Optimization-AI/LibAUC

LibAUC: A Deep Learning Library for X-Risk Optimization

What it solves

LibAUC addresses the difficulty of directly optimizing complex performance measures (X-Risk) in deep learning. Instead of relying on standard proxy losses, it provides tools to optimize metrics like AUROC, AUPRC, and NDCG, which are often more critical for specific real-world tasks but harder to optimize directly.

How it works

The library provides a unified framework and a user-friendly API to implement surrogate losses and specialized optimizers. It abstracts the optimization of compositional loss functions across several domains:

  • Classification of Imbalanced Data (CID): Surrogate losses for AUROC, AUPRC/AP, and partial AUROC.
  • Learning to Rank (LTR): Surrogate losses for NDCG, top-K NDCG, and listwise losses.
  • Contrastive Learning of Representation (CLR): Global contrastive losses.

Who it’s for

It is designed for AI researchers and practitioners working on imbalanced datasets, ranking systems, or contrastive representation learning who need to optimize for specific performance metrics rather than generic cross-entropy loss.

Highlights

  • Broad Application Support: Covers imbalanced classification, learning to rank, and contrastive learning.
  • Diverse Metric Support: Includes optimizations for AUROC, AUPRC, NDCG, and pAUC.
  • Integrated Tooling: Features a dedicated LibAUC Trainer and support for distributed training.
  • Model Integration: Compatible with PyTorch and integrates with Hugging Face models.

Related

  • Project
    google/rax

    A JAX-based Learning-to-Rank library providing off-the-shelf ranking losses, metrics, and transformations for optimizing the relative order of items.

  • Project
    deel-ai/puncc

    A Python library for predictive uncertainty calibration and conformalization, providing rigorous uncertainty estimations for any machine learning model across various tasks.

  • Project
    deepspeedai/DeepSpeed

    A deep learning optimization library that enables the training of models with trillions of parameters by optimizing memory and compute efficiency across distributed systems.

  • Project
    huggingface/lerobot

    A PyTorch-native library providing models, datasets, and tools to standardize real-world robotics, lowering the barrier to entry for training and deploying AI policies.

  • Project
    ntucllab/libact

    A Python package for pool-based active learning that provides various query strategies and a meta-algorithm to automatically select the best strategy.