Page 288 - AI for Good Innovate for Impact
P. 288

AI for Good Innovate for Impact



                      4�3�        Category 3: 5G





                          Use Case 1: AI Full-Process Low-Code Middle Platform Based on

                      the ‘1+1+3+N’ Architecture











                      Organization: China Telecom HongXin Information Technology Co.,Ltd

                      Country: China

                      Contact Person(s): 
                           Shaoshuai Che, cheshaoshuai.js@ chinatelecom .cn


                      1      Use Case Summary Table


                       Item            Details
                       Category        5G

                                       The widespread application of artificial intelligence remains limited due to
                       Problem         high thresholds, including compute power, infrastructure, technical exper-
                       Addressed       tise, data quality, and tool complexity, which hinder its integration across
                                       industries and sectors.

                                       A complete Low-Code Middle Platform Based on the ‘1+1+3+N’ Architecture
                                       is developed to support the full lifecycle of AI model production using the
                                       following components
                                       An integrated platform for annotation, training, and inference, covering the
                       Key  Aspects  of  full lifecycle workflow of AI model production
                       Solution
                                       MaaS model services covering images, text, and videos,
                                       Three AI tool platforms include a large model and agent aggregation plat-
                                       form, a visual AI algorithm platform, voice customer service dialog platform.
                                       The solution supports both localized deployment & cloud deployment

                       Technology      All-target recognition, multimodal parsing, multimodal retrieval, multimodal
                       Keywords        Q&A, big model, cold start

                       Data Availability  Not available

                       Metadata (Type   Text data, image data, video data
                       of Data)

                                       Model training techniques: small sample robust techniques, GPU accelera-
                       Model Training   tion techniques, smart labeling, cold-start and multi-stage training;
                       and Fine-Tuning
                                       Base models: DeepSeek, Qwen 2.5, Star Big Model, etc.







                  252
   283   284   285   286   287   288   289   290   291   292   293