mlsanityData to 7 Oct 2026

Adversarial Robustness in Two-Stage Learning-to-Defer: Algorithms and Guarantees

arXiv 2502.01027 ↗First posted 3 Feb 2025stat.ML cs.LGcited by 16

Not on the map. Why?

Abstract (arXiv)

Two-stage Learning-to-Defer (L2D) enables optimal task delegation by assigning each input to either a fixed main model or one of several offline experts, supporting reliable decision-making in complex, multi-agent environments. However, existing L2D frameworks assume clean inputs and are vulnerable to adversarial perturbations that can manipulate query allocation--causing costly misrouting or expert overload. We present the first comprehensive study of adversarial robustness in two-stage L2D systems. We introduce two novel attack strategie--untargeted and targeted--which respectively disrupt optimal allocations or force queries to specific agents. To defend against such threats, we propose SARD, a convex learning algorithm built on a family of surrogate losses that are provably Bayes-consistent and (R, G)-consistent. These guarantees hold across classification, regression, and multi-task settings. Empirical results demonstrate that SARD significantly improves robustness under adversarial attacks while maintaining strong clean performance, marking a critical step toward secure and trustworthy L2D deployment.

Key facts. Cited by 16 papers in the map (data through 7 Oct 2026). Builds on 31 of its 61 references that are in the map. First posted on arXiv 3 Feb 2025.

Builds on

31 of 61 references.

References · 31

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2015Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networkscited by 9,617
  3. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  4. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
  5. 2013Intriguing properties of neural networkscited by 5,114
  6. 2017MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applicationscited by 3,827
  7. 2018Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labelscited by 994
  8. 2017Selective Classification for Deep Neural Networkscited by 799
  9. 2018Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Surveycited by 459
  10. 2017Robust Loss Functions under Label Noise for Deep Neural Networkscited by 441
  11. 2020Uncovering the Limits of Adversarial Training against Norm-Bounded Adversarial Examplescited by 275
  12. 2020Consistent Estimators for Learning to Defer to an Expertcited by 262
19 more
  1. 2017Predict Responsibly: Improving Fairness and Accuracy by Learning to Defercited by 228
  2. 2023Cross-Entropy Loss Functions: Theoretical Analysis and Applicationscited by 206
  3. 2023Who Should Predict? Exact Algorithms For Learning to Defer to Humanscited by 68
  4. 2022Learning to Defer to Multiple Experts: Consistent Surrogate Losses, Confidence Calibration, and Conformal Ensemblescited by 59
  5. 2021Calibration and Consistency of Adversarial Surrogate Lossescited by 47
  6. 2022Sample Efficient Learning of Predictors that Complement Humanscited by 46
  7. 2020Calibrated Surrogate Losses for Adversarially Robust Classificationcited by 36
  8. 2024Regression with Multi-Expert Deferralcited by 29
  9. 2023In Defense of Softmax Parametrization for Calibrated and Consistent Learning to Defercited by 26
  10. 2024Realizable H-Consistent and Bayes-Consistent Loss Functions for Learning to Defercited by 26
  11. 2024A Universal Growth Rate for Learning with Smooth Surrogate Lossescited by 24
  12. 2024Cardinality-Aware Set Prediction and Top-k Classificationcited by 23
  13. 2024Multi-Label Learning with Stronger Consistency Guaranteescited by 22
  14. 2024H-Consistency Guarantees for Regressioncited by 21
  15. 2024Enhanced H-Consistency Boundscited by 18
  16. 2024A Two-Stage Learning-to-Defer Approach for Multi-Task Learningcited by 16
  17. 2024Optimal Query Allocation in Extractive QA with LLMs: A Learning-to-Defer Framework with Theoretical Guaranteescited by 15
  18. 2022Towards Consistency in Adversarial Classificationcited by 12
  19. 2023On Classification-Calibration of Gamma-Phi Lossescited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 8; 2026: 8.

16 papers.

Sort:

  1. 2025RouterEval: A Comprehensive Benchmark for Routing LLMs to Explore Model-level Scaling Up in LLMscited by 58
  2. 2025Mastering Multiple-Expert Routing: Realizable H-Consistency and Strong Guarantees for Learning to Defercited by 16
  3. 2025Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Datacited by 15
  4. 2025Budgeted Multiple-Expert Deferralcited by 11
  5. 2025Why Ask One When You Can Ask k? Learning-to-Defer to the Top-k Expertscited by 6
  6. 2025Principled Algorithms for Optimizing Generalized Metrics in Binary Classificationcited by 6
  7. 2025Adversarial Robustness in One-Stage Learning-to-Defercited by 5
  8. 2025Identity-Free Deferral For Unseen Expertscited by 4
  9. 2026Learning-to-Defer in Non-Stationary Time Series via Switching State-Space Modelscited by 4
  10. 2026When More Experts Hurt: Underfitting in Multi-Expert Learning to Defercited by 2
  11. 2026Optimized Deferral for Imbalanced Settingscited by 2
  12. 2026Linear-Core Surrogates: Smooth Loss Functions with Linear Rates for Classification and Structured Predictioncited by 2
4 more
  1. 2026Consistent Learning-to-Defer with Expert-Conditional Advicecited by 1
  2. 2026Online Learning-to-Defer with Varying Expertscited by 0
  3. 2026RACER: Role-Aligned Competence Estimation for Human-AI Routingcited by 0
  4. 2026Should I stay or should I show? Learning to selectively disclose informationcited by 0