DeepQueueNet combines discrete‑event and continuous simulation with deep neural networks to deliver highly accurate, generalizable, and GPU‑scalable network performance estimates at packet‑level granularity, outperforming existing DNN‑based estimators across diverse topologies and traffic scenarios.
This article introduces the PanoViT model, a vision‑transformer‑based approach for indoor layout estimation from panoramic images, covering its research background, architectural components, experimental results on public datasets, and step‑by‑step usage within ModelScope.
This article introduces neural style transfer, explains its underlying principles using VGG19 feature extraction, content and style loss definitions, and provides a complete PyTorch implementation with code for loading images, extracting features, computing Gram matrices, and optimizing the output image.
This article explains the background, core principles, preprocessing, feature extraction, matching algorithms, and practical application scenarios of face recognition search, and provides detailed reference implementations with Java and OpenCV code examples for building a complete system.
The Multi‑Granularity Attention Model (MGAM) improves group recommendation by extracting subset, group, and superset preferences through hierarchical attention and graph neural networks, fusing them via self‑attention, and achieves state‑of‑the‑art offline results and a 1.2% online CTR lift in Alibaba’s local‑life services.
An end‑to‑end AI platform integrates feature processing, model training, deployment, and decision orchestration across offline and online layers, leveraging automated pipelines such as AutoML (feature engineering, hyper‑parameter optimization, neural architecture search) built on Ray Tune and NNI, which have already boosted CTR in real‑world advertising and aim to make every user an algorithm engineer.
The paper presents Action Sensitivity Learning (ASL), a framework that models frame‑wise importance at both class‑level (via learnable Gaussian distributions) and instance‑level (using quality scores), integrates these weights into classification and regression losses, adds a contrastive InfoNCE term, and achieves state‑of‑the‑art temporal action localization performance across six benchmark datasets.
AI painting leverages deep‑learning models to analyze and recreate color palettes, enabling machines to mimic famous artists, generate novel hues, and express emotions, while also facing challenges such as limited creativity, human‑AI interaction, and fine‑detail rendering.
The paper introduces StEN, a spatiotemporal-enhanced network for CTR prediction in location-based services, combining static spatiotemporal feature activation, dynamic preference activation, and target attention, achieving state-of-the-art offline results and a 1.6% CTR lift in online tests.
Deep Learningclick-through raterecommendation systems
This article provides a comprehensive overview of OCR technology, detailing the evolution from traditional hand‑crafted methods to modern deep‑learning approaches, describing image preprocessing, text detection and recognition pipelines, summarizing classic machine‑learning algorithms, and presenting a practical OpenCV implementation with Python code.
This article provides a comprehensive overview of Optical Character Recognition (OCR), covering its definition, historical development, classification, real‑world applications, technical pipeline, common challenges, mitigation strategies, popular datasets, model performance comparisons, and leading open‑source platforms.
The paper introduces PEPNet, a plug‑and‑play network that tackles the domain‑seesaw and task‑seesaw problems in multi‑scenario recommendation by using a gated personalization module (GateNU) together with embedding‑level (EPNet) and parameter‑level (PPNet) personalization, and demonstrates its superiority through extensive offline and online experiments on Kuaishou data.
This article explains the challenges of processing sequential data, introduces LSTM as a solution to long‑term dependency problems in RNNs, details its cell state and gate mechanisms, showcases its architecture, and provides Python code examples for time‑series forecasting using Keras.
This article provides a comprehensive overview of deep neural network families, introducing twelve major architectures—including Feedforward, CNN, RNN, LSTM, DBN, GAN, Autoencoder, Residual, Capsule, Transformer, Attention, and Deep Reinforcement Learning—explaining their principles, structures, training methods, and offering Python/TensorFlow/PyTorch code examples.
The article defines the AI architect role and outlines three essential capabilities—mastery of AI technologies and development workflow, deep business understanding with strong abstraction ability, and the design and implementation of efficient, scalable AI solutions—explaining why each is critical for successful AI product delivery.
