I have been hanging out mostly at the network coding sessions. For those who haven't seen the term, the one sentence description of network coding is that you combine routing with coding -- intermediate nodes do not just store and forward, they get to evaluate functions of received packets and send those on as needed. Why aren't there more CS theory people working on network coding (myself included!)? It's a huge relatively new area in information theory and networking. There are lots of interesting, challenging open problems. Lots more info here. Also, there's a Scientific American article this month by three of the leaders in the field -- Michelle Effros, Ralf Koetter, and Muriel Medard (three top names in information theory that CS people should become familiar with if they are not...) -- that gives excellent high-level background.
The most interesting talk so far was the work by Koetter and Kschischang, which reduces network coding with erasure and errors to an interesting coding variant. In this setting, codewords are vector subspaces; erasures delete a basis vector from the subspace and errors add a spurious basis vector to the subspace. What do codes in this world look like? (Hopefully I'm doing the question some justice -- it was a great talk.)
Michael Luby and Amin Shokrollahi won the IEEE Eric E. Sumner Award "For bridging mathematics, internet design, and mobile broadcasting as well as successful standardization." Go Digital Fountain.
I gave a talk on good codes for limited insertion/deletion channels.
At the lunch, I talked with some people about how the information theory community doesn't really have a "competitive" conference -- like a FOCS/STOC, or a SIGCOMM. They were wondering about how to and the impact of adding one. I suggested adding special single-track sessions (as opposed to the eight parallel sessions) for the key papers to ISIT.