This article explains how reinforcement‑learning (Q‑learning) can generate mock interface data for regression testing, how Puppeteer automates UI interactions, and how a DSL‑plus‑runtime approach enables scalable pop‑up components, reducing testing costs in complex e‑commerce interactions.
This article details the end‑to‑end technical architecture of WeChat's "Look" personalized recommendation service, covering data collection, recall, multi‑stage ranking, various CTR and multi‑objective models, reinforcement‑learning based mixing, diversity optimization, and the engineering hurdles overcome to deploy these solutions at massive scale.
This article introduces mainstream budget‑pacing methods for cost‑per‑click online ads, describes the 58.com business scenarios, details the pacing algorithm—including bid modification, probabilistic throttling, and reinforcement‑learning approaches—explains system design with PID control, and presents online experimental results and future directions.
This article explains how the Taobao front‑end team tackled the complexity of the 618 interactive game by using reinforcement‑learning‑driven intelligent testing, Puppeteer‑based automated regression, and a decoupled UI‑logic architecture for scalable popup production, dramatically improving development efficiency and stability.
This article introduces recommender systems and the exploitation‑exploration dilemma, explains common E&E algorithms such as ε‑greedy, Upper‑Confidence‑Bound, and Thompson Sampling, and details their practical deployment for interest‑point eviction, selection, and adaptive recall count optimization in an automotive recommendation platform.
The article reviews recent research on representing massive sparse features in click‑through‑rate (CTR) models, introducing Alibaba's Res‑embedding method and Google's Neural Input Search (NIS) approach, and discusses how these techniques improve embedding efficiency and model generalization in large‑scale recommendation systems.
JD announced that two of its research papers—one on a hierarchical reinforcement‑learning framework for multi‑objective recommendation (MaHRL) and another on an attention‑based model for delayed‑feedback conversion‑rate prediction (pCVR)—were accepted as full papers at the prestigious IJCAI2020 and SIGIR2020 conferences, highlighting the company's strong AI capabilities.
This article details the evolution of AliExpress product ranking from simple DNN scoring to advanced reinforcement‑learning re‑ranking, comparing multiple models, exploring context effects, introducing pointer‑network generators, evaluating various RL algorithms, and reporting significant online gains in conversion and GMV.
This article examines the limitations of traditional new media marketing, explains reinforcement learning fundamentals, and presents a six‑step technical solution—including problem modeling, algorithm selection, action, state, reward design, and model training—that uses RL to optimize budget allocation and achieve over 35% improvement in campaign effectiveness while reducing costs.
The article examines the limitations of the single‑machine APEX framework in the tf2rl reinforcement‑learning library, proposes a cross‑machine distributed architecture using middleware such as Redis, compares alternative frameworks like EasyRL, and outlines expected performance gains and future development plans.
In this article, Wang Zhe addresses fifteen common questions about recommendation systems, covering topics such as building cross‑domain knowledge, the role of deep reinforcement learning, handling sparse or low‑sample data, offline‑online evaluation, knowledge graphs, graph neural networks, model interpretability, large‑scale ID embedding, and career advice for engineers.
The article explains how data products create business value through application, data, and algorithm layers, using examples like 5G infrared temperature screening and ImageNet, and discusses the roles of experimental design, causal inference, and reinforcement learning in building effective AI‑driven strategies.
This article explains how Alibaba leveraged AI‑powered voice robots to handle massive outbound call volumes during COVID‑19, detailing the technology stack, real‑world application scenarios across finance and retail, and the future potential of intelligent voice assistants in customer service.
This article reviews the evolution of dialogue management models from rule‑based systems to deep‑learning approaches, identifies three major challenges—poor scalability, limited annotated data, and low training efficiency—and surveys recent research solutions including semantic matching, knowledge distillation, hierarchical reinforcement learning, model‑based RL, and human‑in‑the‑loop methods.
The article explains why Didi needs advanced dispatch algorithms, describes the complexities of order‑driver matching from simple one‑to‑one cases to large‑scale bipartite matching, and introduces batch matching, supply‑demand prediction, chain dispatch, and AI‑driven optimizations that together improve global efficiency and user experience.
This article surveys neural architecture search, classifying existing methods, describing common search spaces—including global and cell‑based designs—detailing optimization strategies such as reinforcement learning, evolutionary algorithms, surrogate models, one‑shot and differentiable approaches, and highlighting recent results and trends in the field.
This article details Zhihu's search engineering advances over the past year, covering long‑tail query challenges, term‑weight calculation, synonym expansion, query rewriting with translation models and reinforcement learning, and semantic retrieval using BERT‑based embeddings, while outlining future research directions.
This article provides an in-depth overview of sequential recommendation, defining the problem, discussing data preparation, and reviewing various neural architectures—including MLP, CNN, RNN, Temporal CNN, self‑attention, and reinforcement‑learning approaches—while offering practical guidance on model selection and implementation.
