{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-09-22T20:00:59.977Z","headline":"IBM 与 Confluent 发布 Granite 时间序列模型 Early Access，可在流数据上实时预测与检测异常","description":"IBM 与 Confluent 宣布四款 IBM Granite 时间序列基础模型（PatchTST-FM、FlowState、TTM、TSPulse）以 Early Access 形式上线 Confluent Cloud。","url":"https://www.aioga.com/news/cmtk6zwl601osroqryptpbgai/","mainEntityOfPage":"https://www.aioga.com/news/cmtk6zwl601osroqryptpbgai/","datePublished":"2026-09-02T13:49:14.000Z","dateModified":"2026-09-02T13:49:14.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://huggingface.co/blog/ibm-research/real-time-intelligence","https://aihot.virxact.com/items/cmtk6zwl601osroqryptpbgai"],"canonicalUrl":"https://www.aioga.com/news/cmtk6zwl601osroqryptpbgai/","directAnswer":{"@type":"Answer","text":"IBM 与 Confluent 宣布四款 IBM Granite 时间序列基础模型以 Early Access 形式上线 Confluent Cloud，面向流数据提供预测、异常检测、相似历史检索与优化等能力。","url":"https://www.aioga.com/news/cmtk6zwl601osroqryptpbgai/","dateCreated":"2026-09-02T13:49:14.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"huggingface.co source article","url":"https://huggingface.co/blog/ibm-research/real-time-intelligence","datePublished":"2026-09-02T13:49:14.000Z","provider":{"@type":"Organization","name":"huggingface.co","url":"https://huggingface.co/blog/ibm-research/real-time-intelligence"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmtk6zwl601osroqryptpbgai","datePublished":"2026-09-02T13:49:14.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmtk6zwl601osroqryptpbgai"}}],"aggregationSource":"Hugging Face：Blog（RSS）","originalPublisher":{"name":"huggingface.co","url":"https://huggingface.co/blog/ibm-research/real-time-intelligence"},"geoDeepAnswer":null,"article":{"id":"cmtk6zwl601osroqryptpbgai","slug":"cmtk6zwl601osroqryptpbgai","url":"https://www.aioga.com/news/cmtk6zwl601osroqryptpbgai/","title":"IBM 与 Confluent 发布 Granite 时间序列模型 Early Access，可在流数据上实时预测与检测异常","title_en":"","summary":"IBM 与 Confluent 宣布四款 IBM Granite 时间序列基础模型（PatchTST-FM、FlowState、TTM、TSPulse）以 Early Access 形式上线 Confluent Cloud。","source":"Hugging Face：Blog（RSS）","sourceUrl":"https://huggingface.co/blog/ibm-research/real-time-intelligence","aiHotUrl":"https://aihot.virxact.com/items/cmtk6zwl601osroqryptpbgai","publishedAt":"2026-09-02T13:49:14.000Z","category":"行业动态","score":58,"selected":false,"articleBody":["Until now, those decisions have run on outdated economics: one bespoke model at a time and months of expert work on each. So teams model the few hundred series where the money is and cover the rest with safety margins, extra inventory, extra headroom, extra tolerance, acted on after the window has closed. That margin is the cost of a decision nobody could forecast, paid every cycle.","A time series foundation model (TSFM) changes that. Trained once across vast, varied signals, it generalizes to a series it has never seen: give it a window of measurements and it tells you what comes next, how far behaviour sits from normal, which history looks like this one, and which settings best serve a target. Using one does not take an army of data scientists either: a demand planner, a fraud analyst or a process engineer can put these models to work on their own streams. Around the models, IBM is building functions that shift the work left, so forecasting, anomaly