Xiaoxiao-ideas

[xiaoxl3]

Performance accuracy in relation to reaction time. a. Measurement of Angles: Measure the angles between the selected arrow by subjects and the prior motion showed on screen (using the estimated x/y coordinates). We can set up a distribution of the deviation angles from the real orientation to represent the performance index or accuracy. Categorize the deviation angles into specific ranges, such as "0-5 degrees" and "5-10 degrees."

We can consider the following interpretations:

1. Reaction Time as an Indicator of Attention:

• Reaction time could represent the level of attention. By comparing reaction time with deviation angles, we can determine if shorter reaction times predict higher performance accuracy.
• Decoding Performance Accuracy: Utilize reaction time data to decode and understand performance accuracy. It tell us whether quicker responses are generally associated with more precise performance.

2. Learning improvement over study session---- Models: reinforcement learning: sooner better, or slower

   • Improvement Within and Across Sessions: Examine whether the correlation improves within a single session or over multiple days for the same prior settings. We can test whether performance accuracy and reaction time correlation strengthens with repeated practice or if it diminishes by the end of the day due to fatigue. It will tell us if learning and performance accuracy are progressively enhanced through continuous practice or over time.

3. Learning Strategy Adjustment:

   • Response to Novel Stimuli: The adaptation process might occur at the beginning of the training session or at the start of each exposure to new or unfamiliar conditions. we can assess whether response times extend longer when adjusting to novel stimuli, such as a new orientation or reduced sensory inputs.
   • Impact of Learning Progress: Determine if this adjustment period diminishes with learning progress, as the subject becomes more familiar with the rules or gets used to the changes.

[lucasbaldezzari]

Xiaoxiao! Nice test :D

[xiaoxl3]

Reinforcement Learning: An Approach to Decision Making

1. Split the Group into Fast and Slow Learners by Average Response Time and Accuracy (Angle-Diff):

• We can use the first 200 trials of data or the data from the first day.
• Issue: We might not have enough data in one group, potentially resulting in more slow learners and fewer fast learners. The criteria for splitting fast and slow learners are tricky.

2. Compare Performance: Determine whether fast learners are less sensitive to environmental changes (prior or coherence). Assess which model they are likely to use, the BBO or bimodal.

• We can split subjects into two groups (fast vs. slow) according to previous analysis.
• Draw the performance accuracy distribution along the prior and coherence.

[lucasbaldezzari]

Add this plot for each subject.

plt.figure(figsize=(15,6)) sns.lineplot(grups1, x = grups1.index ,y="diff_angles", hue = "prior_std") plt.grid() plt.title(f"Subjetc {subject_id}") plt.show()

[lucasbaldezzari]

Updating

@xiaoxl3 updated in her document here.

1. Plotting Temporal Learning Patterns:

• Sequential trials will be plotted to observe the learning curve over time.
• Analysis will focus on the first 10-20 trials in each block as a critical window for learning.

2. Analysis of Prior Information:

• Trials will be split into segments (e.g., 0-10, 10-20, 20-30, ..., 90-100 trials).
• Differences in response angles will be compared between these segments and the last 10 or 20 trials.
• It is hypothesized that high standard deviation (STD) priors will require more trials to learn, shifting the learning curve to the right. Conversely, low STD priors will result in faster learning.

3. Analysis of Sensory Coherence:

• Data will be reorganized by coherence levels.
• High coherence is expected to result in fast learning and high performance accuracy, with low differences in response angles.
• Low coherence is hypothesized to lead to slower learning, more mistakes, and potentially system crashes (inability to make a choice within 5 seconds, resulting in NaN).