CV
Xiangxiang Chu is a Senior Director at Alibaba and Head of DreamX. He leads a product-facing AI organization at AMAP spanning spatial intelligence, multimodal foundation models, generative AI, reinforcement learning, world models, and agent systems. His work connects original research and reproducible open source with nationwide product deployment. Previously, he built foundational AI teams at Meituan and Xiaomi.
Citations: Google Scholar · Paper count includes published works and confirmed acceptances as of September 27, 2026; see Publications for the counting scope.
Professional Experience
Leads DreamX, an organization with two complementary mandates: building frontier AI systems that reach production, and advancing the core mobility algorithms behind AMAP’s route planning, ETA prediction, and recommendation. The work spans an ecosystem serving 300M+ users daily.
- Scaled DreamX from 30 to 100+ members, including 30 interns; established four technical portfolios from the ground up and built a strong internal leadership bench
- Led the nationwide delivery of Navigation Live, bringing DreamX Agent into a large-scale multimodal navigation product and coordinating dozens of researchers and engineers across multiple business teams
- Led the productization of AIGC and Creator through visual content enhancement for AMAP’s Saojiebang and physically grounded video generation
- Led a research and open-source program producing 60+ papers at leading conferences and journals and a portfolio of public releases through the AMAP-ML GitHub organization
Built the foundational vision team from 0 to 30 members, covering model architecture, multimodal foundation models, detection, and industrial perception.
- Created first-authored research including Twins (NeurIPS 2021), CPVT (ICLR 2023), VisionLLaMA (ECCV 2024), MobileVLM, and QARepVGG (AAAI 2024)
- Led the open-source YOLOv6 industrial detection framework and the department-wide production adoption of Twins and YOLOv6; QARepVGG addressed the quantization bottleneck in RepVGG-style deployment
- Shipped 3D perception systems for autonomous delivery vehicles and drones, reducing annotation and serving costs
Founded and grew Xiaomi’s AutoML team from 0 to 10 members, connecting neural architecture search with mobile deployment.
- Developed the first-authored NAS research line spanning FairNAS (ICCV 2021), FairDARTS (ECCV 2020), DARTS- (ICLR 2021), and FALSR
- Deployed AutoML-designed real-time super-resolution and portrait-segmentation models across 100M+ smartphones, with demonstrations featured in major Xiaomi product launches
- Won 2nd place in Xiaomi’s first “Million Dollar Prize” for Automated Neural Network Design
- Core contributor to the “Complex Power Grid Autonomy — Collaborative Automatic Voltage Control” project, recognized with the State Scientific and Technological Progress Award, First Prize (2018)
- Developed large-scale data analytics and machine learning solutions at IBM China Research Lab
Selected Contributions
First-Author Research
- GPG — Simple and strong reinforcement learning for model reasoning · ICLR 2026 · First author · Adopted by ByteDance’s VERL
- USP — Unified self-supervised pretraining for image generation and understanding · ICCV 2025 · First author
- VisionLLaMA — LLaMA-style vision foundation architecture with auto-scaling 2D RoPE · ECCV 2024 · First author
- MobileVLM — Compact vision-language model designed for real-time mobile deployment · First author
- QARepVGG — Quantization-aware RepVGG for industrial deployment · AAAI 2024 · First author
- Twins, CPVT, and FairNAS — First-authored architecture research recognized on PaperDigest’s Most Influential lists
Systems & Teams Led
- DreamX-World — General-purpose interactive world model with an open-source 5B model and one-minute generation
- DreamX-Creator — Native joint audio-video generation with 2K refinement
- DreamX-Phi — Geometry-aware, action-conditioned video world model for robotic manipulation
- LongHorizon-Harness — Verified long-horizon computer use with durable task state and a Manage-Execute-Audit loop
- MobilityBench — Real-world route-planning agent benchmark · KDD 2026 Oral
- YOLOv6 — Industrial real-time object detection framework with broad open-source and production adoption
→ Full publication list (120+ papers and preprints)
Honors & Recognition
- State Scientific and Technological Progress Award, First Prize, 2018 — core contributor to the Complex Power Grid Autonomy project
- Top 100 AI Scholars, AMiner 2023
- 3 first-authored papers on PaperDigest's Most Influential lists: FairNAS, Twins, and CPVT
- 2nd Place, Xiaomi "Million Dollar Prize" — Automated Neural Network Design
Professional Service
Education
- M.S. in Electrical Engineering, Tsinghua University, 2012
- B.S. in Electrical Engineering, Southeast University, 2010