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 describes how Didi Ride‑Sharing leverages reinforcement learning and graph‑embedding techniques to model and optimize user‑operation marketing, detailing system architecture, algorithm design, experimental ROI improvements, and personalized message delivery for enhanced conversion and cost efficiency.
Professor Song Le presented at the Cloudwise Conference how Ant Financial leverages large‑scale graph neural networks, automated machine‑learning platforms, and multi‑agent reinforcement learning to model complex financial networks, improve risk control, and drive diverse fintech applications.
This article reviews recent progress and open research problems in human‑machine dialogue, covering both open‑domain chat and task‑oriented systems, with focus on reply quality, decoding, retrieval‑augmented generation, controllable and personalized responses, multi‑turn modeling, reinforcement‑learning strategies, low‑resource NLU, and data augmentation techniques.
The project replaces cumbersome manual management of thousands of heterogeneous CPU and GPU nodes for large‑scale reinforcement learning with a TKE‑based, containerized actor‑learner architecture that automates batch start/stop, provides elastic autoscaling, fault‑tolerant processes, shared model storage, and CI‑driven image deployment, cutting costs by up to two‑thirds while dramatically speeding experiment cycles.
This article reviews recent advances in applying reinforcement learning to recommendation systems, explains the fundamental RL concepts, discusses the specific challenges such as large action spaces, bias, and long‑term reward modeling, and summarizes two influential YouTube papers along with practical insights and future directions.
This article presents Alibaba's Xiaomai intelligent dialogue platform, detailing its general system architecture, three-tier NLU approaches for zero‑, few‑, and many‑shot scenarios, platform‑centric dialogue management with TaskFlow, robustness and continuous learning mechanisms, and a user simulator for large‑scale data generation and dialogue diagnosis.
This article explains why modern ride‑hailing platforms need advanced dispatch algorithms, describes the underlying order‑allocation problem, explores simple and complex matching scenarios, and introduces batch matching, supply‑demand prediction, chain dispatch, and AI‑driven techniques used by Didi to improve efficiency and fairness.
Didi’s ride‑hailing dispatch system has progressed from a simple greedy, first‑come‑first‑served matcher to sophisticated batch, chain, and predictive algorithms that use deep‑learning demand forecasts and reinforcement‑learning optimization to assign drivers under complex business rules, boosting response rates and serving over 30 million daily requests.
This article introduces the Exact‑K recommendation problem, highlights its differences from traditional Top‑K approaches, and presents a novel solution combining Graph Attention Networks (GAttN) with Reinforcement Learning from Demonstrations (RLfD), backed by extensive experiments showing superior performance on real-world datasets.
From Steve Jobs’s early work at Atari to modern DeepMind breakthroughs, the article traces how video games have grown into a multibillion‑dollar industry that serves as a testbed for AI research, while highlighting current AI techniques for smarter agents, procedural content generation, and the collaborative challenges shaping the future of game development.
This article discusses the fundamental traits of recommendation technologies, compares UserCF and ItemCF models, explains matrix factorization and FM, explores negative sampling, CTR/CVR modeling, ensemble methods, and practical considerations such as reinforcement learning and exploration strategies for improving recommendation performance in real-world systems.
At SIGIR 2019 in Paris, Alipay presented two AI research papers—one applying reinforcement learning to predict user intent in customer‑service bots and another introducing the unsupervised QUEST method that builds noisy quasi‑knowledge graphs for answering complex multi‑document questions.
RESA is a real‑time evaluation platform that continuously tests multiple Adaptive Bitrate (ABR) algorithms on live user traffic, introduces a multi‑user QoE metric derived from viewing behavior, reveals trade‑offs between clarity and bandwidth, and proposes the RL‑based ABSbc algorithm to steer bitrate distribution and balance user experience with network cost.
This paper presents a sequence‑to‑sequence model that fuses product‑detail tables with customer comments, using gated multimodal attention, copy mechanisms and reinforcement learning to automatically produce high‑quality, context‑aware replies for e‑commerce platforms, and validates the approach with extensive experiments on a large Taobao dataset.
This article examines the limitations of traditional learning‑to‑rank for Ctrip hotel sorting, introduces reinforcement learning as a remedy, outlines three progressive implementation plans (A, B, C) with algorithm choices and engineering trade‑offs, and presents early experimental findings that demonstrate RL's potential to improve conversion rates.
This article presents a model‑based reinforcement learning framework for recommendation systems that uses a generative adversarial user model to simultaneously learn user behavior dynamics and reward functions, enabling efficient Cascading‑DQN policy learning and achieving superior long‑term user rewards and click‑through rates in experiments.
Alibaba’s senior engineers share their curated list of essential technical books—from software testing and design patterns to AI, machine learning, reinforcement learning, Rust programming, and database architecture—explaining why each title is valuable for developers seeking deeper knowledge and practical insights.
