Confidence as hyperparameter tuning for sequential decision-making

Metacognition

Sequential decision-making

• Speed/Accuracy Tradeoff.
• Canonical model: integration of noisy information until a threshold is reached.
• Many refinements: multi-alternative choice, impact of learning, change of mind, etc.

Confidence for decision-making

• Quantifies the degree of certainty associated to a decision.
• Canonical model: post-decisional computation based on accumulated info.
• Can be used to alter the subsequent decisions.

Proposed architecture: confidence as hyperparameter tuning for sequential decision-making

Current & next steps

• Article for ESANN 2025.
• Detailed review of confidence for decision-making (future PhD chapter).
• First experiment on very basic task.

Questions ?