Lingxiao Zhao Email & Phone Number

California, United States

AI Incubation, full-stack LLM, from pretrain to sft and agent. Large-scale data processing.

United States

Working on pretraining large language model.

Working on graph generative model.

Working on more expressive graph neural network.

Yorktown Heights, New York, United States

Working on graph anomaly detection.

Greater Pittsburgh Area

Working with Prof. Leman Akoglu on graph mining and machine learning algorithms.Project 1: Optimal Graph for Semi-Supervised Learning- Designed a hyper-loss over validation set measuring task-based "optimal" graph for SSL.- Proposed an algorithm minimizing the hyper-loss based on hyper-gradient of the graph. - Boosted the runtime 10x faster by using tensor-form update and efficient sparse operation.- Paralleled it by incorporating and modifying HyperBand, an adaptive resource allocation alg.- Paper is accepted by CIMK 2018, 2nd author but equal contribution. Will present it on CIKM. Project 2: Optimal Counterfactual Explanation for Anomaly Detection - Proposed to explain tree-based anomaly detection model using counterfactual explanation.- Designed a cluster and correlation sensitive cost function for optimal counterfactual. - Designed a density-based "possible-world" constraint for optimal counterfactual.- Built the constrained optimization problem as MINLP and solve it using BARON. - Proposed a natural gradient & stochastic computational graph based method to estimate the "gradient" of discontinuous tree-ensemble models. And designed an approximation solver using augmented Lagrangian based on the estimated gradient. Boost the runtime a lot.- This work is targeting SDM 2018. Will be the main contributor.

Doctor Of Philosophy - Phd, Machine Learning Joint Public Policy

I'm doing research under the supervision of Prof. Leman Akoglu. The topics that I'm interested in: - Graph related algorithms, e.g. graph.

Master’S Degree, Electrical And Computer Engineering, 3.90/4.00

Activities and Societies: Specialized at machine learning techniques.

Bachelor’S Degree, Electrical And Electronics Engineering, 92.1/100

Activities and Societies: Working with Prof. Zhaohong Bie on economical dispatch for power system.Rank 7/370