The article presents an in‑depth overview of Alibaba's intelligent service bot Ali Xiaomì, covering its platform evolution, core NLP techniques such as intent recognition and machine reading comprehension, multi‑turn recommendation strategies, transfer‑learning approaches across domains and languages, and future technical challenges.
Tencent’s industrial practice shows how a large‑scale offline‑nearline‑online “Shield” recommendation architecture, powered by the DeepR framework built on RCaffe, uses deep semantic embeddings, massive neural networks and reinforcement‑learning decisions to handle billions of daily requests, demonstrating that data richness and engineering capability, not model depth alone, drive performance in big‑data recommendation systems.
This article details Alibaba.com engineers' application of reinforcement learning, attention mechanisms, and weakly supervised techniques to extract product summaries, improve content quality, and significantly raise SEO rankings, supported by offline experiments, online A/B testing, and future research directions.
This article introduces reinforcement learning fundamentals, contrasts it with supervised learning, and explores its challenges and advantages in natural language processing, including applications such as text classification, relation extraction from noisy data, and weakly supervised topic segmentation, while summarizing key insights and experimental results.
The article explains the four main machine‑learning categories—supervised, unsupervised, semi‑supervised, and reinforcement learning—detailing how each approach handles data labeling, typical algorithms, and common application scenarios.
This article explains how reinforcement learning can be modeled and deployed to enable intelligent, interactive product search on e‑commerce platforms, detailing problem definition, system architecture, training methodology, online results, and future research directions.
This article explains the principles of adaptive bitrate streaming, compares traditional ABR algorithms with a reinforcement‑learning‑based approach, describes its system architecture and training process, and presents QoS evaluation results that show RL‑driven streaming can improve video quality and smoothness.
This article traces the ten‑year evolution of Alibaba’s e‑commerce search system, detailing four major stages—from the early Pora streaming engine to dual‑link real‑time architectures, the integration of deep and reinforcement learning, and the shift to large‑scale online deep learning—while highlighting the technical drivers and future AI‑enabled search vision.
Meituan enhanced its homepage “Guess You Like” recommendation slot by modeling user‑item interactions as a Markov Decision Process and applying an improved DDPG reinforcement‑learning agent that adjusts the ranking trade‑off parameter, uses advantage‑based Q decomposition, shares actor‑critic weights, and runs in a real‑time TensorFlow pipeline, delivering consistent lifts in click‑through, dwell time, and depth.
This article outlines the essential role of a data scientist and introduces ten fundamental machine‑learning algorithms—including PCA/SVD, OLS and polynomial regression, regularized linear models, K‑Means, logistic regression, SVM, feed‑forward, convolutional and recurrent neural networks, CRFs, ensemble trees, and reinforcement‑learning methods—while linking to popular Python libraries and tutorials.
This article explains how reinforcement learning, specifically the DDPG algorithm combined with Transformer-based networks, is applied to improve large‑scale news recall systems, detailing the business scenario, algorithm selection, model architecture, speed optimizations, training challenges, and observed online performance gains.
The article presents a comprehensive overview of how Shumei Technology leverages machine learning—including supervised, unsupervised, and reinforcement learning methods—across its credit scoring, fraud detection, advertising, and audio content moderation services, highlighting practical challenges, model fusion techniques, and future research directions.
The article details JD's third‑generation autonomous delivery robots, covering their multi‑sensor fusion localization, deep‑learning perception, reinforcement‑learning motion control, extensive patent portfolio, and upcoming technical hurdles such as high‑precision mapping and lidar cost, while also inviting public voting for patent awards.
This article reviews reinforcement learning fundamentals, contrasts it with supervised learning, surveys major RL algorithms such as DDPG and DQN, and details how these methods can be modeled for sequential news recommendation, including system architecture, state‑action definitions, and practical challenges.
This article analyzes OpenAI’s Retro Contest on Sonic the Hedgehog, explains why reinforcement learning generalization is crucial for AGI, and details the winning team’s joint PPO pipeline, engineering optimizations, training strategies, and final performance compared to human baselines.
PhoenixGo, an open‑source Go AI built on AlphaGo Zero's reinforcement‑learning algorithm, leveraged Tencent's idle cloud servers to achieve professional‑level play, won the 2018 World AI Go Championship, and was released with a strong model for researchers and hobbyists alike.
On World Book Day, nine Alibaba technology veterans share a curated list of essential technical books—covering software testing, design patterns, AI, machine learning, reinforcement learning, Rust, and database architecture—offering concise reasons why each title is valuable for developers and engineers.
The article surveys three recent approaches—Transition State Clustering, Sequential Windowed Inverse Reinforcement Learning, and Deep Discovery of Continuous Options—that automatically segment long‑horizon surgical‑robot demonstrations into sub‑tasks, learn hierarchical policies from limited data, and achieve markedly higher success rates on da Vinci cutting, tension, and needle‑picking tasks.
