Influences of uncertainty on context guidance and suppression in context learning: neural and computational mechanisms

• Applicants: Dr. Siyi Chen, Ph.D.; Professor Dr. Zhuanghua Shi
• Subject Area: Biological Psychology and Cognitive Neuroscience ; General, Cognitive and Mathematical Psychology
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 505693104

Repeatedly encountering a target object consistently positioned within an invariant arrangement of distractor objects (‘context’) can affect the pre- and post-selective attentional processing and decision-making processes involved in the search and thus significantly expedite the final reaction time - an effect that is attributed to statistical (long-term) context learning ('contextual learning / cueing', see Chun & Jiang, 1998). Previous studies on context learning used almost exclusively constant target-context relationships (the target always appeared at the same place in a given configuration of identical distractors) neglecting the actual dynamics of context learning in the real environment, in which the target-context relations may be subject to fluctuations, i.e., 'uncertainty' or 'volatility'. Hence, it is theoretically important to understand how the uncertainty that accompanies such fluctuations dynamically affects statistical context learning. A distinction must be drawn between three basic forms of uncertainty in statistical learning environments: (1) uncertainty in the predictivity of the spatial distractor configuration with respect to the target position, (2) uncertainty in the predictivity of the distractor identity with respect to the target position or identity, and (3) uncertainty in the temporal sequence of repeated configurations with respect to the extraction and thus the learning of (more or less) invariant target-context relations. In particular, under conditions of more or less extreme (maximum or minimum) un-/certainty in factors (1) and (2), the context effect might be mediated by very different mechanisms: context-based control ('guidance', i.e., 'cueing') vs. context suppression. The research project uses (variations of) Chun & Jiang’s (1998) classical 'contextual-cueing' paradigm as a testbed to systematically examine the three basic sources of uncertainty in context learning and thus to delineate the two mechanisms (guidance vs. suppression) and their interplay by means of a combination of behavioral approaches with neuroscientific methods (EEG and fMRI) as well as mathematical modeling. The ultimate goal is to develop a theoretically coherent neuro-cognitive model of active context learning based on the core concept of ‘predictive coding’.

• DFG Programme: Research Grants