Skip to content
프로젝트

huawei-noah/trustworthyAI has 1,154 GitHub stars and 255 forks. GitHot measured 0 stars in 24 hours, 0 in 7 days, and +1 in 15 days. Last checked 2026-11-19 10:48 UTC. Last push 2026-09-21.

이 수치를 측정하는 방법

측정된 스타 증가

in 24h
0
in 7 days
0
in 15 days
+1
in 1 month
업데이트 지연

서로 겹치는 기간의 순증가량이며 일별 합계가 아닙니다. 측정값이 없으면 비워 둡니다. 스타는 관심도를 나타내며 품질을 보장하지 않습니다.

소개

Trustworthy AI related projects

머신러닝

README.md

Original

Trustworthy AI

This repository is a collection of trustworthy AI related works from Huawei Noah's Ark Lab.


gCastle

  • A causal structure learning toolchain containing various functionalities related to causal learning and evaluation. A tech report describing the toolbox is available here.
  • The package offers a number of causal discovery algorithms, most of which are gradient-based, hence the name: gradient-based Causal structure learning pipeline.

Competition

  • Information and baselines for causality-related competitions arranged by Noah's Ark Lab.
  • Previous competitions were held at PCIC 2021, PCIC 2022, and NeurIPS 2023.

Datasets

  • Real-world datasets released by Huawei Noah's Ark Lab.
  • Code for generating various synthetic datasets.

Research

  • Research works related to causality. We will continuously add new methods here.
  • Currently contains implementations of CausalVAE, GAE, and causal discovery with reinforcement learning.

TabbyTSFM

  • Tabby, a time series foundation model proposed by the Paris team of Huawei Noah's Ark Lab, supporting forecasting, classification, and anomaly detection. Its pretraining code is fully open-sourced here.
  • Data processing: TabbyTSFM/src/tabby/data holds the GIFT-Eval/BLAST shard reader, the KernelSynth Arrow reader, and the online CauKer V2 generator that samples series from structural causal models; shared input preprocessing is in TabbyTSFM/src/tabby/utils.
  • Pretraining: TabbyTSFM/recipes/pretrain contains the distributed mixed-data trainer and the train_165k.sh release entry point.
  • Tests and evaluation: TabbyTSFM/tests covers the release contract, checkpoint loading across every accepted layout, and the GIFT-Eval script end to end; TabbyTSFM/benchmarks holds the benchmark harnesses for GIFT-Eval and TIME (forecasting), UCR (classification), and TSB-AD (anomaly detection).
  • The weights are published on Hugging Face at paris-noah/Tabby