Bayesian Models of Optical Illusions

ISEF Category: Computational Biology and Bioinformatics

Subcategory: Computational Neuroscience  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

Your brain does not record the world like a camera. It guesses, then checks. Optical illusions expose those guesses in a way you can measure. With web-based experiments, you can test why two people can stare at the same image and report different things.

What Is It?

This project studies how your brain decides what you see when the visual signal gets weak or noisy. In motion-induced blindness, a visible object can seem to vanish when other parts of the scene move. In Troxler fading, a steady image can fade from awareness when you keep your eyes fixed. Both effects happen because your brain does not trust raw input alone. It mixes what reaches the eye with prior beliefs, which are your brain's built-in expectations.

A Bayesian observer model is a math way to describe that process. Think of it like a judge weighing two clues, the current evidence and the brain's prior guess. If the evidence gets weak, the prior matters more. Your project asks which priors best predict who sees the illusion sooner, who sees it longer, and who reports different patterns across trials.

Why This Is a Good Topic

This is a strong science fair topic because you can measure real behavior, model it with clear math, and compare predictions against data. You can connect it to vision science, attention, and perception, all of which matter in neuroscience and human-computer interaction. You do not need a wet lab, but you do need careful experimental design, data cleaning, and model comparison. That gives you a real research feel without needing a university lab.

Research Questions

• How does stimulus contrast affect the time until an object disappears in motion-induced blindness?
• What is the effect of fixation stability on the strength of Troxler fading?
• Does prior exposure to a pattern change the probability that a participant reports an illusion?
• To what extent does a Bayesian observer model predict individual differences in illusion strength?
• Which prior assumptions best fit reaction times and disappearance reports across participants?
• What is the effect of background motion speed on illusion frequency?

Basic Materials

• Laptop or desktop computer with a modern browser.
• Stable internet connection.
• Free jsPsych experiment template or another browser-based experiment builder.
• Consent form text and participant instructions.
• Screen calibration image or built-in brightness settings.
• Spreadsheet software for data logging.
• Digital timer or experiment timing built into the task.
• Headphones or quiet testing space to reduce distraction.

Advanced Materials

• Laptop or desktop computer with a modern browser.
• Free jsPsych framework for building experiments.
• GitHub Pages, OSF, or another free hosting platform for the task.
• R with tidyverse, lme4, and brms for modeling and analysis.
• Python with PsychoPy, NumPy, SciPy, and pandas for simulation and preprocessing.
• ImageJ or Python image tools for stimulus preparation.
• Bayesian inference tools such as Stan or PyMC for model fitting.
• Dedicated response logger or survey platform export for participant data.

Software & Tools

• jsPsych: Builds browser-based perception tasks and records response timing and choices.
• R: Fits Bayesian models, compares priors, and makes plots of individual differences.
• Python: Helps you simulate observer behavior, clean response files, and test model code.
• GitHub Pages: Hosts a free experiment site so participants can access the task in a browser.
• OSF: Stores your protocol, data, and analysis plan in one public research archive.

Experiment Steps

1. Define the illusion outcome you will measure, such as disappearance reports, reaction time, or dominance duration.
2. Choose one stimulus feature to vary first, such as motion speed, contrast, or fixation demand.
3. Build a simple browser task that presents the illusion consistently and records participant responses.
4. Plan control trials that separate true perceptual effects from guessing, fatigue, or screen differences.
5. Fit a Bayesian observer model that links evidence, priors, and response probability.
6. Compare candidate priors and check which model best predicts both the group pattern and individual differences.

Common Pitfalls

• Using screenshots from different devices without checking display settings, which changes contrast and breaks comparison across participants.
• Mixing up attention effects with illusion strength, which makes the model explain the wrong cause.
• Letting participants use very different screen sizes and distances without tracking them, which adds noise to the response data.
• Building a task that records only yes or no reports, which gives too little information to fit priors well.
• Skipping a model comparison plan, which leaves you with one fit that may look good but may not beat simpler alternatives.

What Makes This Competitive

A stronger project will do more than show that illusions happen. You can compare several prior structures, test whether one model predicts individuals better than a simple average model, and report fit quality with cross-validation or another out-of-sample test. You can also ask whether the same model works for both motion-induced blindness and Troxler fading, or whether each illusion needs a different prior. That kind of careful comparison makes your work feel like real computational neuroscience.

Project Variations

• Test whether the same Bayesian model predicts motion-induced blindness and Troxler fading in the same participant set.
• Swap in a different sample group, such as students with heavy screen time versus students with less daily screen exposure.
• Add a model comparison layer that tests whether reaction time, disappearance rate, or confidence best reveals individual priors.

Learn More

• PubMed: Search review articles on visual perception, Bayesian brain models, motion-induced blindness, and Troxler fading.
• NIH PubMed Central: Find full-text neuroscience papers that explain perception experiments and analysis methods.
• MIT OpenCourseWare: Look for free neuroscience, statistics, or probabilistic modeling lecture materials.
• Nature Reviews Neuroscience: Search for review articles on Bayesian perception and visual awareness through your school or public library access.
• OSF: Look for open datasets and preregistered studies on visual perception and psychophysics.