The effect of task and training on intermediate representations in convolutional neural networks revealed with modified RV similarity analysis
This work provides insights into neural network training dynamics for researchers, but it is incremental as it builds on existing similarity metrics and training methods.
The study investigated how different training methods affect intermediate representations in convolutional neural networks using a modified similarity metric, finding that freeze training's superior performance may be due to representational differences in the penultimate layer and that inputs and targets anchor representations in extreme layers.
Centered Kernel Alignment (CKA) was recently proposed as a similarity metric for comparing activation patterns in deep networks. Here we experiment with the modified RV-coefficient (RV2), which has very similar properties as CKA while being less sensitive to dataset size. We compare the representations of networks that received varying amounts of training on different layers: a standard trained network (all parameters updated at every step), a freeze trained network (layers gradually frozen during training), random networks (only some layers trained), and a completely untrained network. We found that RV2 was able to recover expected similarity patterns and provide interpretable similarity matrices that suggested hypotheses about how representations are affected by different training recipes. We propose that the superior performance achieved by freeze training can be attributed to representational differences in the penultimate layer. Our comparisons of random networks suggest that the inputs and targets serve as anchors on the representations in the lowest and highest layers.