Experience
Education
Bio
Research Interests and Highlights
Representation Learning
Self-supervised learning: Self-Supervised GAN
Compact representation: KD code, DPQ
Generative Modeling
Better generator: Self-Modulation (latent-depdent batch norm)
Better discriminator: Self-Supervised GAN
Discrete Structures and Efficient Modeling
Graph neural nets: Dissecting GNNs
Graph embedding: Categorical anomaly detection, Heterogeneous network embedding
Efficiency/Compression: Negative Sampling trick, KD code, DPQ, DS-softmax, Mixture of LowRank