detection, optimization and semantic intelligence arrive as capabilities you call rather than projects you build.","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/V5JlP6rnQZG8psqofXSf6.jpeg","Picture one tempering line in a chocolate factory, its temperature, speed and throughput sampled every few seconds and watched against fixed thresholds. Drop a foundation model into that stream and it forecasts the line's output through the evening shift, so the planner sees a shortfall while there is still time to act. It scores today's run against how the line normally behaves on dark chocolate, so a slow drift surfaces before a bar blooms. It finds the closest match in plant history, so the engineer knows how the last runs like it turned out. It conditions on the settings the crew controls, and fine-tunes when the last points of accuracy are worth it. No data science team required, and the same model rolls to every line in every factory.","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/vTTaugPEu3DaLg_SwVfZP.png","IBM ran these models before offering them, in its own products and operations first, then with design partners in cement, steel, pulp and paper, food and telecommunications. The numbers make the case: every point of accuracy is worth millions, productivity gains run 5 to 10×, and work that waited for specialists now sits with the domain experts who own the decision.","Now that proof meets real-time context: IBM brings frontier models that understand how signals behave, 44M+ downloads behind them, and Confluent brings the live state of the business and reach to every system that acts. Together they run stream-native, hosted in Confluent Cloud and called from Flink. Access opens on Confluent Cloud on AWS. Confluent Platform follows, bringing the same models and capabilities to on-premises and hybrid environments.","The months usually spent wiring a model into production are months you keep: Granite reads the signal, Confluent supplies the context, the governance and the delivery to everything downstream.","A signal's value decays with time: a pump caught drifting today is a work order, the same pump next week is an outage.","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/wlsMI5mra9TvaO7wA53CF.png","Forecasting and detection are stateful: the next value only means something against recent history, and an anomaly only exists against a running sense of normal. Flink manages that state, keyed per series and fault tolerant, so each model gets the history it needs without a separate data store or a database hit per call.","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/7szxD-qQx3PDPT_KxWOiC.jpeg","This is where the value compounds. Confluent's data streaming platform puts business data in motion and makes it usable for ML. The platform continuously streams, connects, governs, and processes real-time data, capturing live business signals that IBM Granite Time Series models use for forecasting, anomaly detection, similarity search, classification, gap-filling and optimization. Confluent provides what you need to implement streaming use cases quickly, reliably, and securely, so you can focus on developing real-time ML applications rather than managing data infrastructure.","Confluent Cloud, the cloud deployment of Confluent's data streaming