CV

Lichuan Xiang | 向力川

Education

Ph.D. in Computer Science (2019-2024)

University of Warwick, United Kingdom

• Research Focus: Efficient Neural Architecture Search and Pruning
• Thesis: Efficient Automated Neural Architecture Design
• Supervised by: Prof. Hongkai Wen
• Examiner: External Prof. Tian He, Internal Prof. Victor Sanchez

M.Sc. in Computer Science (2017-2018)

University of Warwick, United Kingdom

• Thesis: Transfer Learning and CNN in Human Pose Estimation
• Graduated with Merit
• Supervised by: Prof. Hongkai Wen

B.Eng. in Digital Media Technology (2012-2016)

Huaqiao University, China

• GPA: 84.1/100 (Top 4 in class)
• Thesis: Human Action Estimation based on RGBD Camera using HOPC Features

Professional Experience

Founding Research Scientist (Full-time) (2024.04-Present)

Collov.ai, Beijing

• Lead Project Indoor 3D space reconstruction from Single Image: related to depth estimation, object recognition and segmentation, point-cloud analysis, camera estimation tasks.
• Lead Diffusion Image-Edit Model Training on different tasks: Object Erase, Object Insert, Super-resolution, Space planning
• Lead Diffusion model performance optimization, on control module pruning, acceleration and pipeline parallelization
• Lead Core Development on Product Virtual Staging Tools $200K MRR from scratch to now, Monthly growth up to 66%

Honorary Research Fellow (Full-time) (2024-2027)

University of Warwick

• Lead advanced research in machine learning and computer vision
• Supervise doctoral research projects
• Collaborate with industry partners on cutting-edge AI solutions

Data Scientist - KTP Associate (Full-time) (2023-2024)

Warwick Business School

• Developed legal large-scale language models and RAG systems
• Conducted deep learning and data analysis training sessions
• Implemented advanced NLP solutions for legal document processing

Academic Collaborator (Part-time) (2019-2023)

Samsung AI Centre Cambridge

• Researched neural architecture search and model compression
• Developed super-resolution algorithms
• Published multiple papers in top-tier conferences

Data Scientist & Research Assistant (Part-time) (2020-2023)

University of Warwick

• Led development of “AI Essay-Analyst” - an academic writing tool featured in Times Higher Education
  • Implemented deep learning and NLP algorithms for automated essay analysis
  • Created visualization tools for argument structure and writing quality assessment
  • Integrated with WBS marking criteria for comprehensive feedback generation
• Co-organized “Streetscape Perception Modelling” project at PLATIAL’21 Conference
  • Collaborated with researchers from University of Warwick on urban perception modeling
  • Focused on theoretical frameworks and methodological approaches in streetscape analysis

Computer Vision Engineer (Full-time) (2018-2019)

4Paradigm, Beijing

• Developed AutoCV deep learning framework
• Implemented automatic data augmentation systems
• Created evaluation tools for AutoDL
• Contributed to AutoDL Challenge platform development
  • Developed AutoCV access platform for automated deep learning challenges
  • Supported challenge organization and technical infrastructure
  • Results published in IEEE TPAMI (DOI: 10.1109/TPAMI.2021.3075372)

Professional Activities

Workshops & Events

• Co-organizer, “Streetscape Perception Modelling – Theoretical Considerations and Methodological Possibilities”, PLATIAL’21 Conference (December 2021)
  • Collaborated with researchers from University of Warwick on urban perception modeling
  • Focused on theoretical frameworks and methodological approaches in streetscape analysis

Selected Publications

Conference Papers

1. Xiang, L., et al. (2024). Towards neural architecture search through hierarchical generative modeling. ICML 2024.
2. Xiang, L., et al. (2023). Zero-cost operation scoring in differentiable architecture search. AAAI 2023.
3. Pham, T.,…Xiang, L., et al. (2023). Towards data-agnostic pruning at initialization: what makes a good sparse mask?. NeurIPS 2023.
4. Xiang, L., et al. (2023). Exploiting network compressibility and topology in zero-cost NAS. AutoML Conference 2023. 🏆 Best Paper Award
5. Xiang, L., et al. (2023). DPaI: Differentiable Pruning at Initialization with Node-Path Balance Principle. ICLR 2023.

Preprints

1. Xu, M., Xiang, L., et al. (2024). No More Adam: Learning Rate Scaling at Initialization is All You Need. arXiv.
2. Fang, Z., Xiang, L., et al. (2025). FlexControl: Computation-Aware ControlNet with Differentiable Router for Text-to-Image Generation. arXiv.

Workshop Papers

1. Xiang, L., et al. (2023). Generating Neural Network Architectures with Conditional Graph Normalizing Flows. AutoML Conference Workshop 2023.
2. Xiang, L., et al. (2021). Temporal kernel consistency for blind video super-resolution. ICCV Workshop 2021.

Honors & Awards

Academic Excellence

• 🎓 Full Ph.D. Scholarship, University of Warwick (2019-2023)
• 🏆 Best Paper Award, AutoML Conference 2023
• 🏆 Top 1 and Top2 Solutions on Zero-Cost NAS Competition, AutoML Conference 2022
• 🥇 Best Presentation Award (Machine Learning Track), Warwick Computer Science Graduate Conference 2021

Software & Innovation Competitions

• 🥇 First Prize, Fujian Software Design Competition (2015)
• 🥇 First Prize, Cross-Strait Information Service Innovation Competition (2013, 2014)
• 🥈 Second Prize, China College Students’ Computer Design Competition (2015)
• 🥈 Second Prize, Fujian “Challenge Cup” Academic Competition (2015)

Industry Recognition

• 🏅 CMB FinTech Training Camp Excellence Award (2018)

Skills

• Programming Languages: Python, C++, PyTorch, TensorFlow
• Research Areas: Neural Architecture Search, Model Compression, Computer Vision, Deep Learning
• Tools & Frameworks: AutoML, ControlNet, Diffusion Models, RAG Systems
• Languages: English (Fluent), Chinese (Native)