Yamin Cheng

Yamin Cheng

Auto-generated profile, based on publicly available information

Yamin Cheng

School of Computer Science and Engineering, School of Computer Science and Engineering,University of Electronic Science and Technology of China,Chengdu,Sichuan,China, China

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
25 October 2022 in IEEE Transactions on Circuits and Systems for Video Technology
Verified
Journal Article
Closed access
YCYamin Cheng
Yamin Cheng
ZWZhi Wang
Zhi Wang
WZWenhan Zhan
Wenhan Zhan
HDHancong Duan
Hancong Duan
Transformers are transforming the landscape of computer vision, especially for image-level recognition and instance-level detection tasks. Human-object interaction detection transformer (HOI-TR) is the first transformer-based end-to-end learning system for human-object interaction (HOI) detection; vision transformers build a simple multi-stage structure for multi-scale representation with single-scale patch and are the first patch-based transformer architecture for image-level recognition and instance-level detection. In this paper, we build a transformer-based multi-scale human-object interaction detector (MHOI), a novel method to integrate Vision and HOI detection Transformer, instead of directly incorporating two types of transformers, since the vision transformer lacks hierarchical architecture to handle the large variations in the scale of visual entities due to the single-scale patch partitioning. Specifically, MHOI embeds features of the same size (i.e., sequence length) with patches of variable scales simultaneously by utilizing overlapping convolutional patch embedding, then introduces an efficient transformer decoder that designs the query based on anchor points and essential auxiliary techniques to boost the HOI detection performance. Numerically, extensive experiments on several benchmarks demonstrate that our proposed framework outperforms prior existing methods coherently and achieves the impressive performance of 29.67 mAP on HICO-DET and 58.7 mAP on V-COCO, respectively.

23 December 2022
Verified
Conference Paper
Open access
ZWZhi Wang
Zhi Wang
HDHancong Duan
Hancong Duan
YCYamin Cheng
Yamin Cheng
Task scheduling plays an important role in query execution, which affects the response time and system throughput of queries. Current database systems use simple heuristic algorithms to determine the order of scheduled tasks and executor allocation. This makes it hard for the scheduler to make full use of the characteristics of the task graph and the state information of executors to optimize the scheduling process dynamically. Especially in heterogeneous environments, it is difficult for heuristic algorithms to generate a better sequence for task execution and balance loads of executors. To address these challenges, we design a DAG scheduler based on graph attention network and reinforcement learning to make scheduling decisions. At first, the scheduler extracts features of each node in the DAG through the graph attention network and utilizes an LSTM module to obtain the high-level representations. Then, the RL agent calculates the probabilities by these representations and selects the node with the maximum probability for each step. Finally, the parameters of the agent will be updated until all the nodes of a DAG have been scheduled. The experimental results based on random DAGs and TPC-H workload reveal that the proposed model can outperform the existing heuristic algorithms by 30% at most on average makespan, which also has significant improvements on TPC-H workload.