platform, provides native inference, which allows you to run IBM Granite Time Series models directly within Apache Flink® on Confluent, providing greater flexibility, security, and cost efficiency for real-time data processing while unifying data and ML workflows. The benefits include:","By bridging operational and analytical estates, Confluent helps teams turn live business events into actionable intelligence, bringing IBM Granite Time Series models into the stream. And because no single model serves a shampoo line, a card network and a retail catalogue alike, IBM and Confluent offer a portfolio rather than a model.","Every decision asks the future a different question. A planning cycle needs a range of outcomes, a trading desk the most accurate number from data at every rate, a fleet of a hundred thousand series a cost that stays rational, and a security team the moment a stream stops behaving like itself and what happened last time it did. The portfolio is four complementary time series foundation models, all in Early Access and called through Confluent's existing AI_FORECAST and AI_DETECT_ANOMALIES Flink SQL functions. Switch models with one SQL parameter, no pipeline redesign.","Change the model value and the same call runs any of the four, with no separate ML stack to build or operate.","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/cPDlF0B1m3D9wB2cgGMh5.png","There is no single best model, so a few questions steer the choice. One series or thousands? One variable or many? Time to train, or out of the box? How far ahead? Forecasting, or anomalies? PatchTST-FM reads a series the way a language model reads text, patch by patch, each variable in its own channel so one noisy signal can't drag the rest down, and returns a full distribution, so a planner can set reorder points off the 90th percentile. FlowState keeps a running summary updated with every point, and because its dynamics are continuous in time it reads seconds-level SCADA and hourly market data alike. TTM drops attention for tiny mixing networks along time and across variables, so a million-parameter model covers a hundred thousand series nightly on CPU. And TSPulse pairs time and frequency views in one small multi-task model for anomaly detection, classification, gap-filling and the question every operator asks: have we seen this before.","Small was a decision, not a compromise: inference runs natively inside Confluent Cloud, or on your own CPUs with the open weights from the Hugging Face Hub, and no cloud ingress or egress keeps architecture simple and cost down. IBM Granite also brings IBM's enterprise AI governance framework, with model provenance and licensing transparency, and a growing set of functions on the models that make each use case more useful out of the box. This is where portfolio and platform come together, where a model stops being a librarian of the past and becomes an optimizer of decisions still in flight.","All four compress the time between an event and knowing about it. In the stream, that gap shrinks from days to seconds, and the signal becomes a trigger for other AI systems, agents and workflows that investigate, triage what matters, and loop in a human, context already gathered, when a decision needs a person. Each of the four is also being packaged as a