@RuiboTu

Ph.D. @KTHuniversity. Work on causality and machine learning.

 Stockholm, Sweden
turuibo.github.io/
Joined October 2016
  • Tweets46
  • Following324
  • Followers94
4Media
5 years! You made it! An excellent research story!
yingzhen@liyzhen21 May 2025
This paper took @BalsellsRodas ~5 years in making: 2020/21: MSc proj, toy exp🐣 2022: added more exp, rejected due to weak theory 😥 2023/24: invented a new proof tech in another proj 🤔 2024/25: revisit, apply new proof tech, resubmit -> accepted 🥳 Persistence pays off indeed👍
1
58
TU Ruibo retweeted
yingzhen@liyzhen21 May 2025
This paper took @BalsellsRodas ~5 years in making: 2020/21: MSc proj, toy exp🐣 2022: added more exp, rejected due to weak theory 😥 2023/24: invented a new proof tech in another proj 🤔 2024/25: revisit, apply new proof tech, resubmit -> accepted 🥳 Persistence pays off indeed👍
Carles Balsells Rodas@BalsellsRodas1 May 2025
Excited to share that our paper "Causal discovery from Conditionally Stationary Time Series" has been accepted to ICML 2025!🥳 Pre-print: arxiv.org/abs/2110.06257 Thank you very much to all my collaborators, persistence pays off! #icml #icml2025

Causal Discovery from Conditionally Stationary Time Series

Causal discovery, i.e., inferring underlying causal relationships from observational data, is highly challenging for AI systems. In a time series modeling context, traditional causal discovery...

arxiv.org
1
2
1
43
3,301
I am pleased to announce that I have updated the online versions of my 2 textbooks (see probml.github.io/pml-book/): I fixed all issues listed on github, added some new references (esp on LLMs), and made a few other small tweaks.

“Probabilistic machine learning”: a book series by Kevin Murphy

“Probabilistic Machine Learning” - a book series by Kevin Murphy

probml.github.io
15
167
8
1,255
68,393
Are you curious about the performance of #ChatGPT on #CausalDiscovery? We do. So we push its limit and explore its ability to answer causal discovery questions by using a causal medical benchmark. The full report serves as an appendix to the benchmark 👇 shorturl.at/bGH45
1
2
10
874
more replies
2. We need to be extremely cautious about using causal claims made by ChatGPT as causal discovery results. This is because causal discovery and causal question answering with large language models are fundamentally different tasks.
1
121
3. ChatGPT can give some correct non-trivial answers serving as a good complementary for causal discovery methods. This might open up new research opportunities on utilizing the large language models to complement, improve and develop better causal machine learning tools.
107
🧐Occam's razor for causality: a causal model can explain the data in a simpler way.
Fermat's Library
@fermatslibrary29 Jun 2022
Solar System a perspective from the Earth. These diagrams show the paths traced by Mercury, Venus, Mars, Jupiter and Saturn as seen from Earth.
4
Augmenting one more dimension => Embedding much more information than before
Jia-Bin Huang
@jbhuang060419 Jun 2022
Cool! With computational approaches, everyone can be a multiview wire art artist! You can even go for three view! See more examples here: cgv.cs.nthu.edu.tw/MVWA_onli…
2
The brilliant team behind the Oceanbird is reducing carbon emissions in the sailing industry by up to 90% using clean wind power technology.

112
43
832
Announced today, @awscloud and @MSFTResearch partnered to create a new @GitHub organization called PyWhy—where novel causal-analysis algorithms developed by Amazon will augment DoWhy, Microsoft's existing causality library. #AmazonScience #AWS

5
31
6
122
OMG, surely embarrassingly they do 😂 In other words, even the most likely corresponding data point of a given noise has a small probability according to the analytical solution. Maybe the magic is from the engineering power of DNNs in the backward process again.
François Fleuret
@francoisfleuret30 Apr 2022
Diffusion models somehow fight entropy.
2
📃 New paper "Optimal transport for causal discovery" by @RuiboTu, Kun Zhang, Hedvig Kjellström & Cheng Zhang has been accepted at #ICRL2022! 👏 Full paper available: arxiv.org/abs/2201.09366
1
1
8
We've released differentiable convex optimization layers for JAX! Code: github.com/cvxgrp/cvxpylayer… Tutorial: nbviewer.jupyter.org/github/… NeurIPS paper: arxiv.org/abs/1910.12430 Blog post: locuslab.github.io/2019-10-2… With @akshaykagrawal, @ShaneBarratt, S. Boyd, S. Diamond, and @zicokolter
5
123
5
552
(1/5) In our machine learning reading group at KTH, we had a very insightful discussion. It is about transfer learning, that is the generalisation without human involved specification what we want ? In other words, is the human involved specification what we don’t want?
1
1
more replies
(4/5) A: For sure, the method in XXX paper captures very important intuitions making its performance excellent, which is great for the field. But the generalisation without specifying how it generalises and how general it can be could be vague and misleading as well.
1
(5/5) A: Not say assumptions doesn’t mean general. There are implicit assumptions of intuitions. If only rely on intuitions without further specifying and understanding underlying assumptions, this would be the most we would achieve, which would be a pity for it and the field.
#Causal #RepresentationLearning. (1/3) How do we understand causal representation learning ? Why it is so hard to imagine what is it? Maybe the reason is sitting in the philosophy of science.
1
1
more replies
(3/3)Thus, if causal representation learning is aiming at something like theoretical identification, maybe causality is not good at it from a methodology perspective.
1
Later on, I realised that there are thousands of ways to model the same thing. We do not actually aim to only model the ground-truth data generation process but a human intelligence level model, which can be understood with human/causal reasoning.
Load more