Shu (@TheWanShu) header
Shu (@TheWanShu) profile picture

Causal machine learning MTS @abel_ai_lab | ASU DMML

US
Joined December 2011
Photo by Shu (@TheWanShu)Photo by Shu (@TheWanShu)Photo by Shu (@TheWanShu)
We built a frontier forecasting model from scratch in 15 days. Introducing Simons. From a minimal baseline, automated research reached a checkpoint ranked ahead of TimesFM-3 on the GIFT-Eval leaderboard in about two weeks. Full post: abel.ai/research/introducing…
1
2
7
159
1/ I wanted to share something I’ve been building. Andrey is a very fast causal discovery package. It's over 100x faster on PC and faster on most other supported methods, compared with other popular packages.
1
1
1
2
240
2/ For a long time, scale has held causal discovery back. This makes experimentation time-consuming and causal discovery on datasets with thousands of variables impractical. That's why Andrey is built, for faster, accurate and agentic-native causal discovery.
1
1
1
55
3/ Andrey is still in alpha. If you work on causal discovery, I’d love to hear which methods you need next. 📦 Try Andrey: uv add andrey-core 🌐 Launch Post: andrey.abel.ai/blog/introduc… 💻 Code: github.com/abel-ai-lab/andre…
27
Easy to see the flaw of WMs in terms of causality, hard to figure out a solution that can scale up to the frontier WMs.
1
4
1,092
In a changing world, time is the common coordinate system of all dynamics. Understanding the world is to read it through time, infer, predict, decide and repeat.
Words have LLMs, pixels have World models, what is the next? Time series. The world is ran by trillions of sensors, signals and movements underneath the surface. Why time series cannot have a world model? The community tried! TimesFM from Google, Chronos from Amazon, Moirai from Salesforce, but they are not the true TSWM we want, cuz unlike LLM and Vision WM, autoregression alone never work for TSWM, even though any variates or covariates are feed in as input, they are never bow-tied in data, representation, hidden space or even transformer architecture. We need to rethink about data curation, representation learning, causal ladders and architecture.
1
30
three causal PFN papers in 6 days is CRAZY ..., and all in neurips templates💀💀💀💀 arxiv.org/abs/2506.07918 arxiv.org/abs/2506.06039 arxiv.org/abs/2506.10914v1 #Causality #NeurIPS2025
143