. Full-time Not Applicable role. See responsibilities, qualifications, and similar jobs on LinkedIn.

linkedin.com
3
Extrapolating DFT Towards the Complete Basis Set Limit: Lessons from the PBE Family of Functionals. doi.org/10.26434/chemrxiv.14… #compchem
3
8
🔥JAX meets Transformers🔥 @GoogleAI's JAX/Flax library can now be used as Transformers' backbone ML library. JAX/Flax makes distributed training on TPU effortless and highly efficient! 👉 Google Colab: colab.research.google.com/gi… 👉 Runtime evaluation: github.com/huggingface/trans…
3
106
16
506
FermiFlow: Ab-initio study of fermions at finite temperature Code: github.com/buwantaiji/FermiF…… Paper: arxiv.org/abs/2105.08644 Animation shows the flow of electrons in a quantum dot towards the so called "Wigner molecule" structure.
2
19
2
87
Load more