神经网络的特定算法可解释性：以纹理为例的案例研究

中文摘要 AI 辅助

我们开发了一个原则性框架，用于构建其特定参数实现具有明确算法解释的神经网络。现有的受算法启发的架构可以解释网络的计算结构，但在标准训练后，学习到的参数不一定与激励算法保持清晰的关系。我们如下解决这一差距。首先，我们将每个数据点建模为类依赖随机过程的一个样本，并假设该过程的统计量可以从单个样本中估计，且这些统计量足以区分各个类别。然后，我们构建一个神经网络，其初始参数精确实现了一种用于估计这些统计量的算法，使网络完全可解释。为了考虑理想化模型与真实数据之间的不匹配，我们在控制网络偏离算法初始化的同时对该网络进行微调。因此，训练后的网络大致保留了初始网络的解释。PAC-Bayesian分析产生了一个统一的泛化界，其复杂度项随微调半径缩放，为我们的方法提供了统计动机。我们使用散射变换来估计判别统计量，将该框架实例化用于纹理分类。

英文摘要

We develop a principled framework for constructing neural networks whose specific parameter realizations admit an explicit algorithmic interpretation. Existing algorithm-inspired architectures can explain the computational structure of a network, yet after standard training the learned parameters need not retain a clear relation to the motivating algorithm. We address this gap as follows. First, we model each data point as a sample of a class-dependent stochastic process and assume that statistics of this process can be estimated from a single sample and these statistics are sufficient to distinguish the classes. We then construct a neural network whose initial parameters exactly implement an algorithm for estimating these statistics, making the network fully interpretable. To account for mismatch between the idealized model and real data, we fine-tune this network while controlling its deviation from the algorithmic initialization. The trained network hence roughly retains the interpretation of the initial network. A PAC-Bayesian analysis yields a uniform generalization bound whose complexity term scales with the fine-tuning radius, providing a statistical motivation for our approach. We instantiate the framework for texture classification using the scattering transform to estimate the discriminative statistics.

发表机构

• Technion – Israel Institute of Technology（以色列理工学院）
• University of Minnesota（明尼苏达大学）

机构由 AI 辅助整理，请以论文原文为准。