This tutorial walks university students through the complete workflow of using the CIFAR‑10 dataset to train a convolutional neural network for image classification and then deploying the model as an online inference service on the UAI‑Train and UAI‑Inference platforms.
This article details a series of reinforcement‑learning experiments on the 2048 game, from random baselines through DQN implementations, classical value‑iteration methods, network redesigns, and Monte‑Carlo tree search, highlighting challenges such as reward design, over‑estimation, and exploration while achieving scores up to 34 000 and tiles of 2048.
This article presents JTM, a joint optimization framework that simultaneously learns a tree‑based index and a deep scoring model to overcome the limitations of traditional recommendation pipelines, demonstrating significant recall improvements on Amazon Books and Alibaba UserBehavior datasets through hierarchical user interest modeling and efficient tree learning.
This article surveys two neural news recommendation approaches—NAML, which uses multi‑view learning to fuse heterogeneous news information, and NPA, which incorporates personalized attention for both words and news items—demonstrating their superior performance over strong baselines on real‑world MSN news data through extensive experiments and visual analyses.
The ICCV 2019 Lightweight Face Recognition Challenge attracted 292 teams and defined four strict FLOP‑ and size‑limited protocols for image and video recognition, with champions employing near‑30 GFLOP EfficientNet‑style backbones, novel loss functions, frame‑fusion, and knowledge‑distilled VarGNet models to balance accuracy and computational budget.
StarNet is a neural network for pedestrian trajectory prediction in large‑scale delivery, using a global dynamic map and a Hub‑Host architecture to model interactions efficiently, reducing complexity from O(N²) to O(N), and achieving higher accuracy with fast inference compared to baseline methods.
This article describes the end‑to‑end development of a multi‑class NLP system for Ctrip airline ticket customer service, covering problem analysis, data preprocessing, sample balancing, model architecture (TextCNN and Bi‑GRU), training strategies, performance evaluation, and online customization to achieve high accuracy in intent recognition.
North Korea’s newly unveiled “Blue Sky” smartphone incorporates a homegrown AI facial‑recognition system built on CNNs, MTCNN, MobileFaceNets and TensorFlow, showcasing how the isolated nation is advancing edge AI despite operating solely on its internal CentOS‑based intranet.
This article explores how Alibaba’s 3D scene‑based recommendation system combines computer‑vision, deep‑learning and data‑driven matching algorithms to create immersive, size‑accurate product visualizations that enhance user experience and drive higher click‑through rates in e‑commerce.
AMap’s ground‑marking recognition has progressed from simple threshold methods to advanced deep‑learning pipelines—including two‑stage R‑FCN, cascade detectors with local regression, corner‑point and segmentation hybrids, and LiDAR‑based 3‑D PointRCNN—achieving over 99 % recall and sub‑5 cm positional accuracy for high‑precision map production.
The article details JD.com's internally developed distributed search engine and recommendation platform, their new architectures, deep‑learning‑driven ranking and recall models, component‑based deployment, extensive performance testing, and coordinated operations that powered the massive 11.11 shopping event.
The Baidu App team showcases how deep‑learning and AI can be deployed on mobile through on‑device and server‑side inference—illustrated by plant‑identification, stylized filters, video subject detection, and AR real‑time translation—while addressing model compression, cross‑platform optimization, and offering a practical guide for engineers.
This article details the 360 display advertising system's architecture and the progressive evolution of its recall module, covering business overview, overall pipeline, various recall strategies—including Boolean, vectorized, and deep‑tree approaches—and the performance optimizations applied to meet real‑time constraints.
The article reviews AR navigation lane‑detection, comparing traditional handcrafted visual pipelines with modern deep‑learning segmentation approaches, proposes an efficient multitask network with weight‑allocation and vanishing‑point anchoring, and demonstrates quantized models achieving real‑time, stable performance on low‑power automotive chips while outlining remaining weather, lighting, and road‑condition challenges.
