A Unified Framework for Behavior-Based Identification and Authentication of Avatars and Digital Twins

This framework provides a unified approach for identifying and authenticating avatars and digital twins based on their behavioral patterns.

Setup Instructions

To setup and test the proposed framework, run the files in the following order:

  1. Update paths in all three files. Replace the placeholder paths with paths on your machine:
    • 01_KAN_AE.ipynb: set BASE_PATH, MODEL_DATA_PATH, MODEL_PATH, and ENC_DATA_PATH.
    • 02_BiGRU_Classifier.ipynb: set BASE_PATH, MODEL_DATA_PATH, and MODEL_PATH.
    • 03_UBM_Template_Auth.py: set DIR_PREFIX, MODEL_DIR, SAMPLE_PATH, and EXP_PATH.

    Confirm that each input directory contains the expected .npy files and that each export directory is writable.

  2. Learn spatial behavior representations. Open 01_KAN_AE.ipynb and run the cells from top to bottom. The notebook converts the time-series signals into timestamp-level spatial samples, trains the KAN autoencoder, and exports the KAN-AE model and encoded data.

  3. Learn temporal behavior representations. Open 02_BiGRU_Classifier.ipynb and run the cells from top to bottom. It loads the encoded data from the KAN-AE step, trains the bidirectional GRU classifier, evaluates it, and exports the trained model.

  4. Enroll and test authentication. Run 03_UBM_Template_Auth.ipynb after the BiGRU model and encoded data have been exported. The notebook creates one prototype/template per enrolled user, tunes the verification threshold on development data, and evaluates the held-out authentication test data.

NOTE: Results are written into the configured experiment output directory. The export includes templates/prototypes, trial scores, summary and per-user metrics, threshold/configuration JSON files, ROC and enrollment-size arrays, and ROC/enrollment-size plots.