Responsibilities
Team introduction: The platform governance algorithm team was established in September 2020. Its main responsibility is to optimize the algorithm and cooperate with the e-commerce business team to conduct comprehensive quality/ecological management of ByteDance e-commerce products, including combating low-quality/risk/violation issues, as well as style optimization, high-quality content support, e-commerce ecological construction, etc. The mission of the platform governance algorithm team: guaranteed operation and delivery, and everything you see and get is good Vision: to create a safe and compliant platform order and establish a healthy ecosystem trusted by users. 1. Responsible for the optimization and iteration of computer vision-related models in e-commerce scenarios, including fine-grained classification of product images, product object recognition, product subject recognition, feature extraction, logo detection, brand recognition, etc., optimizing the process of adding and removing merchant products 2. Responsible for video classification, multi-modal content mining, and multi-modal content understanding of e-commerce short videos and live broadcasts, optimizing the e-commerce short video shopping experience 3. Responsible for algorithm optimization and iteration of e-commerce scene image search, product search by photo, and product deduplication 4. Explore cutting-edge computer vision technologies and be responsible for the iteration and evolution of overall algorithms and systems.
Qualifications
1. Have in-depth research in a certain field of multimedia and computer vision, including but not limited to: image search, image/video classification and recognition, image segmentation, target detection, OCR, graph neural network, multi-modal, unsupervised and self-supervised learning, etc. 2. Familiar with TensorFlow / PyTorch / Training and deployment of one or more framework models in MXNet, and understanding of training acceleration methods such as mixed precision training and distributed training 3. Familiar with the latest research and technical progress of model compression acceleration, including but not limited to model quantification, pruning, knowledge distillation, etc., as well as inference frameworks such as TensorRT 4. Have strong practical ability, and winners in Kaggle, COCO, ImageNet, ActivityNet, ICPC, NOI/IOI and other competitions are preferred 5. Understand AutoML related algorithms, and have experience in network structure or hyperparameter automatic search is preferred 6. Familiar with big data related frameworks and applications such as MR/Spark are preferred 7. Candidates with strong practical skills and have published papers in relevant competitions and top academic conferences (such as CVPR, ICCV, ECCV, etc.) are preferred.
