Guilin Li

Generative Pretraining at Scale: Transformer-Based Encoding of Transactional Behavior for Fraud Detection

CROLoss: Towards a Customizable Loss for Retrieval Models in Recommender Systems

Dropnas: Grouped operation dropout for differentiable architecture search

Autofis: Automatic feature interaction selection in factorization models for click-through rate prediction

Bayesian optimal designs for efficient estimation of the optimum point with generalised linear models

Residual distillation: Towards portable deep neural networks without shortcuts

Revisiting knowledge distillation via label smoothing regularization