Figure 4 - uploaded by René Schlegelmilch
Content may be subject to copyright.
Graphical Bayesian model of the generalized context model (GCM) with separate similarity gradients µ c qr for each set r of exemplars (indexed according to set in Tables 1 and 2). Prior definitions are given for Study 3. The shaded box highlights the tested difference δ c r of the estimated similarity gradients µ c qr on each set r between the reward condition and the corresponding baseline. Circles and squares represent continuous and nominal variables, respectively. The model is embedded in a model selection, and a common cause mechanism predicts both reported feature weights w k and the total number of Tami responses y ik (gray symbols) for each item i in the transfer phase; see text).
Source publication
Reward magnitude is a central concept in most theories of preferential decision making and learning. However, it is unknown whether variable rewards also influence cognitive processes when learning how to make accurate decisions (e.g., sorting healthy and unhealthy food differing in appeal). To test this, we conducted 3 studies. Participants learne...
Contexts in source publication
Context 1
... test which mechanism describes the influence of reward magnitude in exemplar-based categorization decisions, we applied two separate versions of the GCM (i.e., a memory and a similarity version; Nosofsky, 2011) to all decisions from the transfer phase in each study and condition, describing the influence of reward magnitude either by variations on item-specific memory or by the similarity gradient, respectively. 6 Figure 4 shows the graphical model of the similarity version of the GCM, with prior definitions for Study 3. We tested the two GCMs against each other in a hierarchical Bayesian latent mixture framework (see M. D. Lee & Wagenmakers, 2014; p. 212, for an introduction). ...
Context 2
... this, we hierarchically estimated the corresponding model hyperparameters (i.e., population means and standard deviations) for the manipulated item sets r (indexed according to Tables 1 and 2) in each condition (q). Specifically, in the memory GCM, we estimated In the similarity GCM (Figure 4; without fixed values), we hierarchically sampled the exemplar gradients c kj from the estimated population distribution with mean µ c qr and standard deviation σ c qr (see Appendix C for details). As outlined, we would expect the differences between the similarity estimates in the q = baseline condition to indicate narrower generalization than those in the q = reward condition(s) for r = high-reward exemplars (reflected by δ c r > 0 in Figure 4), but to remain unchanged for r = low-reward exemplars. ...
Context 3
... in the memory GCM, we estimated In the similarity GCM (Figure 4; without fixed values), we hierarchically sampled the exemplar gradients c kj from the estimated population distribution with mean µ c qr and standard deviation σ c qr (see Appendix C for details). As outlined, we would expect the differences between the similarity estimates in the q = baseline condition to indicate narrower generalization than those in the q = reward condition(s) for r = high-reward exemplars (reflected by δ c r > 0 in Figure 4), but to remain unchanged for r = low-reward exemplars. ...
Context 4
... because exemplar-specific parameters substantially increase the flexibility of the GCM in all cases, we also included a common-cause constraint, to improve model convergence and theoretical plausibility. In particular, each of the cognitive models implements feature attention/weighting parameters (e.g., ω k in Figure 4), which are essential for making predictions. Thus, the models' performance evaluations were based on both the categorizations and the subjective feature-attention measures (w k in Figure 4). ...
Context 5
... particular, each of the cognitive models implements feature attention/weighting parameters (e.g., ω k in Figure 4), which are essential for making predictions. Thus, the models' performance evaluations were based on both the categorizations and the subjective feature-attention measures (w k in Figure 4). For the CAM, in which the dimension weights were not constrained to being positive and sum to 1 (as in the GCM), we normalized the estimated absolute weights for fitting 7 By definition of the GCM, a model that fixes memory strength for high-reward items and estimates memory strength for low-reward items is formally identical to the reported model. ...
Context 6
... test the hypothesis that a higher reward alters exemplar similarity gradients (i.e., δ c r = 0) we first derived the distribution for the null hypothesis by calculating the differences between two half-Cauchy priors as defined for µ c qr (see also Figure 4). To approximate the likelihood at zero difference we took the posterior density in the [-0.1, 0.1] interval for the null distribution, as well as for the corresponding parameter posteriors for each item set. ...
Context 7
... approximate the likelihood at zero difference we took the posterior density in the [-0.1, 0.1] interval for the null distribution, as well as for the corresponding parameter posteriors for each item set. Please note, due to the MCMC model selection procedure, the participant group assignments determine which participants contribute (and how much) to the estimation of these posteriors (i.e., the GCM posteriors only update for participants k for which z k = 2, see Figure 4). Note. ...
Context 8
... Posterior means for exemplar-similarity differences (δ c r in Figure 4) between the baseline and each reward condition (see sets in Tables 1 and 2) We then calculated the BFs (SD density ratios; Dickey, 1971), which are presented together with the 95% CIs in Table 6. As can be seen, except for the TR condition in Study 3, the similarity gradients for high-reward exemplars were confidently smaller than the corresponding baseline estimates (CIs > 0, and BFs > 6.9). ...