In 2017 Alibaba's AliMe chatbot platform expanded from a single‑company solution to a multilingual, multi‑channel AI service, introducing platform‑level SaaS/PaaS capabilities, a seven‑layer front‑end architecture, modular back‑end design, advanced intent recognition, knowledge‑graph‑driven product management, reinforcement‑learning‑based recommendation, and machine‑reading comprehension for enterprise and consumer use cases.
This article traces the evolution of computer chess from the 18th‑century automaton "The Turk" through early programs by Turing, Shannon, and McCarthy, to landmark systems like Deep Blue, AlphaGo, and AlphaZero, highlighting key algorithms, milestones, and their impact on AI research.
This article explains how deep reinforcement learning, built upon traditional Q‑learning and enhanced with techniques like experience replay, enables agents to play Atari video games directly from raw pixel inputs, illustrating the key differences, processing steps, and the significance of this breakthrough in AI.
This article introduces the fundamental concepts of reinforcement learning, describing its origins, key components such as agents, environments, states, actions, and rewards, explaining the Markov decision process framework, and highlighting common algorithms like Q‑learning, policy gradients, and actor‑critic methods.
This article presents a comprehensive overview of Alibaba’s intelligent chatbot “Ali Xiaomi”, covering industry context, e‑commerce deployment, NLU architecture, intent‑matching layers, deep‑learning‑based intent classification, reinforcement‑learning‑driven recommendation, knowledge‑graph‑enhanced services, and hybrid retrieval‑generation dialogue models, with future outlooks for AI‑driven interaction.
Salesforce’s recent research introduces Seq2SQL, a reinforcement‑learning‑driven sequence‑to‑sequence model that translates natural‑language questions into SQL queries, eliminating the need to learn SQL, and includes the large WikiSQL dataset built from crowdsourced NL‑SQL pairs for training and evaluation.
The article summarizes Dr. Qi Yuan’s presentation at CCAI 2017, detailing Ant Financial’s AI‑driven solutions for financial services—including risk control, intelligent assistants, large‑scale machine learning, reinforcement‑learning marketing, a model‑service platform, and a computer‑vision damage‑assessment system—while highlighting technical challenges, platform architecture, and the company’s open‑tech philosophy.
This article explains how Alibaba's Cainiao network built the STARK VRP engine—an AI‑driven, distributed vehicle‑routing solver that supports dozens of VRP variants, leverages metaheuristics, parallel island models, and deep reinforcement learning to dramatically reduce fleet size and travel distance in Chinese logistics.
The article explores how machine learning models emulate human learning by converting diverse real‑world descriptions into numerical features, illustrating concepts such as one‑hot encoding, supervised, unsupervised, and reinforcement learning, and emphasizing the importance of mapping inputs to outputs for intelligent systems.
The paper introduces LSTM‑Jump, a reinforcement‑learning‑trained LSTM variant that can dynamically skip irrelevant tokens, achieving up to six‑fold speed‑ups over standard sequential LSTMs while maintaining or improving accuracy on various NLP tasks such as sentiment analysis, document classification, and question answering.
Alibaba’s e‑commerce search engine evolved over three years from offline batch models to a sophisticated AI-driven system that integrates real‑time feature ingestion, online learning, deep and reinforcement learning, enabling dynamic personalization and decision‑making that boosts conversion during high‑traffic events like Double 11.
StarCraft, a classic real‑time strategy game, has become a premier testbed for deep reinforcement learning and AI research, and Alibaba’s open‑source Gym StarCraft platform now bridges Python, TensorFlow, Keras and OpenAI Gym to simplify multi‑agent, macro‑strategy development and fair benchmarking.
This article explains Monte Carlo Tree Search (MCTS), covering its origin in AlphaGo, fundamental algorithm steps, node‑selection strategies such as UCB, strengths and weaknesses, enhancements, historical background, and recent research developments in artificial intelligence.
This article breaks down AlphaGo’s breakthrough architecture—four specialized neural‑network modules, Monte‑Carlo Tree Search, and deep reinforcement learning—to explain how the system moved from imitation learning to self‑improvement and ultimately defeated top human Go players.
This NSF‑funded project aims to develop algorithms that incrementally process partially observed data, integrating generative models with reinforcement‑learning policies to decide when to act, applied to simultaneous machine translation and quiz‑bowl style question answering.
After decades of being deemed a 'century‑long' AI challenge, Baidu’s Bingo system achieved amateur‑to‑professional level Go play by introducing optimized Monte‑Carlo tree search, a weakened Alpha‑Beta hybrid, and massive supervised learning, demonstrating how breakthroughs in game AI can ripple into broader Baidu products.
