As refining the MLP beyond what it already does seems no easy task, I thought I would move to the previous problem: To make sure that there is enough temporal diversity in the percepts for the predictor to learn from longer temporal sequences. I’m running a proper test now, but following are a few frames selected from a botched previous test. The percepts are drawn on a black background, and there is no visual difference between perception, mind-wandering or dreaming.
These results show how images could look when they are a combination of regions captured at significantly different times. It certainly cuts against the photo-realism of the individual regions which emphasizes their borders. In looking at these images I realized there is a tension between viewers interest in participating with the image and the requirements for increasing percept diversity. Percepts are diverse because clustering happens much more rarely, only when arousal is very high. Viewers could manipulate the system by forcing large changes in arousal, which would result in the system clustering more often. Determining the threshold for arousal would be difficult when it means balancing rare perception (encouraging diversity) and regular perception (encouraging viewer participation). In framing the work as constructing a landscape, it’s tempting to use a regular interval to trigger perceptual moments during mind-wandering.