This tutorial demonstrates how to download the Fashion‑MNIST dataset, build a four‑layer fully connected neural network with PyTorch, and train it using loss functions, Adam optimizer, learning‑rate strategies, and Dropout to achieve high‑accuracy multi‑class image classification.
This article explains the Slide-Transformer paper, describing how the proposed Slide Attention replaces inefficient Im2Col‑based local attention with depthwise convolutions and a deformable shift module, achieving high efficiency, flexibility, and hardware‑agnostic performance for Vision Transformers.
This article provides a comprehensive tutorial on building the core components of a Transformer model—including multi‑head attention, layer normalization, feed‑forward networks, encoder and decoder layers—and assembles them into a complete PyTorch implementation, supplemented with explanatory diagrams and runnable code.
This article introduces ten widely used Python libraries—NumPy, Pandas, Matplotlib, Scikit‑learn, Requests, Beautiful Soup, Selenium, Pygame, Flask, and PyTorch—each accompanied by a concise real‑world project and complete code examples to help readers understand and apply them effectively.
This article provides an in‑depth, English‑language overview of Vision Transformer (ViT), covering its Transformer‑based architecture, patch‑to‑token conversion, token and position embeddings, fine‑tuning strategies such as 2‑D interpolation, experimental results versus CNNs, and the model’s broader significance for multimodal AI research.
This paper introduces HairStep, a novel intermediate representation combining Strand Maps and Depth Maps, and demonstrates how it reduces domain gap and improves single‑view 3D hair reconstruction accuracy across multiple algorithms, supported by new annotated datasets (HiSa, HiDa) and fair evaluation metrics.
This article explains the architecture, data layers, prediction modules, and deep‑learning framework behind Gaode’s driving ETA service, detailing how static speed profiles, linear models, and the H‑STGCN model combine to forecast travel times and evaluate their accuracy.
This article introduces Alibaba Cloud's PAI‑DLC, a cloud‑native deep learning platform that integrates machine‑learning capabilities, containerized services, AI‑aware scheduling, GPU virtualization, elastic training with EasyScale, data access, and observability, and discusses its architecture, key features, and future directions.
This article explores the rapid evolution of Baidu's semantic search models, the large GPU consumption they entail, and how extensive INT8 quantization, sensitivity analysis, calibration data augmentation, hyper‑parameter auto‑tuning, and advanced methods like Quantization‑Aware Training and SmoothQuant dramatically improve inference performance while preserving business metrics.
ShaderNN 2.0 is a lightweight mobile deep learning inference engine supporting OpenGL and Vulkan, offering texture‑based zero‑copy I/O, hybrid shader implementation, and achieving significant latency and power reductions versus TensorFlow Lite and MNN, thereby enabling real‑time graphics‑AI tasks such as style transfer, denoising, super‑sampling, and Stable Diffusion on smartphones.
To eliminate data fragmentation, feature inconsistencies, and multi‑language implementation challenges, Bilibili built a unified real‑time advertising feature platform that aligns offline, hourly, and online pipelines via a shared C++ library and JNI, boosting CTR prediction accuracy, cutting training costs, and increasing ad revenue by over 1 %.
Amid tightening financial regulations, ICBC's software team proposes a multimodal AI anti‑fraud framework that combines image, video, and structured data to detect deep‑fake, mask, and forged‑document attacks, enriches verification with cross‑modal cues, and outlines future expansion to text and speech modalities.
This article explains the concept of image‑based search, why it matters, and provides a step‑by‑step guide to implement image search using Elasticsearch 8.x, feature‑extraction libraries, and the multilingual CLIP‑ViT‑B‑32 model, including code snippets and architecture overview.
GoldMiner, a new system from Alibaba Cloud’s PAI platform, elastically scales deep learning data pre‑processing pipelines, dramatically improving training performance up to 12.1× and GPU cluster utilization by 2.5×, and its underlying research was accepted at SIGMOD 2023.
