AI & Machine Learning

Sequential sampling theory is the most widely used theoretical framework in cognitive neuroscience to describe how people make decisions. According to this theory, a stochastic process in a bounded domain calculates the decision process, and the first passage-time distribution of this process serves as the model's probability function. Generally, the first passage-time distribution in sequential sampling models can not be expressed analytically, which makes fitting these models challenging and computationally demanding. Therefore, it is critical for the fields of mathematical psychology to develop effective and general approximation techniques for the first-passage time distribution of sequential sampling models. In this work, we use a theory-informed neural network approach for simulating and estimating parameters of the partial differential equation (PDE) which is derived from this theory and well known as the Fokker-Planck equation. This methodology does both optimization and approximation simultaneously while a simulation with a continuous solution and optimization parameter combined. As a result, the computing cost is decreased and the numerical approximations' accuracy is raised.

Much research in memory and perception focuses on quantitative outcome variables. Despite the utility of such metrics, they are often insufficient for understanding individual differences in how people encode and retrieve items. To address this gap, we developed a natural language processing approach to analyze narrative responses to questions about the strategies people used in similarity rating and recognition judgment tasks for a novel set of auditory timbre stimuli. We applied topic modeling to 779 responses to three questions about: how people judged similarity between sounds; how people recognized previously heard sounds; and how people formed impressions of the sounds they heard. 20 topics characterized the similarity responses, 16 characterized the recognition judgements, and 30 characterized people’s impressions. Principal components analysis identified latent themes within each topic set. Individual differences in topic prominence were related to recognition memory accuracy and attention to dimensions in similarity ratings. These techniques represent a general methodology for triangulating quantitative, qualitative, and computational methods in memory research.

Abstract reasoning, the ability to solve large-scale problems by taking away unnecessary details (Clement et al., 2007), is essential for human cognition and behavior. However, there remains a lack of cognitive computational models available to study how abstract reasoning emerges and develops in early childhood. We seek to solve this knowledge gap by testing whether deep learning models can explain the key mechanisms that enable children to develop abstract reasoning. Specifically, we investigated whether the Emergent Symbol Binding Network (ESBN; Webb et al., 2021) would be a suitable model. Higher working memory capacity has been shown to facilitate the development of abstract visual reasoning (AVR) in humans. We explore whether ESBN can simulate AVR developmental phenomena by manipulating its memory architecture and training regime. To test this, we observed ESBN’s accuracy as it solved two abstract visual reasoning tasks with decreasing batch size per condition (32, 16, 8, 4). We also used two possible encoders: a random and convolutional encoder. We predicted the convolutional encoder should perform better than the random one, given it has more layers (Seijdel et al., 2020). Initial results do not show support for the ESBN model as a model of abstract visual reasoning development because the simpler, random encoder fared better than the convolutional encoder for all batch sizes. Further research will be performed to identify a suitable candidate model for explaining abstract visual reasoning development.

Survival rates of malignant melanoma improve with early detection, but accurately distinguishing melanoma from an otherwise benign skin lesion is a difficult perceptual task. Rules-based heuristics, such as the ABCD criteria, direct novice observers to judge skin lesions for signs of melanoma according to the distinct perceptual features of shape asymmetry, border irregularity, colour variance, and diameter. Violations of a rule or multiple rules signal the observer to seek expert evaluation and offer a convenient albeit simplified approach to making such complex perceptual judgments. While helpful, the ABCD heuristic contains two primary limitations. First, it assumes an individual's capacity to judge these features independently. Second, research shows expert observers (i.e., dermatologists) judge skin lesions holistically rather than via distinct ABCD features. Deep Convolutional Neural Networks (DCNNs) are powerful machine learning techniques for image processing that are frequently applied to melanoma identification problems and can help reduce observer error. While powerful, these algorithms suffer from their black-box nature, ultimately limiting their utility in human-AI collaborative decision-making contexts. However, the final activation layers of a DCNN provide insight into the features necessary for the final classifier, which can be leveraged to inform human observers. Our project has two aims: (i) to examine if and how novice observers combine the perceptual information dimensions of shape asymmetry (A), border irregularity (B), and colour variance (C) and (ii) to explore the representational space of a DCNN trained for melanoma classification. We collected data across three experiments, each presenting a unique pairwise combination of features to be judged (AB, AC, and BC). We then modelled novice observers' perceptual space using GRT machinery. We found that participants failed to separate perceptual information across all dimensions, indicating that novice observers may rely on an overall 'ugliness' concept to drive their judgments. We then trained a DCNN for melanoma identification and identified the resultant features essential to this classifier. Our methodological approach and preliminary results are presented and discussed.