On the Information Bottleneck for Stable Variables
In the authors' words
We consider a generalization of the Gaussian information bottleneck problem to jointly stable variables and . Specifically, for , where is stable and independent of , we find optimal stochastic linear encoders of the form , being stable and independent of . We characterize the linear stable information bottleneck in closed-form and its critical value. Furthermore, we show that, except for the Gaussian case, such stochastic linear encoders are strictly suboptimal, however asymptotically optimal at large compression rate. Our linear solution recovers the Gaussian information bottleneck for jointly Gaussian variables.
Main resultLimitation the authors admit
Appeared: Wednesday, October 7. arXiv. Preprint, not yet peer-reviewed.
Authors' comment: 13 pages, 3 figures