Villeurbanne, Auvergne-Rhône-Alpes, France
熟练掌握Python，并在深度学习、大语言模型与智能问答、向量检索、自监督学习以及图神经网络等核心人工智能技术领域拥有深厚的专业积累。我的研究主攻方向为结构化文档自监督学习与面向知识密集场景的对话与问答系统；在博士期间，我主导设计了分层文档检索方法与多文档问答生成管线 ，成功在真实业务场景中交付了基于大模型驱动的智能问答助手原型，相关研究成果已在ACL和EMNLP等顶级学术会议上发表。结合我在Worldline与美亚柏科的实战经验，我不仅具备从零搭建模型训练与评估流程的端到端开发能力，还拥有在移动端部署Deepfake检测模型以及性能调优大规模向量检索系统的丰富工程经验，并能流利使用中、英、法三语胜任跨团队的技术沟通与业务对接。
• Worldline wants to evolve its artificial intelligence engine to be able to integrate various new technologies based on self-supervised learning (SSL). We will explore SSL solutions on structured data and use for e-commerce orders from customers in the retail sector.
• This is a collaborative project where I am working as an employee at worldline while taking my PhD studies with Labo Eric of Lyon 2.
• Study deep learning algorithms around Deep Fake • Propose implementations of Deep Fake videos with respect to use case • Optimize Deep Fake detection models and deploy models to mobile devices with resource constraints • Develop a demo for use cases
• Use python to perform a performance test on an open source vector index library Faiss, which is deployed in the online face recognition engine. • Learn to use Milvus, a vector index library developed based on Faiss, and study whether Milvus can be used to replace Faiss. • Investigate the existing monocular face silent anti-spoof technology, and try to reproduce the paper.