Disentangled Information Bottleneck
This work provides a more stable and effective method for information bottleneck applications, which could benefit researchers and practitioners in machine learning by simplifying model optimization and improving performance across various tasks.
The authors propose Disentangled Information Bottleneck (DisenIB) to address the difficulty of optimizing the traditional Information Bottleneck (IB) method and the trade-off between compression and prediction performance. DisenIB achieves maximum compression of the source variable without losing target prediction performance.
The information bottleneck (IB) method is a technique for extracting information that is relevant for predicting the target random variable from the source random variable, which is typically implemented by optimizing the IB Lagrangian that balances the compression and prediction terms. However, the IB Lagrangian is hard to optimize, and multiple trials for tuning values of Lagrangian multiplier are required. Moreover, we show that the prediction performance strictly decreases as the compression gets stronger during optimizing the IB Lagrangian. In this paper, we implement the IB method from the perspective of supervised disentangling. Specifically, we introduce Disentangled Information Bottleneck (DisenIB) that is consistent on compressing source maximally without target prediction performance loss (maximum compression). Theoretical and experimental results demonstrate that our method is consistent on maximum compression, and performs well in terms of generalization, robustness to adversarial attack, out-of-distribution detection, and supervised disentangling.