Virtual ICCM Fast Talks

Several researchers have postulated process (Wolfe, 2021) and neural (Grossberg, 1994) computational models of visual attention and perception; however, core tenants of these frameworks are seldom integrated into human-inspired algorithms of computer vision. Here, we illustrate a novel computer vision framework that simulates guided visual attention and processing in humans using a biologically-plausible neurocognitive architecture (Leabra; O’Reilly et al., 2017). We provide an initial demonstration of its efficacy in simulating realistic human visual processing in noisy and degraded visual environments and argue for its potential as a simulation of human visual search with cognitive and biological plausibility.

ACT-R (Adaptive Control of Thought - Rational) is a cognitive architecture that provides a framework for developing computational cognitive models that simulate various aspects of human performance in tasks ranging from time perception to decision-making. ACT-R also can be implemented in various programming languages and environments, including graphical user interfaces (GUIs) that facilitate model development, experimentation, and analysis (Anderson 2007, p.135). One of these interfaces is ACT-R graphical user interface (AGI). The compiling time of the AGI can vary depending on several factors related to the design, development environment, and computational resources available (Bothell, 2007). The compilation process for these technologies involves translating the source code into executable binaries or bytecode that can be run on different operating systems with different compiling times (Şahin & Duman, 2023). The fact that AGI’s timing accuracy is very important to obtain highly accurate results, especially in time perception experiments. In this study, a function was developed to enable the AGI interval timing experiment, developed using the Python programming language, to run as close to real time as possible. The function calculates the duration of other functions which are based on the ACT-R model and called to create the user interface during the experiment, then performs the necessary mathematical operations to ensure that this duration is as close as possible to the required trial duration (13s). In the experiment, there is a letter condition with a letter recognition task and an addition condition involving the sum of two numbers. In the experiments conducted before using the function, it was observed that the letter condition, which should take 13s, took an average of 15.67s and the addition condition took 15.18s. According to the results obtained after this function was added, it was observed that the letter condition took 14.11s on average and the addition condition took 13.3s on average. With the inclusion of the function, the trial time was found to be closer to the original time of 13s. As a result, the study provides an improvement in the use of AGI as a graphical user interface in experiments with humans. It also shows that functions related to timing accuracy can be added in Python-based experiment design.

Adapting to dynamic environments poses significant challenges for humans, even in seemingly simple scenarios. Researchers have explored different directions to address this issue, including the use of cognitive models to predict human adaptive capabilities. This research investigates the effectiveness of an intervention and the role that an Instance-Based Learning (IBL) cognitive model could play in facilitating adaptation to changing conditions. We conducted an experiment involving a binary choice task, manipulating the presence of an intervention and the direction of change in outcome payoffs: either increasing (where one option improves over time) or decreasing (where one option deteriorates over time). Our findings reveal that: a) the intervention appears effective primarily in increasing conditions, b) adaptation is better in decreasing conditions, c) the IBL model outperforms human participants in adaptation, and d) the model exhibits greater accuracy in predicting humans' choices in increasing rather than decreasing conditions. These results are discussed in the context of experiential decision-making literature and the potential of using IBL models for intervention to improve human adaptation.

Collaborative learning is when learners reconstruct one's knowledge based on others’ knowledge, and then they gain understanding. However, it is indicated that working memory inhibits the cognitive process learners acquire and use other's knowledge. Additionally, it is challenging to manipulate learners' working memory and capture the cognitive process during collaborative learning in psychological experiments. Therefore, this study investigated how working memory influenced the search for knowledge reconstruction in a psychological experiment and further examined the nature of the cognitive process using ACT-R, a cognitive architecture and theory for human cognition. Both laboratory experiments and simulations revealed a positive correlation between working memory and correct. Also, those revealed no correlation between working memory and incorrect. In model-based simulations, we found that reconsidering based on others’ knowledge was effective when working memory was high. This study contributed to developing pedagogical agents as collaboration members and teachable agents to support collaborative learning.