The talk outlines how Tencent Advertising leverages deep‑learning visual algorithms—including GCN‑based edge refinement, template generation, AutoML‑driven smart review, and a dual‑tower click‑through‑rate model—to automate creative production, improve ad quality, and enhance user experience across creation, review, and playback stages.
This article details Meitu's large‑scale personalized recommendation pipeline, describing the business scenario, challenges of massive data, latency and long‑tail distribution, and the application of deep learning techniques such as Item2vec, YouTubeNet, dual‑tower DNN, NFM, NFwFM and multi‑task learning to improve click‑through rate, conversion and user engagement.
This article describes 58.com’s deep‑learning‑based advertisement image recognition platform, covering its background, system architecture, QR‑code detection, multi‑scale ResNet classification, category fusion, performance metrics, real‑world case studies, and online service statistics.
This article presents a comprehensive overview of short‑video recommendation at UC, covering business background, system architecture, the evolution from LR to Wide & Deep models, multi‑objective loss design with positive‑sample weighting, graph‑embedding fusion, time‑weighted loss, continuity modeling, a Boosting‑based WnD solution, and future research directions.
This article details 58.com’s recruitment recommendation platform, describing its personalized matching challenges, typical recommendation scenarios, a three‑stage ranking framework, optimization goals, the evolution from rule‑based methods to logistic regression, factorization machines, XGBoost, and deep learning models, extensive feature engineering practices, and future research directions.
The paper chronicles Meituan’s three‑stage evolution of delivery‑time estimation—from a hierarchical address tree with local linear regression, through a vector‑retrieval system that boosts recall, to a lightweight end‑to‑end deep‑learning model that meets sub‑5 ms latency while delivering progressively lower error and full coverage.
This article introduces JTM, a joint optimization framework that simultaneously learns deep scoring models and tree-structured indexes, addressing the limitations of traditional recommendation pipelines and demonstrating significant precision and recall gains on large-scale datasets such as Amazon Books and UserBehavior.
This article presents a comprehensive overview of UC's information‑flow video tag recognition technology, detailing tag usage scenarios, the end‑to‑end system architecture, multi‑modal feature extraction, advanced deep‑learning models such as NextVlad, behavior and person tagging methods, and future research directions.
The talk by Meitu senior algorithm expert Chen Wenqiang details how deep‑learning‑driven embedding, recall, and ranking techniques—including Item2vec, twin‑tower DNNs, and multi‑task NFwFM—are applied to improve click‑through rates, follow conversions, and user engagement in Meitu's content community.
The paper presents an efficient scene‑text detector that expands high‑level SSD feature maps and integrates a feature‑pyramid network, using direction‑aware segment‑and‑link predictions to reconstruct arbitrarily long, rotated text, achieving higher recall and precision with real‑time speed and outperforming recent methods on ICDAR benchmarks and a menu‑recognition test.
Learn how to turn an AI robot into a gallery‑ready painter by setting up Docker and nvidia‑docker, installing GPU drivers, and running a neural‑style transfer pipeline that blends photographs with Monet’s brushstrokes, producing high‑quality artwork in minutes.
The article explains how the WeiChe app leverages AI to instantly recognize car dashboard warning lights, describes the underlying deep‑learning infrastructure on UCloud’s UAI‑Inference platform, and shows how the Hot‑Standby feature dramatically cuts GPU costs while maintaining real‑time performance.
This study presents a deep‑learning pipeline that employs a Cascade‑RCNN with Feature Pyramid Network to automatically detect and refine modules and their internal elements on Taobao’s 99‑promotion pages, achieving roughly 98 % precision and recall on a thousand‑image validation set and paving the way for broader e‑commerce event applications.
Training deep neural networks on large‑scale weakly labeled image data suffers from noisy annotations that degrade performance, but a simple algorithm that adjusts the learning‑rate during training can automatically identify up to 90% of noisy samples, improving dataset cleanliness and model accuracy without manual intervention.