Similar publications
In last few decades, reading the human mind is an innovative topic in scientific research. Recent studies in neuroscience indicate that it is possible to decode the signals of the human brain based on the neuroimaging data. The work in this paper explores the possibility of building an end-to-end BCI system to learn and visualize the brain thoughts...
![Figure 6. Boost schematic model. Source: De Almeida et al. (2022) [25].](publication/375769686/figure/fig1/AS:11431281206204600@1700583323391/Boost-schematic-model-Source-De-Almeida-et-al-2022-25_Q320.jpg)
Neurogenic bladder is a dysfunction in the lower urinary tract due to damage to the nervous system. One of the treatments that has shown important results is transcutaneous neuromodulation. The neuromodulation equipment available on the market does not allow remote activation or offer many resources for adjusting the parameters of the generated sti...
A rapid development in personal electronics has raised challenging requirements for portable and sustainable power sources. For example, in wearable technologies, the concept of wearable body area network brings body motion and vital signs monitoring together in synergy. For this, a key aspect is sustainable portable energy, available anywhere, at...
Introduction: Interactive electronic games allow to access virtual environments and interact using a computer or TV screen. Anyone who has played a video game, or seen others playing, is aware of the importance of reaction speed and eye-hand coordination skills.
Objective: To determine the differences in psychomotor performance between professiona...
Transcranial temporal interfering stimulation (tTIS) can focally stimulate deep parts of the brain related to specific functions using beats at two high frequencies that do not individually affect the human brain. However, the complexity and nonlinearity of the simulation limit it in terms of calculation time and optimization precision. We propose...
Citations
... Stimulus generalization indicates that individuals non-randomly react to a novel stimulus after observing the consequences of a similar one (Schlegelmilch and von Helversen, 2020;Shepard, 1987). For board chairs that experienced SARS-induced operating distress, other similarly negative shocks (such as infectious diseases, or even unrelated shocks that can trigger operating distress) can lead to more conservative investment strategies. ...
This study performs empirical tests to show several behavioral patterns in managerial decision-making using Chinese data. Based on the outbreak of Severe Acute Respiratory Syndrome (SARS) in 2003, caused by the same family of viruses as COVID-19, we show that firms with board chairs that experienced SARS during their tenure as senior executives invested less in subsequent periods. This effect is evident for board chairs that personally experienced the pandemic and those who prominently perceived their neighbor's operating distress during the SARS outbreak. The impact persists and is amplified by other negative shocks, such as infectious diseases and economic uncertainty, which supports the validity of stimulus generalization theory in corporate finance. Firms with board chairs that experienced SARS invested even less during the COVID-19 pandemic but performed relatively better during this same period than during normal business operations. However, the benefit was somewhat smaller than the overall loss. This study supports the finding that the long-lasting impact of pandemic-induced operating distress can distort the real economy. Our findings have meaningful implications for assessing the long-term economic effects of COVID-19.
... Reward expectations while perceiving direct gaze may induce this modulation bias for behaviours with relatively high reward probabilities. Cue-reward associations can be generalized (Miendlarzewska, Bavelier, & Schwartz, 2016;Schlegelmilch & von Helversen, 2020). Reward expectations in response to perceptual cues encoded by the hippocampus are generalized, and reward-related brain activations are observed with generalized cues (Aberg, Doell, & Schwartz, 2015;Gerraty, Davidow, Wimmer, Kahn, & Shohamy, 2014). ...
Perceiving direct gaze facilitates social cognition and behaviour. We hypothesized that direct gaze modulates decision-making, particularly calculations of action values. To test our hypothesis, we used the reinforcement learning paradigm in situations with or without direct gaze. Forty adults were recruited and participated in pupil size measurements and a two-armed bandit task. The task was conducted with 70% and 30% reward probabilities for each option. During the task, a female showing the Direct Gaze (DG) or Closed Eyes (CE) condition was presented from the start of each trial. The results showed that behavioural bias to choices with 70% reward probability increased more in the DG condition than in the CE condition and the expected reward value. This bias to choices with 70% reward in the DG condition was predicted by pupil dilation to DG. These results suggest that participants over-evaluated the expected reward value in the DG condition, and this DG effect may be related to subjective expectations of rewarding events indexed by pupil dilations. It is considered that perceiving direct gaze is a driver of reward expectations that modulate action value calculations and then cognitive processing and behaviours are facilitated.
