Normative evidence accumulation in unpredictable environments.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Decision Neuroscience

Background:

• Decisions in dynamic environments require processing noisy stimuli, balancing signal identification and change detection.
• Existing normative models have limitations in explaining faster decision processes.
• Adaptive learning is crucial for navigating unpredictable environments.

Purpose of the Study:

• To present a novel normative formulation of adaptive learning models for decision-making.
• To investigate how humans learn and utilize expectations about changing environments.
• To provide a unified, empirically supported account of decision-making dynamics.

Main Methods:

• Developed a leaky accumulator model with non-absorbing bounds for decision dynamics.
• Derived model dynamics for both discrete and continuous evidence cases.
• Tested the model with human subjects on two distinct tasks.

Main Results:

• Human subjects demonstrated learning of environmental change expectations, though imperfectly.
• Decision-making performance aligned with the predictions of the novel normative model.
• The model successfully balanced signal identification and change detection based on learned expectations.

Conclusions:

• The proposed normative model offers a unified framework for understanding decision-making in unpredictable environments.
• Results provide insights into the role of expectation-driven dynamics in neural signals during decision-making.
• This work bridges normative theory and empirical observations in adaptive learning and decision processes.