This article systematically reviews the history, core concepts, variants, and practical implementations of attention mechanisms—from early additive and multiplicative forms to self‑attention, multi‑head attention, and recent transformer‑based models—highlighting why attention has become fundamental in modern AI research.
This article reviews Baidu's research on semantic computing, covering the evolution of semantic representation, the development and evaluation of the ERNIE and SimNet models, their industrial applications, model compression techniques, and outlines future research priorities in multilingual and multimodal semantic understanding.
This article explains how to use NVIDIA TensorRT to accelerate TensorFlow model inference by describing TensorRT architecture, optimization techniques such as layer fusion and precision calibration, detailing the conversion of frozen_graph and saved_model formats, presenting experimental setup and performance comparisons, and summarizing the achieved speed‑up.
This guide explains how the Real-Time-Voice-Cloning project uses deep‑learning text‑to‑speech techniques to generate a voice clone from a short audio sample, covering the underlying principle, required dataset, setup steps, demo usage, and ethical considerations.
This article describes a systematic, standardized, and automated “4+N” recommendation framework that unifies features, samples, models, and pipelines to accelerate private‑domain marketing recommendations across multiple scenarios while improving accuracy, efficiency, and business impact.
This article explains how Huajiao Live leverages Spark for data preprocessing and TensorFlow (augmented by the TensorSlow project) for distributed deep‑learning training, detailing computational‑graph concepts, forward and backward propagation, loss construction, gradient‑descent optimization, and a step‑by‑step Python implementation of a multi‑layer perceptron.
This article presents a Transformer‑based framework that simultaneously generates visually compatible outfits and personalizes recommendations by leveraging multimodal item embeddings and user behavior, achieving significant gains in compatibility prediction, fill‑in‑the‑blank accuracy, and click‑through rate on Alibaba's iFashion platform.
This article provides a comprehensive technical overview of collaborative filtering, covering traditional user‑ and item‑based similarity methods, matrix‑factorization approaches for implicit feedback, various loss functions, and a suite of deep neural network models such as GMF, MLP, NeuMF, DMF, and ConvMF, together with implementation details, evaluation metrics, and practical deployment considerations.
This tutorial walks readers through the fundamentals of artificial neural networks, perceptrons, logistic regression, activation and loss functions, gradient descent, and provides end‑to‑end Python implementations using NumPy, TensorFlow, and Keras to build and evaluate a cat‑vs‑non‑cat classifier, complete with code snippets, visual explanations, and performance analysis.
This article explains the design and evaluation of RE2, a lightweight yet expressive text‑matching framework for FAQ‑style chatbots, detailing its five‑layer architecture, block‑wise residual connections, experimental results on SNLI, MultiNLI, SciTail, Quora and WikiQA datasets, and its significant performance improvements in Alibaba’s DingXiaoMi service.
Didi’s Elastic Inference Service (EIS) and its IFX AI acceleration engine provide a distributed, cost‑effective inference platform that automatically scales resources based on QPS and latency requirements, supports major deep‑learning frameworks, excels in public‑cloud, private‑cloud, IoT and edge scenarios, and achieved top‑rank DAWNBench latency and cost scores on ImageNet with P4 GPUs.
The article explains why Google created Swift for TensorFlow, highlights Swift's strong backing, built‑in automatic differentiation, high performance comparable to C, seamless interoperability with Python, C and C++, low‑level hardware access, and its future role within the MLIR compiler ecosystem for deep learning.
This article introduces Auto Risk, a deep‑learning risk model for behavior‑sequence data that leverages unsupervised pre‑training with proxy tasks, details its convolution‑attention encoder, demonstrates significant gains across multiple business scenarios, and highlights its strong small‑sample and analogy capabilities.
This article presents a co-designed algorithm‑system solution—MIMN and an independent UIC module—that enables ultra‑long user behavior modeling for click‑through rate prediction, delivering significant offline AUC gains and online CTR/RPM improvements in Alibaba's display advertising platform.
