On the Multi-View Information Bottleneck Representation