At ICASSP 2023, Kuaishou’s audio team presented a low‑latency hierarchical fusion network for full‑band acoustic echo cancellation, detailing its multi‑stage design, asymmetric windowing, loss functions, training strategy, and achieving second place in the non‑personalized AEC Challenge, with real‑world deployment results.
In a Stanford eCorner interview, Ilya Sutskever, chief scientist of OpenAI and architect of GPT, reflects on the history of large language models, compares AI to human cognition, debates open‑source versus closed‑source models, discusses regulation, training strategies, and offers guidance for aspiring AI researchers.
This article introduces the Paraformer model released by Alibaba DAMO Academy on ModelScope, detailing its non‑autoregressive architecture, training strategies, performance on benchmark datasets, and step‑by‑step guidance for fine‑tuning and deploying the model using FunASR and ModelScope pipelines.
The article chronicles the career of Chinese AI pioneer Huang Xuedong, detailing his education, rise at Microsoft, leadership of Azure AI, groundbreaking human‑level speech recognition breakthroughs, the engineering feats behind them—including a ten‑network model and the CNTK framework—and his recent move to Zoom.
In 2023 the Network Intelligent Research Center celebrated its largest PhD graduating class—seven scholars whose dissertations span deep‑vision hand‑gesture estimation, multi‑scenario network transmission, graph alignment, interactive streaming, knowledge‑defined networking, wireless body‑area networking, and more—showcasing significant AI‑driven advances and high‑impact publications.
CF‑Font introduces a content‑fusion module that linearly mixes base‑font content features using a font‑level distance metric, combined with iterative style refinement and a projection character loss, achieving state‑of‑the‑art few‑shot Chinese font generation that outperforms prior methods by over 5% on L1 and FID and is already used to create proprietary Alibaba‑Mama fonts.
Deep Learningcontent fusionfew-shot font generation
This article reviews the evolution of face detection, covering the Widely‑Face dataset, major research directions such as feature fusion, label assignment, auxiliary supervision, anchor‑free methods, NAS‑based designs, summarizes key papers from S3FD to MogFace, introduces ModelScope implementations, and outlines future challenges and opportunities.
This article introduces Huawei's Domain‑Specific Ascend AI architecture, detailing its four‑layer full‑stack design, the five‑layer abstract and three‑layer logical structures of the CANN heterogeneous computing framework, and the AscendCL programming interface with its advantages and application scenarios.
This guide details the end‑to‑end process of installing GPU drivers, setting up a Python environment, deploying the open‑source Alpaca‑LoRA model, fine‑tuning it with Chinese data on a multi‑GPU server, and performing inference, while highlighting practical challenges and performance observations.
This article provides a comprehensive analysis of the Transformer model behind ChatGPT, covering its origin, core mechanisms such as embedding, positional encoding, self‑attention, multi‑head attention, a step‑by‑step translation example, and the broader implications for AI research and industry.
This article explains what ChatGPT is, how it builds on natural-language-processing and the Transformer-based GPT series, details its model-size growth, architectural enhancements, multilingual support, and walks through the tokenization-to-generation pipeline that enables coherent AI-driven conversations.
This article presents Weibo's comprehensive multimedia content understanding pipeline, covering video summarization techniques, quality assessment models, OCR advancements, video embedding strategies, and the performance benefits of CV‑CUDA acceleration, while highlighting real‑world applications and engineering trade‑offs.
AdaSparse introduces an adaptive sparse network that learns a dedicated sub‑network for each advertising scenario, balancing shared and specific knowledge while keeping computational cost low, and achieves +4.63% CTR and -3.82% CPC improvements in Alibaba’s external ad system, as validated on both public and massive production datasets.
The paper introduces SASS, a scenario-adaptive self-supervised recommendation model that uses contrastive pre-training and multi-layer gating to expand global samples and transfer scene-aware parameters, enabling a single model to deliver personalized recommendations across diverse Taobao ‘SuoSuo’ scenarios while mitigating data sparsity and cross-domain challenges.
