SignXAI Documentation

SignXAI2 is a comprehensive cross-framework explainable AI (XAI) library that provides unified access to explanation methods across TensorFlow and PyTorch. With over 200 XAI methods and automatic framework detection, SignXAI makes it easy to understand and interpret your deep learning models.

Key Features

  • Cross-Framework Compatibility: Single API works with both PyTorch and TensorFlow

  • 200+ XAI Methods: Comprehensive collection including gradients, LRP, CAM, and more

  • Automatic Framework Detection: No need to specify framework - SignXAI detects it automatically

  • Parameter Mapping: Consistent parameter names across frameworks

  • Robust Implementation: Extensive testing and validation across frameworks

Quick Start

Install SignXAI2 with your preferred framework:

# For PyTorch users:
pip install signxai2[pytorch]

# For TensorFlow users:
pip install signxai2[tensorflow]

# For both frameworks:
pip install signxai2[all]

# Note: Requires Python 3.9 or 3.10

Generate your first explanation:

from signxai import explain

# Works with any model - automatic framework detection!
explanation = explain(model, input_data, 'gradient')

Documentation Structure

Getting Started

User Guide

Tutorials

API Reference

Development

Method Categories

SignXAI organizes XAI methods into intuitive categories:

Gradient-Based Methods

  • gradient - Basic gradient attribution

  • smoothgrad - Noise-averaged gradients for robustness

  • integrated_gradients - Path integral gradients from baseline

  • vargrad - Variance-based gradient analysis

Layer-wise Relevance Propagation (LRP)

  • lrp_epsilon - Epsilon rule with stabilization

  • lrp_alpha_1_beta_0 - Alpha-beta rule (α=1, β=0)

  • lrp_alpha_2_beta_1 - Alpha-beta rule (α=2, β=1)

  • lrp_z - Z+ rule for positive contributions

Class Activation Methods (CAM)

  • grad_cam - Gradient-weighted Class Activation Mapping

  • guided_grad_cam - Combines GradCAM with guided backprop

Backpropagation Methods

  • guided_backprop - Guided backpropagation

  • deconvnet - Deconvolutional networks

SIGN Methods

  • gradient_x_sign - Gradient × SIGN (μ=0)

  • gradient_x_sign_mu - Gradient × SIGN with custom μ

  • smoothgrad_x_sign - SmoothGrad × SIGN

  • guided_backprop_x_sign - Guided Backprop × SIGN

Why SignXAI?

Unified Interface

One API works across PyTorch and TensorFlow - no more learning framework-specific tools.

Extensive Method Collection

Over 200 XAI methods implemented with consistent interfaces and validated across frameworks.

Research-Grade Quality

Rigorous implementation following original papers, with cross-framework validation.

Easy Integration

Drop-in replacement for existing XAI workflows with automatic parameter mapping.

Active Development

Regular updates, new methods, and community contributions.

Success Stories

SignXAI is used by researchers and practitioners worldwide for:

  • Medical AI: Explaining diagnostic model predictions

  • Computer Vision: Understanding image classification and object detection

  • Time Series Analysis: Interpreting forecasting and anomaly detection models

  • NLP: Analyzing attention patterns and feature importance

  • Research: Cross-framework method validation and benchmarking

Performance & Reliability

  • Cross-Framework Consistency: High correlation between PyTorch and TensorFlow implementations

  • Comprehensive Testing: Extensive test coverage for all major method combinations

  • Performance Optimized: Efficient implementations with minimal overhead

  • Memory Efficient: Smart batching and memory management for large models

Community & Support

  • GitHub Repository: https://github.com/IRISlaboratory/signxai2

  • Issue Tracker: Report bugs and request features

  • Discussions: Community support and method discussions

  • Contributing: Guidelines for contributing new methods and improvements

Citation

If you use SignXAI in your research, please cite:

@article{Gumpfer2023SIGN,
   title = {SIGNed explanations: Unveiling relevant features by reducing bias},
   author = {Nils Gumpfer and Joshua Prim and Till Keller and Bernhard Seeger and Michael Guckert and Jennifer Hannig},
   journal = {Information Fusion},
   pages = {101883},
   year = {2023},
   issn = {1566-2535},
   doi = {https://doi.org/10.1016/j.inffus.2023.101883},
   url = {https://www.sciencedirect.com/science/article/pii/S1566253523001999}
}

License

SignXAI is released under the BSD 3-Clause License. See the License for details.

Indices and tables