论文标题
大脑间选择性选择的量化表示
A Quantized Representation of Intertemporal Choice in the Brain
论文作者
论文摘要
值[4] [5]通常是使用连续表示（即实数）建模的。最近假设了价值的离散表示[6]。实验数据也提出并支持大脑中概率的量化表示[7]。价值和概率通过前景理论相互关联[4] [5]。在本文中，我们假设还可以量化跨期的选择。例如，人们可以（或打折）16天对17天无关。为了测试这一点，我们使用2种新型模型分析了一项跨期任务：量化双曲线折现和量化指数折扣。我们在这里的工作是对先前收集的fMRI研究的行为数据的重新检查[8]。使用AIC和BIC测试比较了量化双曲线和量化指数模型。我们发现，13/20参与者最适合量化的指数模型，而其余的7/20最适合量化的双曲线模型。总体而言，15/20参与者最适合具有5位精度的型号（即2^5 = 32步）。总之，无论双曲线或指数级，这些模型的量化版本都比其连续形式更适合实验数据。我们最终概述了我们发现的一些潜在应用。
Value [4][5] is typically modeled using a continuous representation (i.e., a Real number). A discrete representation of value has recently been postulated [6]. A quantized representation of probability in the brain was also posited and supported by experimental data [7]. Value and probability are inter-related via Prospect Theory [4][5]. In this paper, we hypothesize that intertemporal choices may also be quantized. For example, people may treat (or discount) 16 days indifferently to 17 days. To test this, we analyzed an intertemporal task by using 2 novel models: quantized hyperbolic discounting, and quantized exponential discounting. Our work here is a re-examination of the behavioral data previously collected for an fMRI study [8]. Both quantized hyperbolic and quantized exponential models were compared using AIC and BIC tests. We found that 13/20 participants were best fit to the quantized exponential model, while the remaining 7/20 were best fit to the quantized hyperbolic model. Overall, 15/20 participants were best fit to models with a 5-bit precision (i.e., 2^5 = 32 steps). In conclusion, regardless of hyperbolic or exponential, quantized versions of these models are better fit to the experimental data than their continuous forms. We finally outline some potential applications of our findings.