This article explains the Transformer model—from its encoder‑decoder structure and self‑attention mechanism to multi‑head attention, positional encoding, residual connections, training loss, and inference strategies—providing a clear, visual walkthrough for readers new to modern NLP architectures.
This article describes the design, algorithmic improvements, and engineering performance optimizations of Bilibili's massive video duplicate detection (collision) system, covering challenges of low‑edit‑degree reposts, two‑stage retrieval, self‑supervised feature extraction, GPU‑accelerated preprocessing, and the resulting gains in accuracy and throughput.
Richard Sutton’s “The Bitter Lesson” argues that over the past seven decades the most powerful driver of AI progress has been general‑purpose compute and large‑scale search, which consistently surpasses methods that rely on human‑engineered knowledge across domains such as chess, Go, speech recognition, and computer vision.
This tutorial walks through the complete process of fine‑tuning a Vision Transformer (ViT) model for 14‑class flower image classification on ModelScope, covering dataset preparation, model loading, training configuration, evaluation, and inference with practical code examples.
Bilibili built a massive video‑duplicate detection platform that trains a self‑supervised ResNet‑50 feature extractor, removes black borders, and uses a two‑stage ANN‑plus‑segment‑level matching pipeline accelerated by custom GPU decoding and inference, boosting duplicate rejection 7.5×, recall 3.75×, and cutting manual misses from 65 to 5 per day.
Deep LearningGPU accelerationLarge-Scale Retrieval
This article introduces OneFlow Coop, a memory‑optimization technique that jointly optimizes dynamic‑graph recomputation strategies and GPU memory allocation by analyzing existing DTR limitations, proposing recomputable in‑place, op‑guided tensor allocation, and layout‑aware eviction modules, and demonstrating superior experimental results.
In this talk, AI expert Wu Jun explains why ChatGPT has caused widespread fear, traces the historical development of language models from the 1970s to today, clarifies the massive computational and data requirements, and discusses the real impact and opportunities of large‑scale AI systems.
This article explores how ChatGPT’s remarkable abilities stem from the Transformer architecture, reinforcement learning from human feedback, and the insights presented in the fourth edition of "Artificial Intelligence: A Modern Approach," highlighting key AI milestones and technical foundations.
This article introduces the basic architectures of recurrent neural networks (RNN), LSTM, and GRU, explains PyTorch APIs such as nn.RNN, nn.LSTM, nn.GRU, details their parameters, demonstrates code examples for building and testing these models, and provides practical insights for deep learning practitioners.
In a GTC fireside chat, Nvidia CEO Jensen Huang and OpenAI co‑founder Ilya Sutskever discuss GPT‑4's multimodal advances, the evolution of deep learning from early neural networks to large‑scale models, the pivotal role of GPUs and datasets like ImageNet, and their vision for more reliable, scalable artificial intelligence.
This article chronicles Ilya Sutskever’s two‑decade rise from a young researcher to a leading figure in artificial intelligence, highlighting his early mentorship, breakthroughs in image recognition, language translation, the founding of OpenAI, and the development of GPT and DALL‑E models.
This article provides a comprehensive overview of deep UPLIFT models, covering their fundamentals, key technical challenges such as confounding bias and inductive bias, the evolution of meta‑learner and deep architectures, and practical case studies in financial technology marketing.
Deep LearningMarketing OptimizationUplift Modeling
The article provides a detailed overview of PyTorch 2.0, highlighting its stable and beta features such as torch.compile, accelerated transformers, MPS backend, new quantization support, and prototype parallelism tools, while emphasizing performance improvements for dynamic shapes, distributed training, and CPU/GPU inference.
At the GTC conference, the talk presents Alibaba Cloud’s heterogeneous computing platform and introduces the Open Deep Learning API (ODLA), then details how CUTLASS‑based operator fusion dramatically accelerates attention and MLP layers in large‑scale recommendation models, achieving multi‑fold performance gains in production.
