Yichao Liu

Use imitation learning to improve performance in goaldirected navigation on the CARLA simulation. Feed the multi-task network with sequence images in order to predict environment variables, which affect the decision making. Finetuning the multi-task network to get the best performance.

Working on Model compression and Neural Architecture Search (NAS), which is very useful for embedding neural network into hardware. Extremely improved DARTS (a kind of NAS) performance by diversifying structure. Result: With only less than 1.7 M parameters achieved 0.97 accuracy on CIFAR10.

SW test for Autonomous driving system and finding bugs on daily road testing (mainly on perception and planning parts). Road testing on different condition (highway, downtown road and different kinds of rain days). Based on sampled point cloud, draw the HD map.

Based on NGSIM (a highway dataset released by US transportation department) to create my own dataset. Based on SVM to build up the lane change prediction model for Autonomous driving. Result: With only 5 frames to predict intension after 3s.

research interest: model compression, Neural Architecture Search (NAS)

research interest: autoencoder, unsupervised learning, Graph publication:Graph Regularized Sparse Autoencoders With Nonnegativity Constraints Data Mining Approach to Chinese Food Analysis for Diet-Related Cardiometabolic Diseases