Our visual perception is remarkably accurate. However, the visual input that enters our brains is often ambiguous due to factors such as eye movements and biological noise. A major challenge in neuroscience is understanding how our brains can make accurate visual decisions based on such ambiguous sensory input.
Luckily, our visual world often changes in predictable, regular ways over time. These temporal regularities can be exploited to predict the future from the recent past, thereby facilitating visual decisions. For instance, our computer will not suddenly disappear in front of our eyes, enabling us to predict that it will still be there after briefly looking up from our desks. Crucially, however, natural environments exhibit a multitude of different temporal regularities. For example, a traffic light that recently turned green can be expected to remain green for a while, allowing you to maintain speed while passing a junction. Conversely, a yellow traffic light can rapidly change to red, thus prompting you to decelerate. Exploiting these temporal regularities, therefore, needs adaptation of perceptual decisions to such sequential patterns. Thus far, it has been unclear how brains can achieve this.
The HFSP Fellowship Awardee Matthias Fritsche, from the University of Oxford, UK, has been studying this topic together with his host supervisor, Armin Lak. In a recent paper published in Nature Communications, Fritsche and colleagues have demonstrated for the first time that mice can adapt their visual decisions to temporal regularities, enabling the researchers to probe the computational principles and neural circuits underlying this adaptation. The results led to the surprising discovery that seemingly complex adaptations to temporal regularities can be explained by relatively simple learning algorithms, commonly known as reinforcement learning algorithms. Furthermore, the researchers revealed that the neurotransmitter dopamine, which plays a pivotal role in learning from reward, tracked the behavioral adaptations and signaled key components of the reinforcement learning algorithm.

This study provides important insights into the computational principles and neural mechanisms underlying the brain's ability to exploit temporal patterns in the environment in order to make accurate visual decisions. While such processes appear to function effortlessly in healthy individuals, they are impaired in several psychiatric disorders, such as autism, schizophrenia, as well as Parkinson's disease, in which the dramatic loss of dopamine neurons causes not only a wide range of motor deficits but also perceptual hallucinations and impaired decision-making. The current study, therefore, contributes to an exciting developing picture of how accurately our environment shapes perception and decisions and how these processes may go away in clinical conditions.