PaddleDetection v2.6 expands the PP‑YOLOE family with rotating, small‑object, dense‑object, and ultra‑lightweight edge‑GPU models, upgrades PP‑Human and PP‑Vehicle toolboxes, releases semi‑supervised, few‑shot and distillation learning methods, adds numerous state‑of‑the‑art algorithms, and improves infrastructure with Python 3.10, EMA filtering and AdamW support.
This article provides a comprehensive overview of object detection, describing traditional sliding‑window approaches, deep‑learning based two‑stage and one‑stage models such as R‑CNN, Faster R‑CNN, YOLO series, and discusses current challenges, improvement directions, and future research trends in the field.
The article describes WeChat Games’ scalable recommendation system, detailing its four‑component architecture—offline ML platform, unified management, online DAG‑based engine, and peripheral services—along with a hybrid algorithm library, feature engineering, real‑time monitoring, and solutions that boost engagement across diverse game recommendation scenarios.
This article traces the development of artificial intelligence from its early conceptual roots and the 2012 deep‑learning breakthrough through the rise of self‑supervised large language models like BERT and GPT, explains ChatGPT’s architecture and RLHF training, and discusses its commercial impact and future prospects for fields such as life sciences.
An in‑depth overview of graph neural networks (GNNs) covering their basic concepts, historical development, core models, recent research frontiers, and diverse applications such as recommendation systems, computer vision, NLP, program analysis, and smart cities, based on the book “Fundamentals, Frontiers and Applications of GNNs.”
This article provides an overview of deep learning as a branch of machine learning and presents detailed comparative tables of popular deep‑learning software frameworks, covering creators, initial releases, licenses, platforms, programming languages, supported features such as CUDA, OpenMP, and model‑training capabilities.
Deep Learningartificial intelligencesoftware frameworks
PaddlePaddle 2.4 introduces 167 new APIs—including sparse computing (paddle.sparse), graph learning (paddle.geometric), and audio processing (paddle.audio) modules—enabling efficient sparse model training and inference, graph message‑passing, advanced audio feature extraction, plus fresh loss functions, tensor utilities, and expanded vision transforms.
This article examines how moving image pre‑ and post‑processing to GPU with NVIDIA's CV‑CUDA reduces software complexity, alleviates CPU bottlenecks, and delivers up to thirty‑fold throughput gains for computer‑vision workloads across training and inference pipelines.
This article compiles practical deep learning training and hyperparameter tuning advice from Zhihu contributors, covering model debugging, learning‑rate strategies, optimizer choices, data preprocessing, regularization techniques, initialization methods, common pitfalls, recommended research papers, and ensemble approaches.
Make-An-Audio, a collaborative text‑to‑audio model from Zhejiang University, Peking University and Volcano Speech, uses a Distill‑then‑Reprogram strategy to generate high‑quality, controllable sound effects from any modality, showcasing impressive demos and promising future AIGC applications.
This article explains how moving image preprocessing and post‑processing to GPU with the open‑source CV‑CUDA library dramatically reduces system complexity, eliminates CPU‑GPU bottlenecks, and delivers up to thirty‑fold performance gains for computer‑vision workloads across training and inference stages.
This article presents a log anomaly detection approach that leverages NLP techniques such as Part‑of‑Speech tagging and Named Entity Recognition combined with deep neural networks, detailing a six‑step model, experimental validation on three datasets, and superior performance compared with existing DeepLog and LogClass methods.
A recent paper by Prof. Ma Yi and Turing‑Award winner Yann LeCun introduces a simple, interpretable unsupervised learning approach that combines sparse coding, manifold learning, and slow feature analysis, achieving near‑state‑of‑the‑art performance on MNIST, CIFAR‑10, and CIFAR‑100 without data augmentation or extensive hyper‑parameter tuning.
After attending the DataFun causal inference summit, this article outlines why causal analysis matters, its typical use cases, practical challenges, its relationship with A/B testing, and how it integrates with machine learning and deep learning to improve decision‑making and model robustness.
