UC-Search: Risk-Aware Test-Time Search for Delayed Constrained Time-Series Control

2026-06-24 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Data Science & Analytics · Depth: Expert, medium

Summary

UC-Search is a novel model-agnostic test-time wrapper designed for delayed constrained time-series control, addressing the challenge of deploying models that require decisions under uncertainty and hard feasibility constraints. It operates by taking forecasts or action scores from a backbone model, using a feasibility automaton to roll forward candidate paths, and then employing bounded search to identify the initial action of a risk-adjusted feasible trajectory. The system integrates epistemic, aleatoric, and propagated uncertainty as path-risk terms. UC-Search demonstrated strong performance on a 9-family, 33-series delayed-control suite, achieving normalized threshold gains of +3.1675, +2.3328, and +2.5038 against CEM, MPPI, and risk-aware random, respectively. A 48-series M4 inventory audit also showed substantial improvements, including +13556.7547 over classic base-stock control.

Key takeaway

For Machine Learning Engineers designing or deploying time-series control systems with inherent delays and hard feasibility constraints, you should consider integrating risk-aware test-time search wrappers like UC-Search. This approach can significantly improve decision quality and feasibility by accounting for various uncertainties and non-myopic value. Evaluate its potential to enhance robustness and performance in your specific constrained environments, particularly for applications like inventory management where traditional methods often fall short.

Key insights

UC-Search provides a risk-aware test-time search framework for delayed, constrained time-series control decisions.

Principles

• Delayed feasible-set coupling can create non-myopic value.
• Uncertainty (epistemic, aleatoric, propagated) can be modeled as path-risk terms.
• Test-time search can enhance model deployment under constraints.

Method

UC-Search wraps a backbone model, using a feasibility automaton to explore candidate paths and bounded search to select the first action of a risk-adjusted feasible trajectory.

In practice

• Apply UC-Search to delayed inventory management.
• Integrate risk-aware search for time-series control.
• Use UC-Beam or UC-MCTS for diagnostic analysis.

Topics

• Time-Series Control
• Risk-Aware AI
• Test-Time Search
• Uncertainty Quantification
• Inventory Management
• Feasibility Constraints

Code references

• Time-Search/TimeSearch-R
• XMUDeepLIT/SAAS

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

Related on AIssential

• Deliberative Searcher: Improving LLM Reliability via Reinforcement Learning with constraints — Integrating confidence calibration with retrieval-based search via constrained reinforcement learning improves LLM reliability and accuracy.
• In-Context Learning for Data-Driven Censored Inventory Control — ICGPS uses generative models and in-context learning to improve inventory control under censored demand.
• Multi-Armed Bandits with Arriving Arms: Sequential Screening, Dynamic Regret, and Sublinear Guarantees — UCB-AA addresses dynamic regret in multi-armed bandits where new arms arrive sequentially, improving efficiency.
• StarOR: Synergizing Tree Search and Test-Time Reinforcement Learning for Optimization Modeling — StarOR integrates MCTS and Test-Time Reinforcement Learning to adaptively refine optimization modeling policies without ground-truth labels.
• Conformal Risk-Averse Decision Making with Action Conditional Guarantee — Action-conditional conformal prediction offers explicit safety guarantees for each action, enhancing risk-averse decision making with ML models.
• Standard Acquisition Is Sufficient for Asynchronous Bayesian Optimization — This paper challenges the prevailing "Hypothesis 1" in asynchronous Bayesian optimization (BO), which posits that standard acquisition functions lead to redundant queries and necessitate complex diver…

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.