Humans learn and make decisions in large part through recognizing patterns. Research on the neurocomputational mechanisms of deterministic sequence learning (i.e., learning of patterns in sequences of states), where the decision-maker can infer the underlying structure and use this knowledge to make better predictions, has lagged behind research on neural representations of sequences and probabilistic reinforcement learning. The brain’s ability to discern between deterministic patterns and random sequences is particularly ambiguous. In contrast to probabilistic learning, when it is known that two states are connected, the main challenge in this problem is determining how strong that connection is (i.e., the probability).
While many studies investigate sequences learned over many training blocks, the current study explores how subjects detect the presence of a predictable sequence. Participants in the research were shown 50 series of 12 photographs, sometimes in a pattern and other times in a random sequence, that featured different combinations of the three shots of a hand, a face, and a landscape. When participants picked which photo they believed would appear next, an MRI scanner inside of a lab monitored which regions of their brains were active.
We could see what parts of the brain were activated when participants figured out that there was a pattern – or realized that there was no pattern.[..] If they don’t know what image is coming next, they have to wait a while.[..] But once they figured out a pattern, they responded more quickly and we could see how that was reflected in their brains.
Ian Krajbich
A distinct type of learning model, known as a probabilistic model, has long been researched by scientists. In the probabilistic paradigm, humans gain knowledge by calculating the likelihood that one occurrence would follow another. For instance, you could discover that following a defeat, your favorite sports team often wins two out of three games. But that paradigm does not account for pattern recognition. The way that probabilistic and pattern learning engage the brain is different: with patterns, you can predict when a certain event will occur.
People in our study aren’t just predicting the odds of which photo will show up next. They are learning patterns and developing rules that guide their decision and make them faster and more accurate.
Different brain regions were shown to be engaged in this study based on the two types of uncertainty that the individuals experienced. Uncertainty of the following image was one type of uncertainty. The results revealed—not surprisingly—that the same brain regions involved in learning probabilistic probabilities were also active during this activity. The second type of ambiguity involved whether the displayed pictures had a pattern. The ventromedial prefrontal cortex, a separate region of the brain, became active as the subjects processed this question. Other studies have demonstrated that this brain region is connected to reward, one interpretation being that people may be getting a sense of reward for figuring out whether there is a pattern or not.
Another area of the brain that was particularly active during pattern recognition was the hippocampus, with those who had more hippocampal activity learning quicker. Overall, the research demonstrated that pattern learning and probabilistic learning are processed in the brain in distinct ways.
Neurocomputational Dynamics of Sequence Learning. Arkady Konovalov, Ian Krajbich.
Published: May 2018
DOI: 10.1016/j.neuron.2018.05.013