Interpretable Neural Network Decoupling
arXiv:1906.01166
Abstract
The remarkable performance of convolutional neural networks (CNNs) is entangled with their huge number of uninterpretable parameters, which has become the bottleneck limiting the exploitation of their full potential. Towards network interpretation, previous endeavors mainly resort to the single filter analysis, which however ignores the relationship between filters. In this paper, we propose a novel architecture decoupling method to interpret the network from a perspective of investigating its calculation paths. More specifically, we introduce a novel architecture controlling module in each layer to encode the network architecture by a vector. By maximizing the mutual information between the vectors and input images, the module is trained to select specific filters to distill a unique calculation path for each input. Furthermore, to improve the interpretability and compactness of the decoupled network, the output of each layer is encoded to align the architecture encoding vector with the constraint of sparsity regularization. Unlike conventional pixel-level or filter-level network interpretation methods, we propose a path-level analysis to explore the relationship between the combination of filter and semantic concepts, which is more suitable to interpret the working rationale of the decoupled network. Extensive experiments show that the decoupled network achieves several applications, i.e., network interpretation, network acceleration, and adversarial samples detection.
arXiv comment20 pages, 12 figures
References in corpus (16)
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
- YOLOv3: An Incremental Improvement
- Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
- How transferable are features in deep neural networks?
- Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
- Understanding Neural Networks Through Deep Visualization
- Understanding Black-box Predictions via Influence Functions
- Object Detectors Emerge in Deep Scene CNNs
- On the importance of single directions for generalization
- Fast Decoding in Sequence Models using Discrete Latent Variables
- Conditional Computation in Neural Networks for faster models
- PerforatedCNNs: Acceleration through Elimination of Redundant Convolutions
- Identifying Unknown Unknowns in the Open World: Representations and Policies for Guided Exploration
- Unsupervised learning of object semantic parts from internal states of CNNs by population encoding
- Understanding deep features with computer-generated imagery