Validation & Benchmarks

Empirical validation, performance benchmarks, and experimental results for HoloVec.

This section provides evidence of correctness, performance measurements, and comparisons with theoretical predictions and other implementations.

Contents

Model Validation

VSA Model Validation Results contains comprehensive validation results for all VSA models:

Validation Methodology:

  • Theoretical property verification

  • Similarity distribution analysis

  • Binding/unbinding correctness

  • Bundling capacity tests

  • Noise tolerance measurements

  • Comparative benchmarks

Models Tested:

  • MAP (Multiply-Add-Permute)

  • FHRR (Fourier Holographic Reduced Representations)

  • HRR (Holographic Reduced Representations)

  • BSC (Binary Spatter Codes)

  • BSDC (Block-Structured Distributed Codes)

  • GHRR (Generalized HRR)

  • VTB (Vector-derived Transformation Binding)

Key Results

Performance:

  • NumPy backend: Baseline CPU performance

  • PyTorch backend: 10-100x speedup on GPU

  • JAX backend: JIT compilation benefits

Accuracy:

  • Binding/unbinding: Near-perfect recovery (similarity > 0.99)

  • Bundling capacity: Matches theoretical predictions

  • Noise tolerance: Graceful degradation up to 20% corruption

Comparison:

  • Matches academic implementations

  • Validates theoretical models

  • Confirms encoder properties

See Also

  • Theory & Foundations - Theoretical foundations

  • HoloVec Examples - Practical examples

  • 31_performance_benchmarks - Performance benchmarks example

  • 32_distributed_representations - Capacity analysis example

  • 33_error_handling_robustness - Robustness testing example