Insights on representational similarity in neural networks with canonical correlation

Abstract

Comparing different neural network representations and determining how representations evolve over time remain challenging open questions in our understanding of the function of neural networks. Comparing representations in neural networks is fundamentally difficult as the structure of representations varies greatly, even across groups of networks trained on identical tasks, and over the course of training. Here, we develop projection weighted CCA (Canonical Correlation Analysis) as a tool for understanding neural networks, building off of SVCCA, a recently proposed method (Raghu et al., 2017). We first improve the core method, showing how to differentiate between signal and noise, and then apply this technique to compare across a group of CNNs, demonstrating that networks which generalize converge to more similar representations than networks which memorize, that wider networks converge to more similar solutions than narrow networks, and that trained networks with identical topology but different learning rates converge to distinct clusters with diverse representations. We also investigate the representational dynamics of RNNs, across both training and sequential timesteps, finding that RNNs converge in a bottom-up pattern over the course of training and that the hidden state is highly variable over the course of a sequence, even when accounting for linear transforms. Together, these results provide new insights into the function of CNNs and RNNs, and demonstrate the utility of using CCA to understand representations.

arXiv commentNIPS 2018

References in corpus (9)

• Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding
• Understanding Neural Networks Through Deep Visualization
• Visualizing and Understanding Recurrent Networks
• Deep Neural Networks as Gaussian Processes
• Regularizing and Optimizing LSTM Language Models
• On the importance of single directions for generalization
• Convergent Learning: Do different neural networks learn the same representations?
• PerforatedCNNs: Acceleration through Elimination of Redundant Convolutions
• Measuring the Intrinsic Dimension of Objective Landscapes

Cited by in corpus (9)

• Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML
• How deep is deep enough? -- Quantifying class separability in the hidden layers of deep neural networks
• ModelDiff: Testing-Based DNN Similarity Comparison for Model Reuse Detection
• Unsupervised Model Selection for Variational Disentangled Representation Learning
• PipeTransformer: Automated Elastic Pipelining for Distributed Training of Transformers
• Critical Assessment of Transfer Learning for Medical Image Segmentation with Fully Convolutional Neural Networks
• YASENN: Explaining Neural Networks via Partitioning Activation Sequences
• Representation Transfer by Optimal Transport
• Word Interdependence Exposes How LSTMs Compose Representations