• Tidak ada hasil yang ditemukan

In Partial Fulfillment of the Requirements for the degree of

N/A
N/A
Protected

Academic year: 2023

Membagikan "In Partial Fulfillment of the Requirements for the degree of "

Copied!
113
0
0

Teks penuh

2] Adachi R, Suzuki S, O'Doherty JP (under review) Neural computations mediating the detection of temporal changes in the human brain. How to make decisions in the face of uncertainty about environmental statistics. These are the questions that played a major role in the formation of the "Neuroscience of Decision" field.

In the second chapter, we reviewed the calculations involved in decision making under uncertainty.

NOMENCLATURE

OVERVIEW

The time course of the dDM-estimated accumulator value correlates with activity in frontoparietal network areas including dmPFC and IPS. Our results confirm the involvement of regions in the frontoparietal network in the process of evidence accumulation in value-based binary choice. Temporal change detection is the ability to detect changes in statistics that govern the timing of events.

The result is interesting in itself because the computer models we tested explain the variance in participants' behavior equally well, regardless of the difference in computer complexity between the models, and also because we cannot distinguish computer models exclusively from behavioral data.

NEURAL CORRELATES OF BINARY CHOICE GUIDED BY VISUAL ATTENTION

In particular, we tested neural correlates of the time course of the accumulator value during trials. It scales with the difference in value between the two items compared to a discount factor on the value of an unattended item (i.e., an off-screen item) that reflects the effect of fixation-driven attention to the attended item. The interpretation of the attention bias parameter 𝛼 is that there is a complete attention bias for the attended item when 𝛼 = 0, ignoring the value of the unattended item in the sample evidence.

In the selection phase, the inputs to the two accumulators prescribed by the anDDM necessitate the neural representation of the value of two items being compared. Second, we examined an observation from our model fitting result that the input to the two accumulators was dominated by the value of the processed item at least at the group level (i.e., 𝛼 = 0). A number of previous studies have reported that the activity in the vmPFC represented the value of the chosen item minus the value of the non-chosen item (Boorman et al., 2009; Jocham et al., 2012).

BOLD correlates of evidence accumulation during value-based decision making

In the choice phase, participants participated in 65 binary choice trials with a free reaction time (RT). Two items were displayed on the screen sequentially until participants indicated their choice by pressing the button. The presented item and its location on the screen (left or right) changed every 1 s.

Choice was activated at the onset of presentation of the second item, and participants could choose one of the two items regardless of the item on the screen at the time of the button press. In the price phase, participants reported their WTP for 130 items used in the choice phase with a full amount between 0 and 3 euros.

INTRODUCTION

MODELING

RESULTS

RESULTS (continued)

SUMMARY

Accumulators that collect evidence supporting the selection of the first-presented item (Item 1) and the second-presented item (Item 2) are shown by A1 and A2, respectively. The sampled proof that serves as the input to the accumulator is the value difference of the two items compared and changes depending on which item is displayed on the screen at the time of accumulation. The selection of one of the two elements is made when the corresponding accumulator value reaches the +1 threshold.

Estimated time course when item2 with value 2 is selected between 3-4 seconds of RT;. the item is selected when it has been attended. The time courses in each category were averaged at each time point of the simulation to give the average time course of the accumulator value corresponding to the selected item (blue), unselected item (green), and the sum of the two (red). Across participants correlation between probability of choosing attended item and attentional bias parameter from model fitting to each participant.

Functional connectivity between each of the two regions involved in the evidence accumulation process (dmPFC and left IPS) and the region representing sample evidence from item value (vmPFC) during the decision period (***: 𝑝 = 4.19 ×108K, **: 𝑝 Wilcoxon signed rank test). Joint distribution of choice probability and RT for each of the 16 possible value pairs. The choice of item 1 and item 2 are depicted in the positive and negative domain of the y-axis, respectively.

Here, we provide the parameter grid information used in each iteration and the resulting best-fit parameter values. The significance of the cluster in the left IPS representing the time course of the accumulator value was tested by small volume correction (SVC) on a 10 mm sphere surrounding the mean coordinates of peak activation from previous studies (x=-34, y =-55, z= 44).

Figure 2. Computational model.
Figure 2. Computational model.

COMPUTATIONAL ACCOUNT OF

TEMPORAL CHANGE DETECTION IN THE HUMAN BEHAVIOR

TP is a count of the number of times a participant's button press was within a 5-s window after an objective change point (task-specific) occurred. FN is a count of the number of times a participant's button was not pressed within a 5-second time window. One was an implementation of the Bayesian online change point detection model (BOCPD, Adams, & MacKay, 2007) and another was an implementation of the dynamic belief model (DBM, Yu, & Cohen, 2009).

