CausalNLP

🚀 Flexibility: Researchers can tune the simulator to test for: ✅Unmeasured Confounding ✅ Selection Bias ✅ Missing Data (MCAR, MAR, MNAR) A vital sandbox for moving Causal AI from "toy problems" to real-world medical utility. 📊

🧪 Testing the SOTA: The authors show that even top algorithms (PC, FCI, GES) struggle with this level of clinical logic, highlighting the need for more robust discovery methods.

Sharing "Neuropathic Pain Diagnosis Simulator for Causal Discovery Algorithm Evaluation", one of the most popular datasets used for evaluating causal discovery algorithms! 🩺🧠 @RuiboTu @kunkzhang 📄 arxiv.org/abs/1906.01732 💻github.com/TURuibo/Neurop… 1/n

🚀 Why it matters: Essential for understanding training dynamics, preventing copyright infringement, and enhancing data protection. 🛡️ By observing a small set of instances, we can now map a model's full "Memorisation Profile." Experiments conducted on the Pythia model suite. 🧪

⚙️ Training Dynamics: Learning Rate (LR): Memorisation is heavily driven by the LR schedule. Instances seen during peak LR are memorised much more intensely. 📉 Order: Data order matters, but its effect is largely mediated by the LR and the specific model’s capacity. 🔀

Sharing ACL 2024 Best Paper Winner, "Causal Estimation of Memorisation Profiles"! LMs can reproduce training data verbatim, but measuring this "causally" (what would happen if the model never saw the data?) is hard. This paper fills the gap. link: aclanthology.org/2024.acl-long.… 1/n

2️⃣ Causal AI Scientist: an autonomous agent that performs end-to-end causal inference from a dataset, its description, and a causal query. 🔗 drive.google.com/file/d/1KvxAVI…

1️⃣ CauSciBench: a benchmark assessing whether LLM agents 🤖 can answer scientific causal questions by navigating an end-to-end causal inference pipeline. 🔗 drive.google.com/file/d/17isFKv…

Introducing the latest research highlights from @JinesisLab at the Dagstuhl seminar on Causal LLMs! 1️⃣ CauSciBench: Evaluating LLM Causal Inference for Scientific Research 2️⃣ Causal AI Scientist: Facilitating Causal Data Science with LLMs 🔗 cr-llm.github.io 1/n

We are hosting a Dagstuhl seminar on Causality & LLMs this week (Apr 7–10). Bringing together world experts to explore: 1️⃣ Integrating LLMs 🤖 into causal workflows 2️⃣ Evaluating & improving LLMs’ causal reasoning 🧠 Co-organized w/ @amt_shrma @DominikJanzing @kunkzhang @ZhijingJin 📍Schloss Dagstuhl, Wadern, Germany 🔗 dagstuhl.de/26152 📖 cr-llm.github.io 📅 Apr 7–10 #CausalNLP #LLM #Dagstuhl @CausalNLP @MPI_IS @ELLISforEurope @UofTCompSci @VectorInst @TorontoSRI @CIFAR_News @JinesisLab @EuroSafeAI @ELLISInst_Tue Also joined with my student @rahulbshrestha to present our CauSciBench and Causal AI Scientist work :)!

Our CausalNLP Tutorial🎙️video and 🎞️slides are out! #EMNLP2022 @emnlpmeeting Curious about (1) What Is Causality? and (2) How Causality Can Help #NLProc? Check out our #EMNLP2022 Tutorial with @ZhijingJin @amir_feder @kunkzhang 📺Video: youtu.be/4bq1ZYxXbtg

Curious about "What is Causality & How can it help NLP"? Vote for your top interest areas to hear more about it during our #emnlp2022 Tutorial!

A nice python-based causal discovery package by Prof Kun Zhang and his CMU team 👍