1 views collected around this technical thread.
This article shares Kuaishou’s real‑time, massive‑scale model data processing pipeline, covering model scenarios, recommendation workflow complexity, large‑scale data storage, streaming joins, feature computation, NVM‑based storage solutions, strong consistency mechanisms, and future outlook for AI recommendation systems.
This article presents Baidu's exploration of multimodal content understanding for commercial advertising, detailing the ViCAN pre‑training model, its contrastive and mask‑language learning tasks, integration across recall, ranking and risk‑control pipelines, quantization with MMDict, and future AIGC‑driven generation, all backed by extensive experiments and Q&A.
Tencent’s Angel machine‑learning platform, recognized with the 2023 China Institute of Electronics Science & Technology Award, showcases breakthrough distributed training, high‑efficiency caching, adaptive sampling, multimodal fusion, and graph‑model search technologies that dramatically improve large‑scale AI model performance and cost.
iQIYI’s ad ranking system replaces static, hash‑based embeddings with TFRA dynamic embeddings to efficiently handle massive sparse ID features, eliminates collisions and I/O bottlenecks, isolates memory during hot model swaps, enabling billion‑parameter models that boost revenue by 4.3 % while planning adaptive embedding sizes for future improvements.
This article presents Kuaishou's end‑to‑end engineering solutions for handling massive, real‑time recommendation model data, covering scenario description, complex business pipelines, trillion‑parameter model storage, high‑throughput processing with Flink and NVM, and future directions for cloud‑native scalability.
This article reviews the evolution of Weibo's large‑scale recommendation technology, covering the system's business scenarios, technical roadmap, recent large model iterations, multi‑task and multi‑scenario modeling, feature engineering, consistency between recall and ranking, and emerging techniques such as causal inference and graph methods.
The article presents Alibaba DAMO Academy's large‑scale table pretraining model SPACE‑T, explaining the background and trends of TableQA and Text‑to‑SQL, detailing the model’s design and training data, showcasing its deployment on ModelScope and Alibaba Cloud, and outlining future directions and practical impact.
The award‑winning VLDB 2022 paper introduces HET, a cache‑enabled distributed framework that dramatically reduces communication overhead for sparse trillion‑parameter embedding models, and Tencent Ads has industrialized this technology to train 10 TB‑scale models with up to 7×24‑hour online deep learning.
PaddleBox is Baidu’s GPU‑based ultra‑large‑scale sparse DNN training framework that combines a three‑tier hierarchical parameter server (SSD, DRAM, HBM) with pipelined scheduling and multi‑machine multi‑GPU communication, delivering 5–40× cost‑performance gains over traditional CPU solutions and powering Baidu’s advertising services.
This article reviews the latest advances in open-domain dialogue systems, covering classification, end‑to‑end generation challenges, knowledge‑controlled generation, automated evaluation, large‑scale latent‑space models such as PLATO, and outlines future research directions for building more coherent and controllable conversational AI.