Inference-Time Decision Calibration for Temporal Classification

2026-06-14

AI Summary

This work addresses the limitations in temporal classification performance, which stem not only from inadequate representations but also from suboptimal utilization of available evidence during decision-making. The study identifies such deficiencies as a decision calibration issue and introduces a representation–calibration decoupling framework. At inference time, the backbone network is frozen, and a residual multi-scale auxiliary branch is introduced to generate supplementary logits. A branch-aware calibrator dynamically fuses the original and auxiliary evidence without requiring retraining of the backbone. This approach enables dual intervention—recovering missing evidence and better leveraging underutilized evidence. Extensive experiments on temporal datasets such as FI-2010 and PTB-XL demonstrate significant performance gains, particularly in high-noise or weak-representation scenarios.

Abstract

Temporal classification errors are often treated as representation failures, but they can also arise from how available evidence is converted into decisions. This paper proposes a representation--calibration decomposition for temporal classification. We keep a trained native classifier frozen and separate two inference-time interventions: a conservative residual multi-scale branch that adds auxiliary logits to the native prediction, and a post-hoc branch-aware calibrator that recombines native and residual evidence at decision time. This design distinguishes missing temporal evidence from underused decision-level evidence without retraining the backbone. Across FI-2010, PTB-XL, UCI-HAR, MHEALTH, and HARTH, we find that gains are strongly regime-dependent. Residual multi-scale evidence is most useful in noisy or representation-limited settings, especially short-horizon FI-2010 and weaker recurrent backbones, while branch-aware calibration helps when native and auxiliary logits contain complementary evidence not fully exploited by the raw decision rule. Near-saturated settings show limited gains from either intervention. These results suggest that temporal classification should be understood not only as representation learning, but also as the problem of trusting, combining, and calibrating evidence from multiple views.

Problem

• temporal classification
• decision calibration
• inference-time intervention
• evidence combination
• representation learning

Innovation

• inference-time calibration
• temporal classification
• residual multi-scale branch
• branch-aware calibrator
• decision-level evidence