Tencent Cloud’s AIoT solution combines edge AI processing with a cloud‑edge collaboration framework, using container‑orchestrated microservices, AI chips and IoT connectivity to cut latency to milliseconds, lower bandwidth by sending only structured data, and enable real‑time applications such as smart retail, manufacturing, agriculture and building security.
This article introduces Auto Risk, a behavior‑sequence deep‑learning framework that uses unsupervised pre‑training with proxy tasks to learn universal feature representations from massive unlabeled data, achieving significant gains in risk‑control scenarios, improving AUC, supporting multi‑scene generalization and small‑sample learning.
The article outlines AI’s three historical waves—search and reasoning, expert systems, and machine‑learning/deep‑learning—detailing their core technologies, achievements, and limitations, while emphasizing how past cycles inform today’s narrow AI advances and the renewed relevance of computing power and data‑driven methods.
At CVPR 2019 in Long Beach, Suning’s AI team presented a breakthrough paper on multi‑scale body‑part mask guided attention for person re‑identification, detailing the conference’s selectivity, the challenges of re‑identification, and how their deep‑learning approach achieves state‑of‑the‑art performance.
Gaode Maps uses a two-stage machine‑learning pipeline (Faster‑RCNN with shape‑based region proposal networks and fine‑grained classifiers) to detect hundreds of traffic‑sign types in billions of street‑view images, achieving high recall and precision, scalable updates, and near‑real‑time map data refresh.
This article systematically reviews the evolution of attention mechanisms—from early additive and multiplicative forms to self‑attention and multi‑head variants—explaining their core three‑step framework, key differences, and why they have become essential across NLP, vision, and broader AI applications.
After a three‑month contest co‑hosted by iQIYI and ACM MM, 255 teams competed on the challenging iQIYI‑VID‑2019 multimodal dataset, and the BUPT Automation School team won with a 91.14% person‑recognition accuracy, advancing the field and enhancing iQIYI’s video recommendation and AI services.
This article explains how TensorFlow powers Huajiao Live's recommendation system, introduces the TensorSlow project for demystifying TensorFlow's core, and walks through deep‑learning fundamentals, computational‑graph concepts, forward and backward propagation, loss construction, gradient‑descent optimization, and building a multi‑layer perceptron with Python code examples.
This article introduces deep learning applications at Huajiao Live, explains TensorFlow's computational graph architecture, details core concepts such as placeholders, variables, operations, forward and backward propagation, and provides complete Python-like code examples for building and training a multi-layer perceptron.
The Xianyu team’s new “black‑tech” system automatically transforms UI mockup images into production‑ready code by detecting components with deep‑learning models, extracting layouts via OpenCV, and employing a modular stream‑oriented architecture of units, tasks, and server layers that enables rapid testing, flexible composition, and future enhancements such as improved container recognition and semantic understanding.
This article introduces Alibaba's new e‑book collection of five ICASSP‑accepted papers that showcase advances in speech recognition, synthesis, and emotion detection, detailing novel models like DFSMN, A‑LSTM, and speaker‑adaptation techniques that dramatically improve speed, size, and accuracy.
By integrating YOLOv3 video object detection into Youku’s ad platform, the team replaced traditional subtitle‑based and scene‑based placements with precise object‑level targeting, achieving higher relevance, expanded inventory, and a 20% click‑through increase despite 3.5× higher exposure.
Alibaba engineers detail the PAISoar distributed training framework, showing how RDMA‑optimized hardware, Ring AllReduce algorithms, and user‑friendly APIs boost deep‑learning models—like the GreenNet CNN—to 101‑fold speedups on 128 GPUs, dramatically reducing training time from days to under a day.