function, so more of the work lives in the platform and less with the team using it.","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/OrlAwl1BVrDRC49ALrGCF.jpeg","Most forecasting is still a statistical model and gut feel dressed up as a decision. Bespoke ML has not closed the gap, one model per series, refit by hand and drifting with the horizon, so planning covers the handful of series worth the effort and the rest runs on safety stock.","Follow a demand planner at a grocery retailer, where planning reaches the head of the catalogue and never the tail, which sits on a shelf as working capital. She points one shared model at the whole catalogue: it works on an unseen series immediately, takes drivers like weather and promotions, and hands back a distribution rather than a single line. Nothing is built per SKU, so one model is a model factory: a two-year SKU and a three-month one in the same job, a launch with no history starting from the SKUs it resembles, 100,000 SKUs nightly on CPU, the same play across categories and regions.","The distribution turns service level into a policy she can state out loud. And because each forecast lands on a topic, it is a trigger rather than a report: replenishment fires from it, allocation and pricing read the same numbers, a markdown lands before stock ages and a reorder before the shelf empties. The results land where the business keeps score: fewer stockouts and markdowns, customers who find what they came for, revenue protected on the shelf, and freed working capital, usually the largest line in the business case.","Anomaly detection is the broadest lane, and one missed anomaly is rarely small: in fraud it is a customer's money, in security a breach, in IT operations an outage customers meet first, and each lands on the brand as hard as the balance sheet. In financial services, rule-based detection is enumerable, so adversaries enumerate it, and tightening a rule declines more honest customers, revenue gone and a customer half out the door. Bespoke ML wants labels that are scarce and stale, and false alarms cost more than the crime.","Picture the fraud lead at a retail bank: the model keeps a sense of normal per card and scores every payment through the same AI_DETECT_ANOMALIES call while it is still in flight. Because it also forecasts, it flags the drift toward trouble before the event. A card that bought groceries in the same three postcodes for two years funds a wallet abroad at 3am and the alarm fires before the money moves, while the same customer on an honest holiday sails through.","Protection starts on day one, because what the model learned elsewhere transfers to new products, corridors and asset types with no labelled case. It also has to keep moving, because the adversary does: fraud patterns and attack signatures change monthly, so the model is customized on the bank's own stream, refit as cases are confirmed, improved continuously rather than rebuilt yearly. And as banking and commerce turn agentic, with agents initiating payments at machine speed, the rhythm of normal shifts and volumes climb, so live context and in-flight scoring matter even more. Every score enables the next move: block the payment, escalate to an analyst with the closest past cases attached, hand it to an agent.","The same