This article introduces deep learning as a powerful AI technique, explains its core algorithms—including supervised, unsupervised, and semi‑supervised approaches—and provides concrete examples such as CNN, RNN, autoencoders, GAN, self‑supervised and transfer learning, illustrated with visual demos.
The paper presents CapOnImage, a novel image‑on‑image captioning task that generates location‑specific decorative text for product images, introduces the 2.1‑million‑image CapOnImage2M dataset, and proposes a mixed‑modality transformer with position‑aware pre‑training and progressive training, achieving superior accuracy and diversity and already deployed in Alibaba’s advertising platforms for measurable business impact.
Tencent AI Lab deputy Yu Dong was named a 2022 ACM Fellow for his groundbreaking work in speech processing and deep‑learning applications, boasting over 100 patents, multiple best‑paper awards, and technologies now embedded in many of Tencent’s products.
The article reviews the rapid rise of AI‑generated content (AIGC) in 2022, outlines its historical development stages, highlights Baidu’s breakthrough video‑generation and search technologies, and discusses the severe AI talent shortage in China along with Baidu’s extensive training initiatives.
Experts Wang Shenshe and Chen Jing discuss how deep‑learning‑based video coding is reshaping traditional compression by offering modest quality gains but facing theoretical, hardware, and standardization hurdles, while debating hybrid versus end‑to‑end designs, rate control, 3‑D support, and the balance between human‑centric perception and machine‑oriented efficiency.
The QQ Browser search relevance team describes a unified, billion‑scale architecture that combines a main and vertical subsystem, a pyramid‑shaped ranking pipeline (recall, coarse, fine), a dedicated GPU‑accelerated relevance service, and hybrid semantic‑matching models (dual‑tower, BERT, matrix fusion) evaluated with offline and online metrics to deliver accurate, fresh, and authoritative results for diverse content and long‑tail queries.
Deep LearningEvaluation Metricsinformation retrieval
This article reviews recent research on digital human creation, covering graph‑based facial portrait editing (fat‑slim adjustment, double‑chin removal, hair removal), a high‑quality eyelash segmentation dataset with the EyelashNet pipeline, and a deep‑learning framework for real‑time animation of loose clothing using virtual skeletons and RBF networks.
The paper introduces ABPN, an Adaptive Blend Pyramid Network that achieves precise, high‑quality skin retouching and garment wrinkle removal on 4K‑8K photos in real time by combining a context‑aware local retouching layer with a novel adaptive blend pyramid layer, addressing challenges of artifact‑free detail preservation and efficient high‑resolution processing.
Bilibili’s IEEE‑VCIP‑2022‑accepted two‑pass deep‑learning bitrate‑factor predictor achieves 98.8% accuracy with only 1.55 encoding passes, enabling constant‑quality segment encoding that reduces bitrate consumption by over 15% while maintaining visual quality in its narrow‑band HD transcoding pipeline.
This article provides an extensive overview of graph neural networks, tracing their evolution from early RNN‑based models to modern message‑passing frameworks, discussing complex graph types, diverse real‑world applications, scalability challenges, design spaces, training platforms, and promising research directions.
This article reviews the challenges of ultra‑large‑scale face recognition, presents existing solutions such as metric learning, PFC and VFC, and details the proposed FFC framework with dual loaders, ID groups, probe and gallery networks, plus experimental results showing its cost‑effective performance.
This article details the evolution of 58.com’s deep learning inference platform, describing the transition from the original SCF‑based architecture to a cloud‑native Istio gateway (architecture 2.0), and explains design choices, traffic‑management, adaptive rate‑limiting, observability, model pre‑warming, and performance improvements.
Adaptive Parameter Generation Network (APG) dynamically creates sample‑specific model parameters for click‑through‑rate prediction using low‑rank factorization, parameter sharing, and over‑parameterization, achieving up to 0.2% AUC improvement, 3% CTR lift, and up to 96.6% storage reduction with faster inference.