Tencent Cloud OCR leverages deep‑learning‑based text detection and recognition, including Compact Inception and multi‑layer RNN refinements, to overcome challenges such as complex backgrounds, low resolution, and multilingual layouts, delivering over 90% accuracy for ID cards, bank cards, business licenses, handwritten text, and powering fast, cost‑saving applications in logistics, QQ, and WeChat Work.
This article reviews the historical development of natural language processing language models, covering expert rule‑based systems, statistical n‑grams, smoothing techniques, neural network models such as NNLM, RNN, word2vec, GloVe, ELMo, and the transformer‑based breakthroughs of GPT, BERT and GPT‑2, and summarizes their impact on modern NLP tasks.
The Youku Video Enhancement and Super‑Resolution Challenge invites teams to develop models that restore low‑resolution, noisy video to high‑definition quality using a 10,000‑pair industry dataset, offering up to RMB 100,000 in prizes and a recruitment pathway, with registration open through June 16 and competition phases spanning May to September.
This article walks through the development of deep learning for natural language processing, starting with basic neural cells and shallow networks, then exploring CNNs, RNNs, LSTMs, TextCNN, ESIM, ELMo, and culminating with the Transformer‑based BERT model, its training objectives, fine‑tuning strategies, and performance comparisons.
This article introduces Grammatical Error Correction (GEC), explains its main error categories, outlines four key challenges, reviews evaluation metrics and the evolution of NLP approaches, and showcases practical solutions and product applications developed by Liulishuo.
At QCon Beijing 2019, four Beike technology experts presented the practical use and challenges of machine learning for user profiling, deep‑learning‑based house‑quality scoring, intelligent customer‑service systems, and AI‑driven floor‑plan generation, summarizing the architecture, data pipelines, model evolution, and future improvement directions.
This article explains the various roles of data scientists, compares data science with machine learning, deep learning, AI, statistics, IoT, operations research, and applied mathematics, and clarifies the distinctions and overlaps among these fields.
The Youku Video Enhancement and Super‑Resolution Competition challenges teams of up to five to develop 4× upscaling models that also remove noise and compression artifacts, using a 10,000‑pair dataset, with prizes up to ¥100,000 and recruitment opportunities, running from May to September 2019.
This article explains the historical development, definitions, and relationship between artificial intelligence and machine learning, highlighting key milestones, foundational concepts such as deep learning and neural networks, and their modern applications across industries.
This article walks beginners through the fundamentals of artificial intelligence, machine learning, and deep learning, using a real‑world attendance dataset to illustrate how to label data, construct a simple linear model, and expand it into a neural network for workday prediction.
This article explains how MXNet supports data‑parallel training on single‑machine multi‑GPU and multi‑machine multi‑GPU setups, describes KVStore modes, outlines the worker‑server‑scheduler architecture, and shows how to launch large‑scale distributed training using Kubeflow and the mxnet‑operator.
Alibaba’s open‑source MNN is a lightweight, high‑performance deep‑learning inference engine optimized for edge devices, supporting multiple model formats and backends, offering portability across iOS, Android, and IoT, with detailed architecture, performance benchmarks, roadmap, and real‑world application examples.
The article details Meituan's large‑scale O2O search platform, describing its current coverage, challenges such as heterogeneous POI data and user intent diversity, and the deep‑learning‑driven solutions—including intelligent matching, business recognition, component analysis, semantic models, real‑time features, and future personalization directions.
This article reviews why deep neural networks are over‑parameterized, outlines the challenges of deploying them on mobile and embedded devices, and presents six major strategies—pruning, low‑rank approximation, filter selection, quantization, knowledge distillation, and novel architecture design—to accelerate and compress models while preserving performance.
Ren Haibing’s Youku Cognitive Lab talk reviews object segmentation’s motivation, explains semantic and instance concepts, presents UNet‑based and category‑agnostic methods—including fast video segmentation with motion cues—and reports high IoU results while outlining future edge‑aware, label‑free, and non‑online video segmentation research directions.