Algebraic reasoning, particularly concerning literal symbols, poses significant challenges for learners and educators alike. This study investigates the potential of eye-tracking technology to enhance understanding and instructional approaches in algebraic reasoning research. Through two experimental sessions involving students aged 9-11, eye movements, and fixations were analyzed while engaging with algebraic tasks. Results reveal distinct patterns in cognitive processing, highlighting the utility of heatmaps, and eye movement videos in elucidating cognitive load and areas of difficulty. These insights inform the development of targeted instructional interventions to support learners in navigating algebraic concepts. While promising, the study acknowledges limitations in sample size and environmental control, emphasizing the need for further research. Overall, eye-tracking technology shows promise as a transformative tool in algebraic reasoning research, offering valuable insights into students' cognitive processes and informing effective pedagogical strategies tailored to the challenges posed by literal symbols.

Humans have cognitive vulnerabilities that can be leveraged to influence individuals. One such vulnerability is the continued influence effect (CIE), where misleading information can have a lasting effect even after corrections or factual discrediting information is presented. The CIE has been addressed experimentally and memory-based explanations exist. However, no current cognitive models specify cognitive mechanisms for prediction, simulation, and detailed testing of hypotheses. Here, we discuss relevant literature and propose a novel cognitive model to investigate memory mechanisms underlying the CIE. We demonstrate the utility of the model using simulations which show how the CIE emerges from memory processes and discuss plans for future research.

In recent years, the spread of fake news in social media has heightened concerns over widening divisions in public opinion. Localized echo chambers within closed communities exacerbate polarization, presenting significant challenges to trust in digital information. This research utilizes simulations of dilemma situation to explore reliable online communication. By analyzing communication patterns and the spread of fake news across virtual and physical domains, it aims to uncover strategies for mitigation. Research on communication in a two-layered environment includes the analysis of communication dynamics in dilemma situation (Inoue et al., 2022). This study employs a dilemma game task, which adds a dilemma element to collaborative task in a grid-world constructed by Konno et al. (2013) to observe the formation process of communication systems. Building upon this environment, our research expands the scope by incorporating Instance-Based Learning Theory (Simon & Langley, 1981). The simulation entails augmenting agent populations and introducing localized clustering mechanisms to emulate the emergence of echo chambers. Additionally, a novel operational definition of fake news, rooted in game outcomes, is introduced to quantify its diffusion, and unravel the dynamics of group interactions. Through this approach, the study seeks to clarify the mechanisms underlying echo chamber formation and fake news diffusion, thereby offering valuable insights into rebuilding trust in online communication. As the research progresses, further refinement of simulation methodologies and the integration of real-world social media datasets are anticipated, aiming to provide more nuanced analyses and actionable recommendations for addressing these pressing societal challenges.

The tendencies of cognitive errors are considered to be influenced by various factors such as individual attributes and emotional states, which interact and fluctuate. This study examines the interaction of internal factors related to memory errors by appropriately estimating multiple parameters of a cognitive model. Instances of memory errors in a 10-digit span memory task were collected through crowdsourcing. 50 participants were recruited, and they performed the task for 30 trials. After the task, they responded to questionnaires regarding personal attributes and emotional evaluations using the Japanese version of PANAS (Kawahito et al, 2011) during task execution. The cognitive architecture ACT-R was adopted to create the cognitive model. Using this architecture, we created a model to recall 10-digit sequences and searched for ACT-R parameters that matched individual error tendencies. In this study, We adjusted two parameters within the ACT-R model: mismatch penalty (mp) controlling the weight of similarity between chunks to be recalled, and retrieval threshold (rt), controlling the activation threshold of chunks. Then, we investigated correlations between the most fitting parameters for individuals and their attributes and emotional states. In our previous presentation at this conference (Shimbori et al., 2023), the range of parameters to be explored was arbitrarily selected. However, in this study, we adjusted the range to fit the entire experimental data by exploring the median and range of each parameter. As a result, several correlations between model parameters and emotional states were observed, suggesting a potential correspondence between memory error tendencies and internal states.