technique reapplies wherever an entity has a rhythm: IT latency, cell-site KPIs, the tempering line from the opening.","Every plant runs on a model of itself, and keeping it honest is hard: statistical models drift, rule-based control holds a setpoint but never improves it, and bespoke ML explains nothing, so optimization stays in pilots and the plant runs on margins.","Andrés runs process at a shampoo plant whose mixing line streams temperature, agitator speed, dosing rate and viscosity into Confluent. He puts a foundation model on the stream and it works out of the box: months of bespoke modelling become days, productivity gains of up to 10×, and he customizes only where the line demands it. Conditioned on what he controls, the forecast becomes a simulator: energy at this mixing speed, throughput at this temperature and dosing rate, viscosity in spec or not. An optimizer searches that space against a KPI he names, respects his constraints, and explains what it recommends, because a recommendation he cannot interrogate he will not act on. Andrés is a process engineer, not a modeller, and the person who knows the line steers it.","Optimization does not freeze at go-live, it re-optimizes as inputs change: a surfactant supplier switches, a fragrance batch behaves differently, demand shifts from the 400ml bottle to the travel size, and this quarter the objective is throughput rather than energy. Restate the objective and constraints, and the line runs on the next best setpoint rather than last year's, starting from the closest past runs and their fixes. The gains land in the currency his CFO tracks: one point on an operation turning over hundreds of millions is a seven-figure line, and one food manufacturer starts with one process and 400 factories behind it.","Production optimization anchors something larger: quality prediction and equipment condition come next, and as AI moves into manufacturing, robotics and physical systems, it is the first of many in a large, disruptive market.","Every lane above ends on the same question: have we seen this before? The models answer it with embeddings, compact vectors that capture the shape of a window in time and frequency, so two episodes that look alike land close together whatever their scale or offset. In the stream, each window is embedded as it arrives and matched against past episodes and their outcomes, so what comes back is a precedent, not a score: the runs that drifted this way and what fixed them, the demand curves a new SKU most resembles, the confirmed fraud cases this session rhymes with. The same embeddings drive classification and gap-filling, and index the context an agent retrieves before it acts.","This is only the beginning for time series. As with language models, the pace is accelerating: new architectures, new data sources, agentic and user experiences built on the models, and more of it ready to use on day one. IBM and Confluent will keep innovating the way they started, with design partners and clients on real enterprise use cases, for productivity, accuracy and responsiveness across every lane.","Available now in Early Access on Confluent Cloud: forecasting and anomaly detection directly on your data streams, with no model training, feature engineering or AI/ML expertise required. Confluent