... This theoretically commits CAL to the idea that abstraction is mainly driven by the rule-learning network, and strong memorization is more akin to stimulus identification. In other words, in exemplar models (e.g., GCM Nosofsky, 1986), if the memory strength parameter of an exemplar becomes stronger, an increase in its recall accuracy is predicted, while a decrease in accuracy for exemplars from other categories is also predicted (see also Hendrickson, Perfors, Navarro, & Ransom, 2019;Homa et al., 2019;Schlegelmilch & von Helversen, 2020), similar to a recall bias. In CAL, increasing the memory strength of a stored instance increases its recall accuracy and decreases its interfering influence on category inferences for dissimilar instances. ...
We introduce the CAL model (Category Abstraction Learning), a cognitive framework formally describing category learning built on similarity-based generalization, dissimilarity-based abstraction, two attention learning mechanisms, error-driven knowledge structuring, and stimulus memorization. Our hypotheses draw on an array of empirical and theoretical insights connecting reinforcement and category learning. The key novelty of the model is its explanation of how rules are learned from scratch based on three central assumptions. (1) Category rules emerge from two processes of stimulus generalization (similarity) and its direct inverse (category contrast) on independent dimensions. (2) Two attention mechanisms guide learning by focusing on rules, or on the contexts in which they produce errors. (3) Knowing about these contexts inhibits executing the rule, without correcting it, and consequently leads to applying partial rules in different situations. The model is designed to capture both systematic and individual differences in a broad range of learning paradigms. We illustrate the model’s explanatory scope by simulating several benchmarks, including the classic Six Problems, the 5--4 problem, and linear separability. Beyond the common approach of predicting average response probabilities, we also propose explanations for more recently studied phenomena that challenge existing learning accounts, regarding task instructions, individual differences in rule extrapolation in three different tasks, individual attention shifts to stimulus features during learning, and other phenomena. We discuss CAL's relation to different models, and its potential to measure the cognitive processes regarding attention, abstraction, error detection, and memorization from multiple psychological perspectives.
It is known that humans and animals can learn and utilize category information quickly and efficiently to adapt to changing environments, and several brain areas are involved in learning and encoding category information. However, it is unclear that how the brain system learns and forms categorical representations from the view of neural circuits. In order to investigate this issue from the network level, we combine a recurrent neural network with reinforcement learning to construct a deep reinforcement learning model to demonstrate how the category is learned and represented in the network. The model consists of a policy network and a value network. The policy network is responsible for updating the policy to choose actions, while the value network is responsible for evaluating the action to predict rewards. The agent learns dynamically through the information interaction between the policy network and the value network. This model was trained to learn six stimulus-stimulus associative chains in a sequential paired-association task that was learned by the monkey. The simulated results demonstrated that our model was able to learn the stimulus-stimulus associative chains, and successfully reproduced the similar behavior of the monkey performing the same task. Two types of neurons were found in this model: one type primarily encoded identity information about individual stimuli; the other type mainly encoded category information of associated stimuli in one chain. The two types of activity-patterns were also observed in the primate prefrontal cortex after the monkey learned the same task. Furthermore, the ability of these two types of neurons to encode stimulus or category information was enhanced during this model was learning the task. Our results suggest that the neurons in the recurrent neural network have the ability to form categorical representations through deep reinforcement learning during learning stimulus-stimulus associations. It might provide a new approach for understanding neuronal mechanisms underlying how the prefrontal cortex learns and encodes category information.
Based on theoretical and empirical considerations, Bröder et al. (2017) proposed the RulEx-J model to quantify the relative contribution of rule- and exemplar-based processes in numerical judgments. In their original article, a least-squares (LS) optimization procedure was used to estimate the model parameters. Despite general evidence for the validity of the model, the authors suggested that a strong bias in favoring the rule module could arise when there is noise in the data. In this article, we present a hierarchical Bayesian implementation of the RulEx-J model with the goal to rectify this problem. In a series of simulation studies, we demonstrate the ability of the hierarchical Bayesian RulEx-J model to recover parameters accurately and to be more robust against noise in the data, compared to an LS estimation routine. One further advantage of the hierarchical Bayesian approach is the direct implementation of hypotheses about group differences in the model structure. A validation experiment as well as reanalyses of two experiments from different labs demonstrate the usefulness of the approach for testing hypotheses about processing differences. Further applications for judgment research are discussed.
A complete understanding of decision-making in military domains requires gathering insights from several fields of study. To make the task tractable, here we consider a specific example of short-term tactical decisions under uncertainty made by the military at sea. Through this lens, we sketch out relevant literature from three psychological tasks each underpinned by decision-making processes: categorisation, communication and choice. From the literature, we note two general cognitive tendencies that emerge across all three stages: the effect of cognitive load and individual differences. Drawing on these tendencies, we recommend strategies, tools and future research that could improve performance in military domains – but, by extension, would also generalise to other high-stakes contexts. In so doing, we show the extent to which domain general properties of high order cognition are sufficient in explaining behaviours in domain specific contexts.
