Adaptive Learning needs Attention, Meta-learning and Working Memory

We tested which model mechanisms best explain how six animals learn attention sets and found a common set of most-important behavioral mechanisms that account for learning success.
When learning attention sets is easy value based reinforcement learning and working memory are powerful, but when learning problems are more complex learning is more efficient with attention and a meta-learning process that help speeding up learning when errors accumulate. (See our paper Womelsdorf at al. (2022) Learning at variable attentional load requires cooperation between working memory, meta-learning and attention-augmented reinforcement learning. Journal of Cognitive Neuroscience 34(1) 79-107.)

Related News

New 3D-object type with controllable feature space published with open-sourced code

The lab has a new publication showcasing and describing details of Quaddles: A multidimensional 3D object set with parametrically-controlled and customizable features. Quaddles have 5+ feature dimensions, each with multiple possible feature values that can be parametrically morphed, making it possible to generate a near arbitrary number of unique objects. Thanks to Marcus and Milad […]

Acetylcholine, Dopamine, and Glutamate measured in three brain areas with a new SPME-probe

Our new paper shows reliable Multi-Neuromodulator measurements in the awake nonhuman primate in prefrontal cortex, premotor cortex and the basal ganglia using a new chemical sensing probe. The probe uses principles of Solid-Phase-Microextraction and is a development by the SPME pioneer and collaborator Prof. Janusz Pawliszyn from the University of Waterloo. The paper can be […]

ACC causally supports learning -difficult- attention sets

We used focused ultrasound (FUS) sonication of the anterior cingualte and striatum to disrupt local processing during learning. FUS in ACC slowed down learning of atetntion sets – but only when the attentional demands were high and the task included the risk of loosing already attaiuned reward tokens. Under these cognitive and motivaitonally challenging conditions […]