This article explains the fundamental differences between CPUs and GPUs, outlines key GPU specifications such as CUDA cores, memory capacity, bandwidth, and floating‑point precision, and reviews NVIDIA's major GPU series and architectural evolution for high‑performance and AI workloads.
This article reviews the evolution of autonomous‑driving research, explains how to design end‑to‑end deep‑neural‑network models such as PilotNet, and outlines a reinforcement‑learning based decision system, highlighting key architectures, performance metrics, and future challenges.
This article provides a beginner-friendly overview of artificial intelligence, explaining its core concepts, the relationship between AI, machine learning and deep learning, common real-world applications such as search and recommendation, and practical steps and resources for newcomers to start learning AI with Python and basic statistics.
The article details a practical workflow for Chinese sentiment classification in Tencent’s Goose Man product, covering data preparation, word‑segmentation challenges, a six‑layer multi‑LSTM architecture with word embeddings, training results achieving roughly 96 % accuracy, and its deployment for automatic detection of misleading and high‑impact user reviews.
This article reviews the fundamentals of automatic speech recognition, compares traditional MFCC‑GMM‑HMM pipelines with modern deep neural network approaches such as DNN‑HMM, LSTM‑CTC, and attention‑based models, and illustrates each evolution step with flowchart diagrams and key references.
This article outlines the end‑to‑end text‑mining workflow—from data acquisition and preprocessing to feature extraction, algorithm selection, and model evaluation—while demonstrating a sentiment‑analysis case study that combines LDA topic modeling with deep‑learning classifiers.
This article reviews the evolution of click‑through‑rate (CTR) estimation models for recommendation ranking, covering logistic regression, feature‑engineering tricks, factorization machines, deep neural networks, wide‑and‑deep architectures, and the Deep & Cross Network, while discussing their strengths, limitations, and future research directions.
This article summarizes a computer‑vision salon where Dr. Ji Yongnan explains imaging pipelines, traditional feature‑based methods, deep‑learning breakthroughs, Tencent Cloud AI services, real‑world case studies, and answers audience questions about machine‑vision versus computer‑vision and data‑scarcity challenges.
This article explores how deep learning techniques such as RNN, LSTM, 3D‑CNN, and CNN‑RNN hybrids can be applied to predict network bandwidth and traffic, comparing traditional time‑series methods with modern AI approaches and highlighting the potential of graph neural networks for future improvements.
This article walks through creating a Python‑based neural network to predict breast cancer using the Wisconsin dataset, covering network architecture, weight and bias initialization, back‑propagation, gradient descent, and the role of activation functions such as sigmoid, tanh, ReLU and Leaky ReLU.
This article reviews Jia Yangqing’s insights on why deep learning finally succeeded—highlighting the roles of big data and high‑performance computing—while examining its current limitations, emerging challenges, and future opportunities across AI engineering, AutoML, and hardware‑software co‑design.
This article reviews the challenges of media‑domain named entity recognition, outlines the evolution from rule‑based methods through traditional machine‑learning and deep‑learning models to attention‑based Transformers, and details Sohu’s practical Bi‑LSTM‑CRF system with data‑annotation strategies and performance results.
This article explains how real‑time bidding works in computational advertising, defines the budget‑constrained bidding problem, models it with reinforcement learning, and presents a deep‑network implementation together with visual analysis and key references.
This article explains how to build both unsupervised and supervised deep‑learning models that compute item similarity from user behavior, covering prod2vec embeddings, skip‑gram architectures, loss function design, and practical training steps for modern recommender systems.
The AI Scientific Frontier Conference 2019, co‑hosted by the Chinese Academy of Sciences AI Alliance and Beijing Institute of Technology, gathers leading researchers to present cutting‑edge talks on AI theory, deep learning, computer vision, robotics, NLP, big data, and related applications, with detailed schedules, speaker bios, venue information, and registration details provided.
