neurolang-experiment

Human Behavioral Experiment

Phosphene pattern learning experiment comparing Symbol vs Pixel encoding.

Files

human_experiment/
├── generate_stimuli.py   # Export phosphene images from trained encoders
├── experiment.html       # jsPsych experiment (open in browser)
├── analyze_results.py    # Statistical analysis of collected data
├── stimuli/              # Generated phosphene images (after running generate_stimuli.py)
│   ├── symbol/           # Symbol-encoded patterns
│   ├── pixel/            # Pixel-encoded patterns
│   └── manifest.json     # Stimulus metadata
├── data/                 # Raw CSV data from participants
└── figures/              # Analysis output figures

Step 1: Generate Stimuli

Run on the server where trained models exist:

cd human_experiment
python3 generate_stimuli.py

This exports 200 phosphene images per condition (8 classes x 25 samples).

Step 2: Deploy Experiment

Option A - Local (simplest):

cd human_experiment
python3 -m http.server 8000
# Open browser: http://localhost:8000/experiment.html?condition=symbol&pid=P001

Option B - GitHub Pages (for remote participants):

1. Push this folder to a GitHub repo
2. Enable GitHub Pages in repo settings
3. Share the URL with participants

Step 3: Run Participants

Each participant gets ONE link (between-subject design):

• Symbol group: experiment.html?condition=symbol&pid=P001
• Pixel group: experiment.html?condition=pixel&pid=P002

Assign participants randomly. Use odd PIDs for symbol, even for pixel (or any scheme).

Aim for 10-20 per group (20-40 total). Each session takes ~25 minutes.

After completion, participants download a CSV file automatically. Collect these files.

Step 4: Analyze

Put all CSV files in data/, then:

python analyze_results.py --data-dir data/

Outputs:

• figures/learning_curves.png - Learning curves with 95% CI
• figures/test_accuracy.png - Box plot with t-test
• figures/stats.json - Full statistical results

Attention Checks

The experiment includes:

• Practice trials (5) to ensure interface understanding
• Progress feedback every 50 trials
• Subjective difficulty rating at end

Exclude participants with:

• Test accuracy < chance (12.5%) AND learning phase accuracy < chance
• Total experiment time < 10 minutes (not paying attention)
• Incomplete data (< 250 trials)

IRB Notes

This experiment requires ethics approval. Key points for your application:

• Minimal risk (screen-based visual task only)
• Anonymous data (no personal info collected)
• Voluntary participation with right to withdraw
• Normal/corrected vision required (self-reported)
• Duration: ~25-30 minutes