忻斌健
Deep Learning Practioner & Software Developer
Biography
Binjian Xin is a deep learning practitioner and software developer. His interests include deep learning, artificial intelligence and software design.
Interests
- Deep Learning
- Autonomous Driving
- Time Series Modeling and Processing
- Computer Vision
Education
PhD in Image Processing, 2008
University of Karlsruhe
MSc in Robotics, 2002
Tongji University
BSc in Electrical Engineering, 1998
Tongji University
Skills
Technical
Python
C/C++
Data Science
MLOps
Languages
Chinese
English
German
Hobbies
Hiking
Reading
Photography
Cats
Experience
Senior Technical Director, Advanced Development
- Reinforcemnt learning based BEV controller optimization in multimodal complex environments.
- Timeseries anomaly detection and battery state of safety prediction based on generative models.
- Software design & development of streaming data pipelines
Senior Manager, Autonomous Driving
- Advanced hardware and software design of autonomous driving systems.
- Development of intelligent charging and automatic parking assistance system.
- Application and operation of Intelligent Connected Vehicle (ICV) road test in Shanghai and Beijing.
Technical Manager, ADAS
- System & software architecture design for active safety domain unit (ADU).
- PATAC ADU A sample: system and software architecture of embedded platform.
- Software architecture of SAIC-MAXUS SV73 highway assist.
- Camera based driver monitoring system.
Software Manager
- SOP project of instrument clusters.
Senior Manager, Software
- SOP project of BCM and PEPS.
- Platform project of PEPS, BCM, BSW.
Senior Manager, ADAS
- Development of video based ADAS system.
- SOP projects of camera based parking systems.
- Supervision of ADAS advanced research.
Projects
*
Python CAN application wrapper:
- encapsulation with cached properties,
- validation via Pydantic,
- virtual CAN testing via local Actions,
- a2l file handling with lazy loading and dynamic properties,
- lazy loading by streaming json processing.
Time series analysis with generative models:
- Vanilla GAN
- Wasserstein GAN
- Wasserstein GAN with gradient penalty
Time sequence data pipleline framework for deep reinforcement learning:
- coordinated ETL and ML pipelines with asynchronous data pipelines,
- online local/distributed training,
- reinforcement learning models with regular and recurrent models,
- offline reinforcement learning with Implict Diffusion Q-learning.
Publications
Quickly discover relevant content by filtering publications.
(2024).
Natural language interface and large language model based autonomous driving decision module.
(2024).
Natural language interface and multimodal foundation model based autonomous driving desicion system.
