A new FTPL algorithm with self-concordant perturbations achieves tilde O(sqrt(T)) regret for all bounded proper losses and O(log T) regret for bounded smooth proper losses in U-calibration.
Calibration error for decision making, 2024
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 5verdicts
UNVERDICTED 5roles
background 1polarities
background 1representative citing papers
First approximate calibration results for discrete properties in multiclass settings via Lipschitz intermediaries for strongly orderable discrete properties.
Algorithms achieve adaptive calibration bounds of order min{sqrt(T) + (T C)^{1/3}, sqrt(K T)} for l1 error and min{(1+C)^{1/3}, K} for l2 and pseudo-KL error, where K and C are unknown non-stationarity measures.
Task calibration aligns LLM distributions in latent task spaces to make MBR decoding provably optimal and improve generation quality.
Introduces decision-alignment to evaluate uncertainty metrics against downstream decision utilities and proposes prior-weighted proper scoring rules that align better in benchmarks and case studies.
citing papers explorer
-
Toward Simultaneously Optimal Regret in U-Calibration
A new FTPL algorithm with self-concordant perturbations achieves tilde O(sqrt(T)) regret for all bounded proper losses and O(log T) regret for bounded smooth proper losses in U-calibration.
-
Smoothed Elicitation Complexity for Approximate $\Gamma$-calibration of Discrete Classification Tasks
First approximate calibration results for discrete properties in multiclass settings via Lipschitz intermediaries for strongly orderable discrete properties.
-
Adaptive Calibration in Non-Stationary Environments
Algorithms achieve adaptive calibration bounds of order min{sqrt(T) + (T C)^{1/3}, sqrt(K T)} for l1 error and min{(1+C)^{1/3}, K} for l2 and pseudo-KL error, where K and C are unknown non-stationarity measures.
-
Task-Aware Calibration: Provably Optimal Decoding in LLMs
Task calibration aligns LLM distributions in latent task spaces to make MBR decoding provably optimal and improve generation quality.
-
Decision-Aligned Evaluation of Uncertainty Quantification
Introduces decision-alignment to evaluate uncertainty metrics against downstream decision utilities and proposes prior-weighted proper scoring rules that align better in benchmarks and case studies.