Cloud is the starting point, and Confluent Platform is next, so the same models and capabilities reach on-premises and hybrid estates. The feedback loop is the point: what you find on your own streams teaches the models what to become next."],"articleImages":[{"sourceUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64e8143f6de557454220921e/aNouKBrKm1dnN9pI1JXlp.jpeg","alt":"","afterParagraph":0,"url":"/media/articles/cmtk6zwl601osroqryptpbgai/f257760baa1a5507.webp"},{"sourceUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65b12ed52be9660f0b7e5f72/r230jdYzTFj_CYmKgpz78.jpeg","alt":"","afterParagraph":0,"url":"/media/articles/cmtk6zwl601osroqryptpbgai/de2554755ad470a5.webp"}],"mediaStatus":"ok","articleBodyZh":["直到现在，那些决策都是基于过时的经济学：一次一个定制模型，每个模型都需要专家数月的工作。因此，团队只对几百个有资金的系列进行建模，其余的则通过安全边际、额外库存、额外空间、额外容差来覆盖，这些都是在时间窗口结束后采取的措施。这个边际成本是无法预测决策的代价，每个周期都要支付。","时间序列基础模型（TSFM）改变了这一点。它经过一次在庞大且多样的信号上训练，就能推广到从未见过的系列：提供一个测量窗口，它会告诉你接下来会发生什么，行为偏离正常的程度，历史中哪些情况类似，以及哪些设置最能实现目标。使用一个模型也不需要一大群数据科学家：需求规划员、欺诈分析师或流程工程师都可以在自己的数据流上使用这些模型。在这些模型周围，IBM正在构建能够向左移动工作的功能，使预测、异常检测、优化和语义智能成为你调用的能力，而不是需要你建设的项目。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/V5JlP6rnQZG8psqofXSf6.jpeg","想象一下巧克力工厂中的一条回火生产线，其温度、速度和产量每几秒钟采样一次，并与固定阈值进行监控。如果在该数据流中引入基础模型，它可以预测整晚班的产量，使规划员在还有时间采取措施时就能看到可能的短缺。它会根据生产线在黑巧克力上的正常表现对当天的运行进行评分，因此缓慢的漂移会在巧克力条开始膨胀前被发现。它会在工厂历史中找到最接近的匹配，使工程师知道之前类似运行的结果如何。它会根据工作人员可控制的设置进行条件处理，并在最后的准确点值得调整时进行微调。不需要数据科学团队，并且同一个模型可以应用到每条生产线的每个工厂。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/vTTaugPEu3DaLg_SwVfZP.png","IBM 在提供这些模型之前，先在自己的产品和运营中运行它们，然后与水泥、钢铁、纸浆和造纸、食品和电信行业的设计合作伙伴一起运行。数字说明了一切：每一个准确度点都价值数百万，生产力提升达到 5 到 10 倍，原本需要专家处理的工作现在由拥有决策权的领域专家直接操作。","现在，证明遇到了实时场景：IBM 提供前沿模型，可以理解信号的行为，背后有超过 4400 万次下载；Confluent 提供业务的实时状态，并触达每个执行系统。它们一起以流原生方式运行，托管在 Confluent Cloud 中，并从 Flink 调用。在 AWS 上的 Confluent Cloud 开放访问。随后是 Confluent 平台，将相同的模型和能力带到本地和混合环境中。","通常用于将模型接入生产的数月时间，现在被保留下来：Granite 读取信号，Confluent 提供上下文、治理以及对下游一切的交付。","信号的价值随时间衰减：今天发现漂移的泵是一个工单，下周同样的泵则变成停机。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/wlsMI5mra9TvaO7wA53CF.png","预测和检测是有状态的：下一个值只有在与近期历史对照下才有意义，异常只有在持续感知的正常状态下才存在。Flink 管理这些状态，每个序列都有键值，且容错，因此每个模型都能获取所需的历史数据，而无需单独的数据存储或每次调用访问数据库。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/7szxD-qQx3PDPT_KxWOiC.jpeg","这就是价值复合的地方。Confluent 的数据流平台让业务数据动起来，并使其可用于机器学习。该平台连续流动、连接、管理和处理实时数据，捕捉 IBM Granite 时间序列模型用于预测、异常检测、相似性搜索、分类、缺口填充和优化的实时业务信号。Confluent 提供实现流式用例所需的一切，快速、可靠、安全地实施，从而让你可以专注于开发实时机器学习应用，而不是管理数据基础设施。","Confluent Cloud 是 Confluent 数据流平台的云端部署，提供原生推理功能，使您能够直接在 Confluent 上的 Apache Flink® 中运行 IBM Granite 时间序列模型，为实时数据处理提供更高的灵活性、安全性和成本效率，同时统一数据和机器学习工作流。其优势包括：","通过桥接运营和分析资产，Confluent 帮助团队将实时业务事件转化为可执行的洞察，把 IBM Granite 时间序列模型引入数据流中。并且，由于单一模型无法同时适用于洗发水产品线、信用卡网络和零售目录，IBM 和 Confluent 提供的是一个组合模型，而不是单一模型。","每个决策都向未来提出不同的问题。一个计划周期需要一系列结果，交易台需要从数据中获得各个利率下最准确的数值，一支拥有十万辆车队的车队需要保持合理成本，安全团队则需要在数据流行为异常时及时获知，并了解上次出现问题的情况。该组合包含四个互补的时间序列基础模型，均处于早期访问阶段，并可通过 Confluent 现有的 AI_FORECAST 和 AI_DETECT_ANOMALIES Flink SQL 函数调用。通过一个 SQL 参数即可切换模型，无需重新设计数据管道。","更改模型值，同样的调用即可运行四个模型中的任意一个，无需构建或操作单独的机器学习堆栈。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/cPDlF0B1m3D9wB2cgGMh5.png","没有单一最佳模型，因此几个问题会引导选择。一条系列还是成千上万？一个变量还是多个？训练时间还是开箱即用？预测多久？预测还是异常检测？PatchTST-FM 读取序列的方式就像语言模型读取文本一样，逐块处理，每个变量在自己的通道中，这样一个嘈杂的信号不会拖累其他信号，并返回完整的分布，因此计划者可以根据第90百分位设置补货点。FlowState 会保持一个随每个点更新的运行摘要，并且由于其动态在时间上是连续的，它可以读取秒级 SCADA 数据和小时级市场数据。