Machine Learning Engineer

• May 2017 - Present

Tech Lead on machine learning modeling and infrastructure task force (from 06/2018). Designed and implemented machine learning pipeline, introduced ML modeling best practices into the team. All those efforts hugely shortened the average period of model incubation. Built Content-based image retrieval platform to further support specific business. Combined Object Detection, Embedding Feature vectors training, and Approximate Nearest Neighbors. Optimized platform performance with batch inference, static graph transform, parameters quantization, etc. Implemented algorithms from popular papers, like ‘lottery ticket hypothesis’, ‘one cycle policy’, etc. Built abnormal behavior detection models on video frame records of 4A Security Audit to catch operators violating rules. 中文: 建模和基础能力专项组Lead. 领导一个5人小组, 设计并实施机器学习建模pipeline, 引入并定期更新建模最 佳实践. 着力于缩短模型孵化平均时间, 提升效能 参与公司京凯项目(防欺诈), 负责判定用户上传的人脸照片是否合规. 构建基于表征/嵌入(embedding)的图像/人脸检索匹配平台，以进一步支持特定的业务需求(是否本人, 是否黑名单用户等)。平台采用并结合目标识别，嵌入特征向量训练和近似KNN等多种技术. 上线半年, 有效识别用户近百人 参与公司‘提升经营质量, 量化经营指标’攻坚, 作为模型Lead, 主导判别线下营业厅服务质量、经营效率的量化工作. 依托营业厅监控视频, 通过目标检测判断营业厅人流量情况. 通过批处理推断，静态图变换，参数量化(Parameters Quantization)等优化生产平台性能. 在内部数据集上复现若干经典论文, 如‘ResNet’, ‘Faster-rcnn’, ‘U-Net’, ‘lottery ticket hypothesis’等. 在4A安全审核的视频帧记录上标记并训练异常行为检测模型，帮助捕获定位违反规则行为. Show less

Machine Learning Engineer

• Sep 2015 - Mar 2017

Senior Software Engineer

• Aug 2011 - Aug 2015

Software Engineer

• Jul 2006 - Jul 2011