Networked data often demonstrate the Pareto principle (i.e., 80/20 rule) with skewed class distributions, where most vertices belong to a few majority classes and minority classes only contain a handful of instances. When presented with imbalanced class distributions, existing graph embedding learning tends to bias to nodes from majority classes, leaving nodes from minority classes under-trained. In this paper, we propose Dual-Regularized Graph Convolutional Networks (DR-GCN) to handle multi-class imbalanced graphs, where two types of regularization are imposed to tackle class imbalanced representation learning. To ensure that all classes are equally represented, we propose a class-conditioned adversarial training process to facilitate the separation of labeled nodes. Meanwhile, to maintain training equilibrium (i.e., retaining quality of fit across all classes), we force unlabeled nodes to follow a similar latent distribution to the labeled nodes by minimizing their difference in the embedding space. Experiments on real-world imbalanced graphs demonstrate that DR-GCN outperforms the state-of-the-art methods in node classification, graph clustering, and visualization.

For adversarial but computationally bounded models of error, we construct appealingly simple and efficient cryptographic encoding and unique decoding schemes whose error-correction capability is much greater than classically possible. In particular: 1) For binary alphabets, we construct positive-rate coding schemes that are uniquely decodable under a 1/2 - γ error rate for any constant γ > 0. 2) For large alphabets, we construct coding schemes that are uniquely decodable under a 1 - R error rate for any information rate R > 0. Our results for large alphabets are actually optimal, since the "computationally bounded but adversarial channel" can simulate the behavior of the q-ary symmetric channel, where q denotes the size of the alphabet, the capacity of which is known to be upper-bounded by 1 - R. Our results hold under minimal assumptions on the communication infrastructure, namely: 1) we allow the channel to be more powerful than the receiver and 2) we only assume that some information about the sender-a public key-is known. (In particular, we do not require any shared secret key or joint local state between sender and receivers).