PANG Kaicheng's Homepage

/home

/CV

/Publications

/Blog

/Gallery

Home > Curriculum Vitae (CV)

Curriculum Vitae (CV)

Kaicheng PANG #

[email protected] | +86 18963695685 | +852 55335759
Personal Website | Google Scholar

Profile #

Currently serving as a Postdoctoral Research Associate and Technical Project Manager at the Laboratory for Artificial Intelligence in Design (AiDLab) and Chief Technology Officer (CTO) at Code-Create Limited, an AiDLab spin-off. Holds a Ph.D. in Fashion and Textile Design and an M.Sc. in Electrical Engineering from The Hong Kong Polytechnic University, as well as a B.Eng. in Automation from Nanjing University of Aeronautics and Astronautics. Research interests focus on Agentic AI, Vision-Language Models (VLM), and Open-Ended Subjective Sequential Decision-Making. Dedicated to exploring scientific challenges including Preference-based Reinforcement Learning, Sparse Reward Modeling, and Subjective Alignment Evaluation. In commercial implementation, successfully led an 11-member cross-functional team to develop the interactive design agent (AiDA), driving Code-Create to secure a $2 million strategic investment and accumulating over 10,000 professional users globally.

Education #

The Hong Kong Polytechnic University | 2021 – 2024

  • Doctor of Philosophy (Ph.D.), Fashion and Textile Design (GPA: 3.57 / 4.00)
  • Supervisor: Prof. Calvin Wong
  • Thesis: Personal Stylist: An Intelligent Evaluation Model for Personalized Styling Advice

The Hong Kong Polytechnic University | 2016 – 2017

  • Master of Science (M.Sc.), Electrical Engineering (GPA: 3.55 / 4.00)

Nanjing University of Aeronautics and Astronautics | 2012 – 2016

  • Bachelor of Engineering (B.Eng.), Automation (GPA: 2.9 / 5.0)

Professional Experience #

Laboratory for Artificial Intelligence in Design (AiDLab) | 2024 – Present
Postdoctoral Fellow and Technical Project Manager

  • Investigated non-linear synergistic effects in multimodal heterogeneous entities. Addressed the scientific challenge of quantitative evaluation and explicit modeling by proposing the Hierarchical Outfit Network (HON) based on higher-order data structures. Constructed Aesthetic 100, an open-source evaluation benchmark covering multiple scenarios, achieving a 3.6% accuracy improvement over state-of-the-art baselines.
  • Reconstructed the sequential composition task paradigm in high-dimensional decision spaces. Addressed the challenges of sparse rewards and dynamic termination in subjective scenarios by proposing the Unified Aesthetic Foresight Model (UAFM). Formulated the outfit composition process as a deterministic Markov Decision Process (MDP) under the Preference-based RL framework, and introduced afterstate temporal difference to resolve the credit assignment problem. Empowered multimodal agents with human-preference-aligned potential evaluation and dynamic termination capabilities, outperforming baselines by 18.09% in human evaluation tests.
  • Developed FashionM3, a multimodal, multi-round interactive agent for the fashion domain. Constructed FashionRec, an open-source fine-tuning dataset containing over 330,000 high-quality multimodal dialogue trajectories. Applied supervised fine-tuning to align Vision-Language Models (VLM), enabling FashionM3 with multi-tool calling capabilities such as personalized recommendation and item image generation, significantly enhancing real-world interaction quality and recommendation capabilities.
  • During Ph.D., focused on cross-modal matching and cognitive learning based on human physical constraints. Addressing the oversimplification of traditional physical priors, proposed the Visual Body-shape Aware Network (ViBA-Net) integrating visual and multi-dimensional parameters, and designed a visual-semantic mapping framework based on Multi-Label Graph Convolutional Networks (ML-GCN). By constructing multimodal joint embedding modules and attention mechanisms, achieved precise feature alignment under complex conditions and equipped predictions with attribute-level explainable attribution. Spearheaded the creation of condition-aligned datasets such as O4U and Body-Diverse, providing open-source data and algorithm baselines for fine-grained conditional recommendations.

Code-Create Limited | 2024 – Present
Chief Technology Officer (CTO)

  • Orchestrated the technical strategy for Code-Create, an AiDLab spin-off. Led an 11-member cross-functional team of doctoral students, research assistants, and developers as CTO to develop AiDA (Artificial Intelligence-based Design Assistant), fully directing the core agent architecture and foundational R&D. Drove the company to secure a $2 million strategic investment from Eschange Capital (Report). Successfully promoted AiDA’s international commercial partnerships and educational integration, with core clients including renowned global brands and institutions such as ANTEPRIMA, BESFXXK, and Paichai University (Report). The platform has served over 10,000 professional designers.