This article provides a thorough examination of computational advertising, covering the oCPM pricing model as a superset, classic system architecture, the evolution of core modules such as ad ranking, pacing, bidding, federated learning, calibration, and conversion‑delay handling, and concludes with career advice for algorithm engineers.
This article presents a comprehensive overview of multimodal pre‑training, describing its motivation, architecture choices, large‑scale Chinese image‑text dataset construction, training optimizations, performance benchmarks, downstream applications, and a Q&A session that highlights practical deployment considerations.
The article proposes a three-stage contrastive learning framework—pre‑training, fine‑tuning, and debiasing—to generate unbiased, fine‑grained image embeddings for mobile Taobao CTR prediction, achieving higher accuracy, fairness, and a 4‑5% CTR lift in large‑scale offline and online evaluations.
This article explains the evolution of Ant Group's xNN edge‑AI framework, detailing its four‑layer model‑optimization space, the lightweight modeling of version 1.0, and the transition to scalable modeling in version 2.0 to better exploit fragmented device compute resources.
Vivo tackles the complex problem of mixing heterogeneous content queues—ads, games, and organic items—in its information‑flow and app‑store by evolving from rule‑based weighting to Q‑learning and deep‑learning position models that respect product constraints, preserve ordering, and balance short‑term revenue with long‑term user experience, while planning deeper personalization and on‑device solutions.
The Adaptive Domain Interest Network (ADIN) introduces a shared backbone with scenario‑specific subnetworks, domain‑specific batch normalization and SE‑Block attention to capture both commonalities and divergences across recommendation scenarios, and, combined with self‑supervised training, consistently outperforms baselines, delivering a 1.8% revenue lift in Alibaba’s display‑ad platform and now runs in production.
Deep LearningSelf-Supervised Learningdomain adaptation
This article summarizes the motivations behind causal inference, its typical business applications such as intelligent decision‑making and prediction, the practical challenges of validation and data, and its relationship with A/B testing, machine learning, and deep learning, providing a concise overview for newcomers.
This article explains the architecture of a video search ranking system that combines a deep encoding network, multi‑task expert networks, and a bias‑correction module to jointly optimize relevance, click‑through rate, and watch time for streaming platforms.
This article presents a comprehensive overview of a personalized news recommendation system that leverages knowledge graphs to improve accuracy, explainability, and user satisfaction, detailing background motivations, graph construction methods, model architecture, experimental results, and practical insights from a Meituan research perspective.
An extensive survey ranks and compares a wide range of machine learning libraries and frameworks—both deep and shallow learning—detailing their languages, types, GPU acceleration, distributed computing capabilities, and typical academic and industrial applications, based on Google search popularity as of May.
This article explains the motivation behind Spectral Normalization for stabilizing GAN training, derives its mathematical foundation, shows how it enforces the Lipschitz constraint, and provides a complete PyTorch implementation with detailed code explanations.
This article introduces JD Daojia's machine learning platform, detailing its architecture, implementation principles, and practical applications in various business scenarios, achieving significant improvements in recommendation and search systems.
This article introduces the EasyModeling framework built on Alibaba Cloud's PAI platform, detailing its modular design, high reusability, integration with deep‑learning libraries, automated hyper‑parameter tuning, deployment scenarios, and a real‑world case study using RoBERTa for dish‑name standardization, demonstrating significant performance gains.
This article examines the challenges of deploying dozens of deep‑learning models at Zuoyebang and compares three serving architectures—Gunicorn + Flask + Transformers, Tornado + PyTorch, and Tornado + Triton—highlighting performance trade‑offs and presenting a final high‑concurrency, low‑latency solution in production.
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.
STARDOM is an end‑to‑end deep hierarchical forecasting model that jointly learns hierarchical constraints, query semantics via pretrained BERT, and a calibration matrix within an encoder‑decoder architecture, using a distilled reconciliation loss and hierarchical sampling to accurately predict large‑scale search traffic and outperform state‑of‑the‑art baselines.
Deep Learninghierarchical modelingsearch advertising