skwdro: a library for Wasserstein distributionally robust machine learning

2025-12-31 · Source: JMLR · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, quick

Summary

skwdro is a new Python library designed to simplify the training of robust machine learning models, released in 2026. It utilizes distributionally robust optimization with Wasserstein distances, a technique common in optimal transport and machine learning. The library offers a wrapper for PyTorch modules, allowing users to robustify model losses with minimal code modifications. Additionally, skwdro provides scikit-learn compatible estimators for several popular objectives. Its core implementation incorporates entropic smoothing of the robust objective, which enhances model flexibility. The library's code is accessible on GitHub, and comprehensive documentation is available via Read the Docs.

Key takeaway

For AI engineers and research scientists aiming to build more resilient machine learning models, skwdro offers a streamlined approach. You should consider integrating this Python library to robustify your PyTorch models or leverage its scikit-learn compatible estimators, as it simplifies the application of distributionally robust optimization with minimal code changes, potentially improving model stability against data shifts.

Key insights

skwdro simplifies robust ML model training using Wasserstein distributionally robust optimization with PyTorch and scikit-learn compatibility.

Principles

• Robustify models via Wasserstein distances
• Entropic smoothing enhances flexibility

Method

skwdro wraps PyTorch modules to robustify loss functions and provides scikit-learn compatible estimators, using entropic smoothing for flexibility.

In practice

• Integrate with existing PyTorch models
• Use scikit-learn estimators for objectives

Topics

• skwdro
• Distributionally Robust Optimization
• Wasserstein Distance
• Robust Machine Learning
• PyTorch

Code references

• iutzeler/skwdro
• JmlrOrg/v27

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• Robust Local Polynomial Regression with Similarity Kernels — RSKLPR enhances LPR robustness by using similarity kernels that weight data based on both predictor and response variables.
• $L^2$ over Wasserstein: Statistical Analysis for Optimal Transport — The $L^{2}$ over Wasserstein space unifies optimal transport under statistical uncertainty, extending its geometric and flow properties to random measures.
• Distributionally Robust Set Representation Learning Under Inference-Time Element Corruption — SW-DRSO enhances set representation learning robustness by optimizing for worst-case inference-time corruption using a barycentric adversary.
• I-SAFE: Wasserstein Coherence Metrics for Structural Auditing of Scientific AI Models — Scientific AI model performance benchmarks are insufficient to confirm alignment with domain-relevant structural behavior.
• SPRI: SVD-Partitioned Residual Initialization for Data-Constrained MoE Upcycling — SPRI leverages SVD-partitioned residuals from pretrained FFNs to create diverse MoE experts, improving upcycling under data constraints.
• Distance-Matrix Wasserstein Statistics for Scalable Gromov--Wasserstein Learning — DMW offers a scalable, provable relaxation of Gromov--Wasserstein for comparing structural data via sampled distance matrices.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by JMLR.