The virtual loss function in the summary perception of motion and its limited adjustability

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Visual Perception

Background:

• Humans efficiently perceive average features in visual ensembles.
• The precise mathematical form of this summary perception and its modifiability by feedback remain unclear.
• A loss function framework can model how summary statistics relate to ensemble feature distributions.

Purpose of the Study:

• To infer the implicit loss function governing human ensemble perception.
• To investigate whether this inferred loss function can be altered through feedback.
• To characterize the summary statistic perception in a random-dot motion task.

Main Methods:

• Two experiments using a random-dot motion estimation task with skewed direction distributions.
• Participants estimated the average motion direction of dot ensembles.
• Experiment 1: No feedback; Experiment 2: Feedback provided (mean or mode as correct).

Main Results:

• Estimates in Experiment 1 fell between the mean and mode, closer to the mean, modeled by an inverse Gaussian loss function.
• The deviation towards the mode increased linearly with the mode-to-mean distance.
• Feedback in Experiment 2 showed minimal influence on participants' estimates, failing to specifically alter the loss function.

Conclusions:

• The virtual loss function for summary perception of motion is approximately inverse Gaussian.
• This loss function demonstrates significant resistance to modification by explicit feedback.
• Findings suggest a stable underlying mechanism for ensemble motion perception.