A change point is detected if 𝐷𝑉q≥ 𝜃q and the value of z is updated to the location of the change point (i.e. 𝑧 = 𝑛). We calibrated each of the three computational models to predict the timing of button presses in participants' behavior. Finally, we compared and selected the best submodel for each of the three models.

Histogram of the average number of button presses predicted by each model, compared to the participant's timing of button presses. Comparison of the number of button presses per trial of actual participants vs. For each of the 72 trials in the main task, we generated temporal change points (see Results) and counted the number and recorded the timing of the change points.

Open and filled circles indicate the two sets of sequences used in the main task of the experiment. For each of the 72 trials in the main task, we generated temporal change points (see STAR Methods) and counted the number and recorded the timing of change points. Overall, the number of button presses for participants was comparable to that of the optimal Bayesian agent (right).

Comparison of the performance of the delta rule model against the three models analyzed in the main text (See Figure 7C).

Figure 6. Illustration of task structure.
Figure 6. Illustration of task structure.

NEURAL COMPUTATIONS MEDIATING

TEMPORAL CHANGE DETECTION IN THE HUMAN BRAIN

In the tables, the reported coordinates are in MNI coordinates of the peak voxel in each of the activation clusters. We tested for brain regions that represented the time course of the decision variables derived from each of the BOCPD, DBM, and TAM models. We found that the time course of the decision variable of TAM showed a significant positive correlation with activity in the frontoparietal network of the right dlPFC and bilateral IPS (Figure 9, p<0.05 whole brain corrected at the cluster level; see Table S8 for other activated regions).

Plot of the effects of button pressing in the main task and the control task in the right dlPFC ROI, identified by cross-validation procedure excluding one subject. Neural correlates of the time course of the decision variables derived from each of the three computational models for temporal change detection. Activity in the right dlPFC and bilateral IPS showed a significant correlation with the model-predicted time course of the TAM decision variable (p<0.05 whole brain corrected at the cluster level; see Table S8 for other activated areas).

The model-predicted time course of the decision variable from BOCPD or DBM showed no significant positive correlations with either brain region (see Table S8 for activated regions in other model-related contrasts). Brain regions that uniquely correlated with the time course of the decision variables derived from TAM. Activity in the right dlPFC and right IPL comprising the IPS showed a significant correlation with the model-predicted time course of the decision variable from TAM when the decision variable from the other two models was included as a confounder (p<0.05 whole-brain cluster-corrected; see Table S9 for other activated regions ).

We show the time course of the decision variable derived from each model for two example trials in one participant. We report the minimum, 25th percentile, median, 75th percentile, and maximum of the decision variable for each of the three computational models.

Figure 8. Brain areas showing greater activity associated with button presses in the main task  than in the control task
Figure 8. Brain areas showing greater activity associated with button presses in the main task than in the control task

BIBLIOGRAPHY

Daw ND, O'Doherty JP, Dayan P, Dolan RJ, Seymour B (2006) Cortical substrates for exploratory decisions in humans. Gläscher J, Hampton AN, O'Doherty JP (2009a) Determining a role for the ventromedial prefrontal cortex in encoding action-based value signals during reward-related decision making. Hampton AN, Bossaerts P, O'Doherty JP (2006) The role of the ventromedial prefrontal cortex in abstract state-based inference during decision making in humans.

Hampton AN, Bossaerts P, O'Doherty JP (2008) Neural correlates of mentalizing computations during strategic interactions in humans. Heekeren HR, Marrett S, Bandettini PA, Ungerleider LG (2004) A general mechanism for perceptual decision making in the human brain. Larsen T, O'Doherty JP (2014) Uncovering the spatio-temporal dynamics of value-based decision making in the human brain: a combined fMRI-EEG study.

Lee SW, Shimojo S, O'Doherty JP (2014) Neural computations underlying the arbitrage between model-based and model-free learning. Lopez-Persem A, Domenech P, Pessiglione M (2016) How prior preferences determine decision frames and biases in the human brain. McNamee D, Rangel A, O'Doherty JP (2013) Category-dependent and category-independent goal-value codes in human ventromedial prefrontal cortex.

Payzan-LeNestour E, Dunne S, Bossaerts P, O'Doherty JP (2013) The neural representation of unexpected uncertainty during value-based decision making. Polanía R, Krajbich I, Grueschow M, Ruff CC (2014) Neural oscillations and synchronization support differential evidence accumulation in perceptual and value-based decision making.

Gambar

Figure 2. Computational model.
Figure 3. Behavioral results.
Figure 4. Fixation-driven attentional effect on the stimulus value representation in vmPFC
Figure 5. Neural correlates of the time course of accumulator value and PPI result.
+7

Referensi

Dokumen terkait

Candidate Linguistics and Applied Linguistics, College of Chinese, Wuhan University, Hubei, P.R.China MTCSOL Teaching Chinese to Speakers of Other Languages, Yunnan Normal University,