Using 35.4 million responses, we measured human observers’ spatial working memory across 80,000 patterns. A convolutional neural network (CNN) was trained to explain the data. Notably, a relatively simple 67-parameter cognitive encoding (CE) model performed nearly as well as the CNN. This CE model offers a unifying framework for spatial working memory mechanisms, integrating both established and novel principles. 

Huang, L. (2023). A quasi-comprehensive exploration of the mechanisms of spatial working memory. Nature Human Behaviour, 7, 729–739. https://doi.org/10.1038/s41562-023-01559-z .

We analyzed 40 million responses to assess human observers’ working memory of colors across 10,000 patterns. A neural network was developed to account for the data. Strikingly, a simpler 57-parameter CE model outperformed the neural network. This model provides a tentative overarching framework for visual working memory, bridging classic and contemporary theories. 

Huang, L. (2025). Comprehensive exploration of visual working memory mechanisms using large-scale behavioral experiment. Nature Communications, 16, 1383.  https://doi.org/10.1038/s41467-025-56700-5 .

Using the same patterns as in Huang (2023), we collected 42 million aesthetic preference judgments. A CNN explained nearly all (98.3%) of the explainable variance in the data. Remarkably, a very simple CE model also captured the majority (93.6%) of the explainable variance, revealing that aesthetic preferences for spatial patterns are driven by regularity, with three key components: proximity, continuity, and linearity. 

Huang, L. (2025). Aesthetic preferences among spatial patterns: Large-scale experiment, comprehensive exploration, and a three-component regularity-based model. Psychology of Aesthetics, Creativity, and the Arts. https://doi.org/10.1037/aca0000774 .

In an ongoing project, we investigate the cognitive mechanisms underlying shape perception by examining why some shapes are perceived as fundamentally distinct while others appear similar. Using a carefully selected set of 400 shapes, we systematically measured perceptual similarity across all possible pairwise combinations (400×399 = 159,600 pairs). Our research aims to develop a computationally interpretable model that explains these variations in shape similarity judgments.

In an ongoing collaboration with Jeremy Wolfe, we investigate the mechanisms underlying color-based visual search. We have measured observers' search performance across hundreds of thousands of pre-generated visual displays. While our initial analyses confirm established findings, they also reveal substantial unexplained variance in search efficiency. Our research aims to develop a computationally interpretable model that precisely characterizes the underlying mechanisms governing color search behavior.

In collaboration with Ke Zhou, we investigate the core psychological dimensions underlying facial expression perception. Using over 24 million naturalistic similarity judgments across 750 facial images, we train neural networks to learn a low-dimensional, interpretable embedding. This space allows us to determine which dimensions are categorical, which are continuous, and how they together shape human perception. Our aim is to reconcile competing emotion theories through a unified, data-driven representational framework.

In collaboration with Ming Meng, we investigate human intuitive problem-solving using the Traveling Salesman Problem (TSP). By testing 150,000 location sets, we discovered that humans are remarkably skilled at intuitively solving the TSP—yet they systematically deviate from optimality in specific scenarios. These deviations likely stem from the use of efficient but imperfect heuristic strategies. Our current work focuses on precisely identifying, characterizing, and modeling these strategies.

In an ongoing project, we investigate the cognitive mechanisms underlying aesthetic perception in visual art. Using a diverse dataset of hundreds of thousands of artworks—encompassing both historical masterpieces and AI-generated pieces—we systematically measure observers' aesthetic preferences. Our research aims to develop a computationally interpretable model that elucidates the fundamental principles governing human aesthetic judgment in visual art appreciation.

In an ongoing project, we investigate how humans perceive poetic beauty using Lüshi (律詩), a classical Chinese poetic form. By analyzing hundreds of thousands of poem samples—including both traditional masterpieces and AI-generated compositions—we measure observers’ aesthetic preferences. Our goal is to develop an interpretable cognitive model that reveals the underlying mechanisms of aesthetic judgment in poetry appreciation.

In an ongoing project, we use Texas Hold'em as an experimental framework to investigate how humans infer event probabilities in complex, uncertain scenarios. By generating enumerous card combinations and measuring players' probability judgments, we aim to investigate how human predictions systematically deviate from true statistical likelihoods. Our aim is to develop an interpretable cognitive model of these inferential biases, and to explore gamification as a promising avenue for future cognitive experiments at scale, building on the success of CE.