huawei-noah/trustworthyAI

What it solves

This repository serves as a hub for trustworthy AI research and tools from Huawei Noah's Ark Lab, focusing primarily on causal learning and time series foundation models.

How it works

The project is organized into several specialized components:

• gCastle: A toolchain for causal structure learning that uses gradient-based algorithms to discover causal relationships.
• TabbyTSFM: A time series foundation model designed for forecasting, classification, and anomaly detection, including full pretraining code and evaluation benchmarks.
• Research & Datasets: A collection of causality-related research implementations (such as CausalVAE and GAE) and both real-world and synthetic datasets.

Who it’s for

Researchers and developers working on causal discovery, time series analysis, and the development of trustworthy AI systems.

Highlights

• Gradient-based Causal Learning: Provides a dedicated pipeline for causal structure learning and evaluation.
• Time Series Foundation Model: Includes the open-sourced pretraining code and weights for TabbyTSFM.
• Comprehensive Benchmarking: Includes harnesses for forecasting (GIFT-Eval, TIME), classification (UCR), and anomaly detection (TSB-AD).