Visualization of the gaming behavior of humans and Ape-X.
<p>To provide a visualization of the action selection by humans and the DQNs, we created a video, which can be found in the GitHub repository at <a href="https://github.com/SHaberland15/Arcade_DQN_Research" target="_blank">https://github.com/SHaberland15/Arcade_DQN_Re...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | , |
| Published: |
2025
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <p>To provide a visualization of the action selection by humans and the DQNs, we created a video, which can be found in the GitHub repository at <a href="https://github.com/SHaberland15/Arcade_DQN_Research" target="_blank">https://github.com/SHaberland15/Arcade_DQN_Research</a>. The video provides a comparison between the actions chosen by the participant and those by Ape-X. The left plot in the video depicts the gaming behavior of a randomly selected participant playing Space Invaders. On the right plot, each bar represents the Q-values for every frame, each associated with one of the six possible types of actions. These values have been preprocessed using a softmax function, enabling them to be interpreted as probabilities. The actions performed by the subject in the current frame, as shown on the left, are highlighted by the purple-colored bar. The action chosen by the DQN for a particular frame, indicated by the maximum Q-value across all types of actions, does not always align with the action chosen by the subject. Therefore, as the second step in our analysis, we introduced a GLM that fits the generated time series of features generated by the DQN to the time series of human actions.</p> <p>(AVI)</p> |
|---|