Learning to Defer in Non-Stationary Time Series via Switching State-Space Models

2026-01-30

AI Summary

This work addresses the challenge of modeling non-stationary time series when expert availability varies over time and only partial feedback is observed. To this end, the authors propose L2D-SLDS, a factored switching linear Gaussian state-space model that jointly captures shared global dynamics and expert-specific latent states. The approach models expert residuals and enables context-aware dynamic routing, while supporting online expert registration and pruning. A routing policy based on information-directed sampling (IDS) is introduced to balance predictive accuracy against the information gain from querying individual experts. Experimental results demonstrate that the proposed method significantly outperforms contextual multi-armed bandit baselines and ablated variants without shared factors, achieving marked improvements in predictive performance.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling
• Intelligent Robots: State Estimation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We study Learning to Defer for non-stationary time series with partial feedback and time-varying expert availability. At each time step, the router selects an available expert, observes the target, and sees only the queried expert's prediction. We model signed expert residuals using L2D-SLDS, a factorized switching linear-Gaussian state-space model with context-dependent regime transitions, a shared global factor enabling cross-expert information transfer, and per-expert idiosyncratic states. The model supports expert entry and pruning via a dynamic registry. Using one-step-ahead predictive beliefs, we propose an IDS-inspired routing rule that trades off predicted cost against information gained about the latent regime and shared factor. Experiments show improvements over contextual-bandit baselines and a no-shared-factor ablation.

Problem

• Learning to Defer
• Non-Stationary Time Series
• Partial Feedback
• Time-Varying Expert Availability
• Expert Routing

Innovation

• switching state-space models
• learning to defer
• non-stationary time series
• information-directed sampling
• dynamic expert routing