Inference-Time Decision Calibration for Temporal Classification
For researchers in temporal classification (e.g., time-series, activity recognition), this work provides a framework to diagnose whether errors stem from representation or decision-level issues, but the gains are incremental and regime-specific.
The paper proposes a representation-calibration decomposition for temporal classification, using two inference-time interventions (residual multi-scale branch and branch-aware calibrator) to separate missing temporal evidence from underused decision-level evidence. Across five datasets, gains are regime-dependent, with residual evidence most useful in noisy settings and calibration helping when complementary evidence exists, while near-saturated settings show limited gains.
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.