Flexible Variational Information Bottleneck: Achieving Diverse Compression with a Single Training
This addresses a computational bottleneck for researchers and practitioners using IB in classification tasks, offering a more efficient approach.
The paper tackles the computational expense of the Information Bottleneck (IB) framework, which traditionally requires multiple training cycles to find trade-off parameters, by introducing Flexible Variational Information Bottleneck (FVIB) that achieves optimal models for all parameter values with a single training, empirically matching effectiveness and outperforming other methods in calibration performance.
Information Bottleneck (IB) is a widely used framework that enables the extraction of information related to a target random variable from a source random variable. In the objective function, IB controls the trade-off between data compression and predictiveness through the Lagrange multiplier $β$. Traditionally, to find the trade-off to be learned, IB requires a search for $β$ through multiple training cycles, which is computationally expensive. In this study, we introduce Flexible Variational Information Bottleneck (FVIB), an innovative framework for classification task that can obtain optimal models for all values of $β$ with single, computationally efficient training. We theoretically demonstrate that across all values of reasonable $β$, FVIB can simultaneously maximize an approximation of the objective function for Variational Information Bottleneck (VIB), the conventional IB method. Then we empirically show that FVIB can learn the VIB objective as effectively as VIB. Furthermore, in terms of calibration performance, FVIB outperforms other IB and calibration methods by enabling continuous optimization of $β$. Our codes are available at https://github.com/sotakudo/fvib.