Ronald van den Berg

Research interests

For a complete list of published work, see the publications page.

Models of Cognition

Broadly speaking, there are two classes of models of human cognition. The first class is fueled by the work of Kahneman and Tversky, suggesting that cognitive processes are based on simple heuristics. The other class has its roots in mathematical theory and claims that cognitive processes are based on statistical inference. Over the past few years we have performed several studies to try tease apart these two theories. Using rigorous model comparisons, we found very little evidence for the heuristical models. However, we also found evidence against fully optimal (or: Bayesian) decision making. Altogether, our findings suggest that human cognitive systems are grounded in statistical principles, but suffer from imperfections.

key publications

E Stengård, P Juslin, U Hahn, R van den Berg (2022).
On the generality and cognitive basis of base-rate neglect.
Cognition (in press)

E Stengård, P Juslin, R van den Berg (2022).
How deep is your Bayesianism? Peeling the layers of the intuitive Bayesian.
Decision (in press)

Models of Visual Perception

Perception is an inference problem: the brain is interested in world states (e.g., "is a predator present?"), but only has access to incomplete and ambiguous information about these world states. As a result, perfect accuracy is typically out of reach. However, there is a strategy that guarantees to give the best possible accuracy when making decisions under uncertainty: probabilistic Bayesian inference. There is overwhelming evidence showing that human behavior is consistent with probabilistic Bayesian inference when making simple perceptual inferences, such as combining two information cues (see for example Knill & Pouget 2004). In a series of experiments and model comparisons, we examined inference strategies in more complex, multi-element tasks, including visual search, sameness judgment, change detection, and change localization. Somewhat surprisingly, we found that human strategies are compatiable with the optimal strategy in all tested tasks. This suggests that optimality in perceptual inference extends to tasks well beyond simple cue combination.

key publications

E Stengård, R van den Berg (2019). Imperfect Bayesian inference in visual perception. PLoS Computational Biology PDF

R van den Berg, M Vogel, K Josic, WJ Ma (2012). Optimal inference of sameness. PNAS 109 (8), 3178-83. DOI PDF

WJ Ma*, V Navalpakkam*, JM Beck*, R van den Berg*, Pouget A (2011). Behavior and neural basis of near-optimal visual search. Nature Neuroscience 14 (6), 783-90. DOI PDF

Models of Visual Working Memory

Working memory provides us with the ability to briefly remember pieces of information, which is essential for our functioning (e.g., imagine trying to read a book without remembering any of the words you read except the one you're looking at). The classic slot theory of working memory claims that working memory consists of a small number of slots, each of which can be used to store an item (e.g., Zhang & Luck 2008). In a series of psychophyiscal experiments and rigorous model comparisons, we have found that the classic slot theory is untenable. Instead, we found strong evidence indicating that working memory is a continuous quantity that is roughly equally spread across items, with variability in memory precision. Moreover, we have found some initial evidence that visual working memory might be a "resource-rational system", optimizing its resource allocation to the demands of the task.

key publications

R van den Berg, WJ Ma.
A resource-rational theory of set size effects in visual working memory (2018).
eLife. DOI PDF

R van den Berg, H Shin, WC Chou, R George, WJ Ma (2012).
Variability in encoding precision accounts for visual short-term memory limitations.
PNAS 109 (22). DOI PDF

R van den Berg, E Awh, WJ Ma (2014).
Factorial comparison of working memory models.
Psychological Review. 121(1), 124-149. DOI PDF

Models of Metacognition

Our decisions and judgments about the world are usually accompanied by a sense of confidence. This is often referred to as "metacognition", because it is essentially "cognition about cognition". In my research, I am interested in uncovering the computational mechanisms that underlie confidence formation in the brain: from what kind of information is our sense of confidence derived and through what kind of computations is this information mapped to a confidence judgment?

key publications

R van den Berg, A Yoo, WJ Ma (2017). Fechner's law in metacognition: a quantitative model of visual working memory. Psychological Review DOI PDF

R van den Berg, A Zylberberg, A Zylberberg, R Kiani, M Shadlen, D Wolpert (2016). Confidence is the bridge between multi-stage decisions. Current Biology DOI PDF

R van den Berg, K Anandalingam, A Zylberberg, R Kiani, M Shadlen, D Wolpert (2016). A common mechanism underlies changes of mind about decisions and confidence. eLife 14 (6), 783-90. DOI PDF

Visual crowding

When a visual item is surrounded by other items, we have difficulty recognizing its identity. To experience this effect, look at the + below and try to read the text to the right:

+        A

+     X A B

You will notice that the "A" is easy to recognize when presented in isolation, but not when it is surrounded by other letters. This "crowding" phenomenon places a fundamental constraint on human vision that limits performance on numerous tasks. While an overwhelming amount of data and qualitative theories are available, very little quantitative modeling has been performed to make sense of those data and theories. One of the qualitative theories is that crowding results from spatial feature integration necessary for object recognition. To test this theory, we developed a quantitative and physiologically plausible model for spatial feature integration. We found that this model coherently accounts for several fundamental properties of crowding, including critical spacing, “compulsory averaging”, and a foveal-peripheral anisotropy. Our results suggest that rather than being a visual defect, crowding may be a signature of optimal feature integration.

key publications

R van den Berg, A Johnson, A Martinez, AL Schepers, FW Cornelissen (2012). Comparing crowding in human and ideal observers. Journal of Vision, 12(8):1-15. DOI PDF

R van den Berg, JBTM Roerdink, FW Cornelissen (2010). A neurophysiologically plausible population code model for feature integration explains visual crowding. PLoS Computational Biology, 6(1):e1000646. DOI PDF

R van den Berg, JBTM Roerdink, FW Cornelissen (2007). On the generality of crowding: Visual crowding in size, saturation, and hue compared to orientation. Journal of Vision, 7(2):1-11. DOI PDF