Natural language processing, a cornerstone of AI, relies on language models to capture linguistic features; this article reviews classic pre‑training models—ELMo, GPT, and BERT—explaining their architectures, training objectives, and how they boost downstream NLP tasks despite data‑scarcity challenges.
This article surveys the evolution of object detection models—comparing one‑stage and two‑stage approaches, their performance trade‑offs, and recent state‑of‑the‑art methods—while also outlining key concepts and breakthroughs in natural language processing, highlighting the impact of deep‑learning models such as BERT.
This article explains how the lightweight xNN-OCR engine achieves high accuracy and real‑time performance on mobile devices through deep‑learning model compression, novel detection and recognition techniques, and showcases its practical applications such as bank‑card, gas‑meter, license‑plate, and ID recognition.
This article interprets Alibaba’s “Image Matters” paper, explaining how the Deep Image CTR Model (DICM) introduces user‑side visual preference modeling with image embeddings, why traditional Parameter Servers struggle with large image vectors, and how the Advanced Model Server (AMS) compresses embeddings to enable efficient distributed training.
This article explores the evolution of Generative Adversarial Networks' objective functions, detailing the shift from Jensen‑Shannon divergence to f‑divergence, IPM‑based approaches, and auxiliary losses, while highlighting their impact on stability and performance across image, audio, and text generation tasks.
This article reviews the evolution of Taobao’s search and recommendation technology, tracing its journey from simple statistical models and rule‑based systems through large‑scale machine learning and real‑time online learning to modern deep‑learning and cognitive intelligence approaches that drive e‑commerce innovation.
This article introduces Variational Autoencoders (VAEs), compares them with GANs, explains the underlying variational inference principle, and details how VAEs are trained using the evidence lower bound, complemented by visual diagrams and key equations.
This comprehensive guide explores anomaly (outlier) detection across diverse methods—including time‑series analysis, statistical tests, distance metrics, matrix factorization, graph approaches, behavior‑sequence modeling, and supervised machine‑learning models—highlighting their principles, formulas, and practical use cases such as fraud prevention and system monitoring.
This article traces the evolution of convolutional neural networks from the pioneering AlexNet through VGG, Inception, ResNet, Inception‑v4, Inception‑ResNet and ResNeXt, highlighting architectural innovations, performance gains, and the underlying biological inspirations that shaped modern deep learning models.
This article explores how deep‑learning‑based image dewarping techniques, from traditional hardware methods to modern U‑Net, Stacked U‑Net and Dilated U‑Net architectures, can correct warped document photos, improve OCR accuracy, and support intelligent verification in high‑throughput business scenarios.
This article introduces Alibaba’s intelligent matting editor and its Semantic Human Matting (SHM) algorithm, detailing the integration of semantic segmentation and deep matting networks, the fusion module, training strategy, experimental results, and the deployment of an online fully‑automatic cutout tool for designers.
The UI2CODE system tackles UI layout analysis by first extracting backgrounds with Sobel, Laplacian and Canny edge detection plus a flood‑fill algorithm, then isolating foreground components through connected‑component analysis and a Faster R‑CNN classifier, and finally fusing both pipelines to achieve superior precision, recall and IoU on Xianyu app screenshots.
The article details the design and iterative improvement of 58.com’s intelligent QABot, covering knowledge‑base construction, feature engineering, three generations of classification models—including FastText, Bi‑LSTM, and deep semantic matching—and evaluation metrics that achieve high accuracy and automation rates.
This article explains how Alibaba’s algorithm team tackles the massive challenge of extracting high‑quality buyer‑generated images from Taobao by combining statistical features, GBDT classification, deep CNN visual assessment, aesthetic scoring, and dirty‑data filters to improve content moderation efficiency.
This article introduces TensorFlow, explains when neural networks are appropriate, outlines the general workflow for solving image‑based problems, and provides a step‑by‑step Python implementation of a multilayer perceptron that classifies handwritten digits, while also discussing TensorFlow's strengths, limitations, and alternatives.
