Directly optimizing content of intermediate layers with information bottleneck approach?
Directly optimizing content of intermediate layers with information bottleneck approach?