Laboratory for Artificial Intelligence in Design (AiDLab) | 2020 – 2021
Research Assistant

  • Trained explainable fashion compatibility models utilizing CNN visual features and Bi-LSTM sequential modeling, introducing gradient penalty regularization to achieve visual attribution of predictions.

Baidu, Inc. | 2018 – 2020
Senior Algorithm Engineer

  • Contributed to the Valet Parking system of the Apollo Autonomous Driving project, responsible for the architecture design and algorithm implementation of the vehicle localization system based on semantic information.

Department of Mechanical Engineering, The Hong Kong Polytechnic University | 2017 – 2018
Research Assistant

  • Participated in the development of an autonomous navigation system for special vehicles based on X-shaped suspension structures, responsible for developing path planning algorithms in complex terrains.

Publications #

  • Kaicheng Pang, Xingxing Zou, Ruohan Xu, Waikeung Wong. “Autonomous Fashion Outfit Composition via Unified Aesthetic Foresight Model”. Preprint arXiv:2608.13888, 2026. (Under review for ICLR 2027)
  • Kaicheng Pang, Xingxing Zou, Zowie Broach, Waikeung Wong. “Towards intelligent online cross-selling”. Expert Systems with Applications, 2025.
  • Kaicheng Pang, Xingxing Zou, Waikeung Wong. “FashionM3: Multimodal, Multitask, and Multiround Fashion Assistant based on Unified Vision-Language Model”. Technical Report, 2025.
  • Kaicheng Pang, Xingxing Zou, Waikeung Wong. “Learning visual body-shape aware embeddings for fashion compatibility”. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024.
  • Kaicheng Pang, Xingxing Zou, Fangjian Liao, Waikeung Wong. “M-VTON: Multi-layer Virtual Try-on System”. 12th International Conference on Design and Semantics of Form and Movement, DeSForM, 2023.
  • Kaicheng Pang, Xingxing Zou, Waikeung Wong. “Dress Well via Fashion Cognitive Learning”. The 33rd British Machine Vision Conference (BMVC), 2022.
  • Kaicheng Pang, Xingxing Zou, Waikeung Wong. “Modeling fashion compatibility with explanation by using bidirectional LSTM”. CVPR Workshops, 2021.
  • Dongmei Mo, Xingxing Zou, Kaicheng Pang, Waikeung Wong. “Towards private stylists via personalized compatibility learning”. Expert Systems with Applications, 2023.
  • Xingxing Zou, Kaicheng Pang, Wen Zhang, Waikeung Wong. “How good is aesthetic ability of a fashion model?”. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022.

Patents #

  • Method, apparatus, electronic device, and storage medium for intelligent clothing matching. Granted Country: China, Patent No.: CN202210470116, Granted Date: 2023.11.01.

Projects #

  • HKSAR Government “InnoHK Research Clusters”, AiDLab/RP1.1, AI-Powered Interactive Marketing and Customer Engagement Solution, 2025.07 - 2030.06, Participant. Responsible for market research and requirements analysis of cutting-edge AI technologies, and coordinated the development of core interactive systems.
  • HKSAR Government “InnoHK Research Clusters”, AiDLab/RP1.2, Designer-led AI system for Fashion Accessory and Lifestyle Product Design, 2025.07 - 2030.06, Participant. Orchestrated engineering architecture design and full-lifecycle development management of AI generative systems.
  • HKSAR Government “InnoHK Research Clusters”, AiDLab/RP3-1, Artificial Intelligence based fashion design assistant, 2020.07 - 2025.06, Participant. Led the overall system technical architecture design and the R&D and deployment of foundational core AI algorithms.
  • HKSAR Government “InnoHK Research Clusters”, AiDLab/RP3-2, Intelligent Fashion Aesthetic Evaluation System, 2020.07 - 2025.06, Participant. Independently implemented multimodal dataset construction, aesthetic evaluation model training, underlying algorithm engineering deployment, and end-to-end productization (Core topic of Ph.D. research).
  • Shanghai Pujiang Program (Shanghai Magnolia Talents Plan), 23PJ1404100, Deep Learning-based Metamaterial Design, 2023.10 - 2025.09, Participant. Assisted in deep learning model architecture design and algorithm training optimization in metamaterial scenarios.

Invited Talks #

  • International Academic Week on Fashion and Technology at Zhejiang Sci-Tech University, Keynote Presentation: The Evolution of Intelligent Stylist: From Perceptual Compatibility to Multimodal Generative Fashion, 2025. (Details | News Report)