EegFun

Appearance

RSA ​

Overview ​

This demo demonstrates Representational Similarity Analysis (RSA) for analyzing neural representational geometries.

What is RSA? ​

RSA quantifies how neural patterns represent different stimuli/conditions:

  1. Compute dissimilarity: Calculate distance between conditions (e.g., correlation distance)

  2. Create RDM: Representational Dissimilarity Matrix showing all pairwise distances

  3. Compare geometries: Relate neural RDMs to behavior, models, or other brain regions

RSA Workflow ​

  1. Extract patterns: Get spatial or spatiotemporal patterns for each condition

  2. Compute RDM: Calculate dissimilarity between all condition pairs

  3. Statistical testing: Assess significance of representational structure

  4. Model comparison: Compare neural RDMs to theoretical predictions

Applications ​

  • Compare representational geometry across time

  • Test computational models of cognition

  • Relate neural to behavioral similarity

  • Cross-modality comparisons

Advantages ​

  • Multivariate (uses all channels)

  • Model-agnostic representational space

  • Can compare across different measurement modalities

Code Examples ​

Show Code
julia
# Demo: Representational Similarity Analysis (RSA)
# Shows RSA analysis with multiple participants, dissimilarity measures,
# grand averaging with noise ceiling, RDM heatmaps, and timecourse visualization.

using EegFun

@info EegFun.section("Representational Similarity Analysis (RSA)")
println("Basic RSA analysis...")

# Create test data for multiple participants
all_participant_epochs = [
    EegFun.create_test_epoch_data_vector(
        conditions = 1:3,
        n_epochs = 100,
        n_channels = 3,
        fs = 250,
        n = 250,  # 1 second of data at 250 Hz
    ) for _ = 1:5
]

# Perform RSA for each participant using batch method
all_rsa_results = EegFun.rsa(all_participant_epochs; dissimilarity_measure = :correlation, average_trials = true)
all_rsa_results =
    EegFun.rsa(all_participant_epochs; dissimilarity_measure = :correlation, average_trials = true, normalize_method = :zscore)

# Compute grand average with noise ceiling
grand_avg_rsa = EegFun.grand_average(all_rsa_results)

# RSA Plots
EegFun.plot_rdm_heatmap(grand_avg_rsa, time_point = 0.0, title = "Grand Average RDM")
EegFun.plot_rdm_heatmap(grand_avg_rsa, title = "Grand Average RDM (time-averaged)")
EegFun.plot_rdm_timecourse(grand_avg_rsa, title = "Dissimilarity Over Time")

See Also ​