Exposition on Over-Squashing Problem on GNNs: Current Methods, Benchmarks and Challenges

The episode reviews the paper "Exposition on Over-Squashing Problem on GNNs," which details how information gets distorted when traversing large networks. Hosts discuss various solutions, including spatial, spectral, and implicit rewiring methods. The discussion highlights the need for a unified definition of over-squashing and emphasizes the trade-off between solving this problem and avoiding over-smoothing.

Open the full episode

More from AI Radio