Research map / Learning theory and algorithms / Combinatorial and search optimization

Reduced Cost Influence Functions for Predict-then-Optimize under Noisy Data

NeurIPS 2026 · Learning theory and algorithms › Combinatorial and search optimization · approach: Constrained neural network optimization

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes reduced-cost influence functions to identify and down-weight noisy training samples in predict-then-optimize pipelines where a model predicts LP cost vectors from a large noisy dataset plus a small clean set.

Problem

Predict-then-optimize pipelines degrade when the upstream cost-prediction model is trained on large amounts of noisy data, while only a small clean set is available.

Approach

Adapts influence functions to the decision-focused objective by using reduced costs of the downstream linear program, so that influence on decision quality (rather than prediction error) scores each noisy training sample.

Experimental setup

Builds on: influence functions

Closest papers by meaning

• Leaving the Nest: Going beyond Local Loss Functions for Predict-Then-Optimize (AAAI 2024 · similarity 0.87)
• Overcoming the Optimizer's Curse: Obtaining Realistic Prescriptions from Neural Networks (ICML 2024 · similarity 0.86)
• DFF: Decision-Focused Fine-Tuning for Smarter Predict-Then-Optimize with Limited Data (AAAI 2025 · similarity 0.85)
• What Data Enables Optimal Decisions? An Exact Characterization for Linear Optimization (NeurIPS 2025 · similarity 0.85)
• A Versatile Influence Function for Data Attribution with Non-Decomposable Loss (ICML 2025 · similarity 0.84)
• Solver-Free Decision-Focused Learning for Linear Optimization Problems (NeurIPS 2025 · similarity 0.84)