2 个仓库
Awesome GitHub RepositoriesAttribution Algorithm Benchmarks
Comparing and evaluating custom interpretability methods against built-in attribution algorithms for reliability and performance.
Distinct from Algorithm Benchmarking Libraries: Distinct from general Algorithm Benchmarking Libraries: specifically benchmarks attribution and interpretability algorithms, not general ML or RL algorithms.
Explore 2 awesome GitHub repositories matching artificial intelligence & ml · Attribution Algorithm Benchmarks. Refine with filters or upvote what's useful.
Awesome Attribution Algorithm Benchmarks GitHub Repositories
用 AI 发现最棒的仓库。我们将通过 AI 为您搜索最匹配的仓库。
- jacobgil/pytorch-grad-cam
jacobgil/pytorch-grad-cam
12,893在 GitHub 上查看该项目是一个用于 PyTorch 的计算机视觉可解释 AI 库和框架，提供了一套工具来可视化和审计深度神经网络的内部决策过程。它作为一个神经网络归因工具和调试实用程序，用于识别哪些图像区域驱动了模型预测。 该库以其对基于梯度和无梯度归因方法的支持而著称，允许在无需修改原始模型源代码的情况下生成视觉热力图和归因图。它通过视觉概念发现进一步脱颖而出，使用矩阵分解将内部激活分解为可解释的模式，并将潜在嵌入映射到像素重要性。 该框架涵盖了广泛的能力，包括热力图生成和细化、针对视觉 Transformer 等架构的空间转换，以及针对目标检测和语义分割等多任务视觉目标的适配。它还包括一个模型保真度评估套件，采用扰动分析、消融研究和定位测量来量化生成解释的忠实度。 该项目提供了用于动态激活钩子、自定义架构适配和目标驱动目标配置的机制，以将可解释性工具连接到各种模型输出。
Benchmarks various attribution algorithms like GradCAM and ScoreCAM to evaluate their reliability and performance.
Python - pytorch/captum
pytorch/captum
5,652在 GitHub 上查看Captum is an open-source library for explaining model predictions by attributing them to input features, neurons, and layers using gradient-based and perturbation-based methods. It provides a modular framework for implementing, evaluating, and combining a range of explanation techniques, including gradient-based attribution, perturbation-based analysis, game-theoretic Shapley value approximation, and surrogate model explanations, with support for parallelization and noise stabilization. The library distinguishes itself through its breadth of attribution methods and its support for advanced in
Compares and evaluates custom interpretability methods against built-in attribution algorithms for reliability and performance.
Python
