Yang Tiancheng, Xie Chenghan, Wang Xuehe
School of Artificial Intelligence, Sun Yat-sen University, Zhuhai, China.
College of Engineering, Department of Information Engineering, The Chinese University of Hong Kong, Hong Kong, China.
Front Psychol. 2025 Jun 5;16:1510151. doi: 10.3389/fpsyg.2025.1510151. eCollection 2025.
The Iowa Gambling Task (IGT) is a widely used paradigm for evaluating decision-making and executive functioning, yet existing computational models seldom account for the phenomenon of forgetting, which is critical to understanding dynamic decision processes.
We developed the Exploitation and Exploration with Forgetting (EEF) model, which integrates a dynamic forgetting parameter (λ) and participants' first-choice priors into a unified reinforcement-learning framework. The EEF model was fitted to choice data from 504 healthy individuals performing the standard 100-trial IGT. Model performance was assessed via goodness-of-fit comparisons (BIC/AIC/Free Energy), parameter- and model-recovery simulations, and behavioral validation.
Across multiple cohorts, the EEF model achieved superior fit relative to five established models. We introduce two novel metrics-Sequential Exploration Decay (SED) and Forgetting Interval (FI)-to quantify how forgetting shapes exploratory behavior. The EEF model's SED and FI values closely matched empirical data, and further analyses revealed systematic effects of age and gambling frequency on forgetting and decision strategies.
Our findings underscore the fundamental role of forgetting in complex decision-making environments. By explicitly modeling information decay, the EEF framework offers novel insights into cognitive dynamics across the lifespan and behavioral contexts, and provides a parsimonious yet powerful tool for future computational and empirical research.
爱荷华赌博任务（IGT）是一种广泛用于评估决策和执行功能的范式，但现有的计算模型很少考虑遗忘现象，而遗忘对于理解动态决策过程至关重要。
我们开发了带遗忘的利用与探索（EEF）模型，该模型将动态遗忘参数（λ）和参与者的首选先验整合到一个统一的强化学习框架中。EEF模型被拟合到504名执行标准100次试验IGT的健康个体的选择数据上。通过拟合优度比较（BIC/AIC/自由能）、参数和模型恢复模拟以及行为验证来评估模型性能。
在多个队列中，EEF模型相对于五个已建立的模型实现了更好的拟合。我们引入了两个新指标——顺序探索衰减（SED）和遗忘间隔（FI）——来量化遗忘如何塑造探索行为。EEF模型的SED和FI值与实证数据紧密匹配，进一步分析揭示了年龄和赌博频率对遗忘和决策策略的系统性影响。
我们的研究结果强调了遗忘在复杂决策环境中的基本作用。通过明确模拟信息衰减，EEF框架为整个生命周期和行为背景下的认知动态提供了新的见解，并为未来的计算和实证研究提供了一个简洁而强大的工具。