TTM 在时间和变量之间使用微小的混合网络来替代注意力机制，因此一个拥有百万参数的模型可以在CPU上夜间覆盖十万条系列。而 TSPulse 将时间和频率视图结合在一个小型多任务模型中，用于异常检测、分类、缺口填充以及每个操作员都会问的问题：我们之前见过这个吗。","“小型”是一个决定，而不是妥协：推理原生运行在 Confluent Cloud 中，或者在你自己的 CPU 上使用 Hugging Face Hub 提供的开放权重运行，并且无需云输入或输出，使架构保持简单并降低成本。IBM Granite 还带来了 IBM 的企业 AI 治理框架，提供模型来源和许可透明度，并在模型上提供日益丰富的功能，使每个用例开箱即用时更有用。这正是投资组合和平台汇聚的地方，模型不再只是过去的图书管理员，而是正在进行决策的优化器。","所有四种模型都压缩了事件发生与知晓之间的时间。在流数据中，这一延迟从几天缩短到几秒，并且信号成为其他 AI 系统、代理和流程的触发器，用于调查、分流重要事项，并在需要人工决策时调入人类，且上下文已被收集。每个模型也都被打包为功能模块，使更多工作在平台中完成，而使用它的团队承担的工作更少。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/OrlAwl1BVrDRC49ALrGCF.jpeg","大多数预测仍然是统计模型和直觉的包装，打扮成决策。定制的机器学习并没有弥补这个差距，每个系列一个模型，由人工重新拟合，并随着预测期的延长而漂移，因此规划只覆盖那些值得付出努力的少数系列，其余系列依赖安全库存运行。","跟随一家杂货零售商的需求计划员，规划只覆盖目录的顶部，从不触及底部，底部库存作为流动资金堆在货架上。她给整个目录使用一个共享模型：它可以立即在未见过的系列上工作，考虑天气和促销等因素，并返回一个分布而不是单一预测。没有按SKU构建模型，因此一个模型就是一个模型工厂：同一任务中处理一个运行了两年的SKU和一个只有三个月历史的SKU，用类似SKU启动的新产品，每晚CPU处理10万个SKU，在不同类别和地区使用相同的方法。","该分布将服务水平转化为她可以口头表述的策略。由于每个预测都落在一个主题上，它是一个触发器而不是报告：补货由此触发，分配和定价使用相同的数字，降价在库存老化前发生，补货在货架清空前完成。结果在业务记分的地方显现：缺货和降价减少，客户找到他们需要的商品，货架上的收入得到保护，释放的流动资金通常是商业案例中最大的一项。","异常检测是最广的赛道，而一个被漏掉的异常很少是小事：在欺诈中，它是客户的钱，在安全中，它是一次泄露，在IT运营中，它是客户首先遇到的故障，每一个都会影响品牌，就像影响财务一样。在金融服务中，基于规则的检测是可枚举的，因此对手也可以枚举它，而收紧规则会导致更多诚实客户流失，收入减少，客户流失一半。定制机器学习需要标签，而标签稀缺且过时，且误报的成本高于实际犯罪。","想象一下零售银行的欺诈负责人：模型对每张卡片保持一种正常状态感知，并在支付进行中通过同一个 AI_DETECT_ANOMALIES 调用对每笔支付进行评分。因为它还能进行预测，它会在事件发生之前标记出潜在问题的漂移。一张在过去两年内在相同三个邮政编码购买日用品的卡片，在凌晨 3 点向海外钱包转账时，警报会在资金移动之前响起，而同一客户在诚实的假期中则会顺利通过。","保护从第一天就开始，因为模型在其他地方学到的东西可以转移到没有标记案例的新产品、通道和资产类型上。它也必须持续前进，因为对手不会停：欺诈模式和攻击特征每月变化，因此模型会在银行自己的数据流上定制，根据确认的案例重新拟合，持续改进，而不是每年重建。随着银行和商业变得具有代理性，代理以机器速度发起支付，正常节奏发生变化，交易量上升，因此实时上下文和在途评分更为重要。每一次评分都推动下一步动作：阻止支付，将其升级到附带最接近过往案例的分析师，或者交给代理。","同样的技术可以重新应用于任何具有节奏的实体：IT 延迟、基站 KPI、开线校正线。","每个工厂都依赖于自身模型运行，而保持其准确性很难：统计模型会漂移，基于规则的控制维持设定点但从不改进它，而定制化的机器学习无法解释任何东西，因此优化仍停留在试点阶段，工厂依赖利润运行。","Andrés 在一家洗发水工厂运营工艺，其混合生产线将温度、搅拌器速度、投料速率和粘度流入 Confluent。他在数据流上应用了一个基础模型，并且开箱即可使用：数月的定制建模缩短为几天，生产力提升可达 10 倍，他仅在生产线需要时进行定制。基于他可控的条件，预测变成了一个模拟器：不同混合速度的能耗，不同温度和投料速率下的产量，粘度是否符合规格。优化器在他命名的 KPI 上搜索该空间，尊重他的约束，并解释其建议，因为他无法质询的建议不会被采纳。Andrés 是一名工艺工程师，而不是建模专家，了解生产线的人才掌控它。","优化不会在上线时冻结，而是随着输入变化而重新优化：界面活性剂供应商更换，香氛批次行为不同，需求从400ml瓶转为旅行容量，本季度目标是吞吐量而非能源。重新设定目标和约束，生产线从最近的过去运行及其修正开始，采用下一个最佳设定点而非去年的。收益体现在他CFO追踪的货币中：一个业务点数亿美元的运营，一个点成了七位数的收益，一家食品制造商起步时只有一个工艺，后面有400家工厂。","生产优化支撑着更大的领域：质量预测和设备状况紧随其后，随着人工智能进入制造业、机器人和物理系统，它成为一个庞大且颠覆性市场中的第一个。","上述每条通道都以同一个问题结束：我们以前见过这种情况吗？模型用嵌入来回答这个问题，这些紧致的向量捕捉窗口在时间和频率上的形状，使得两个看似相似的事件无论规模或偏移如何，都相近地落在附近。在流中，每个窗口在到达时被嵌入，并与过去的事件及其结果进行匹配，因此回传的是一个先例，而非评分：那些向此方向漂移的运行及其修复因素，新SKU最相似的需求曲线，以及本会谈中押韵的已确认欺诈案件。同样的嵌入驱动分类和填补缺口，并索引代理在行动前获取的上下文。","这仅仅是时间序列的开始。与语言模型一样，节奏正在加快：新的架构、新的数据源、基于模型构建的代理和用户体验，并且更多内容在首日即可使用。IBM和Confluent将继续创新，继续与设计合作伙伴和客户合作，在实际企业场景中提升生产力、准确性和响应性，涵盖各领域。","Confluent Cloud 现已在抢先体验中提供：可直接在您的数据流上进行预测和异常检测，无需进行模型训练、特征工程或 AI/ML 专业知识。Confluent Cloud 是起点，Confluent Platform 是下一步，因此相同的模型和功能可以覆盖本地和混合环境。反馈循环是关键：您在自己的流中发现的内容会教会模型下一步该如何发展。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"IBM 与 Confluent 宣布四款 IBM Granite 时间序列基础模型以 Early Access 形式上线 Confluent Cloud，面向流数据提供预测、异常检测、相似历史检索与优化等能力。","background":"来源称，模型将在 Confluent Cloud 上托管，并通过 Flink 调用，初始访问开放于 AWS；Confluent Platform 后续将支持本地部署与混合环境。IBM 表示已在自身产品、运营及多个行业设计合作中测试这些模型。","viewpoint":"Aioga 判断：此次发布的核心变化，是将时间序列基础模型与实时业务流结合，并以可调用能力呈现给领域专家。当前仍处于 Early Access，实际效果与适用边界需要通过公开使用结果进一步验证。","implications":"可能影响：企业评估流式预测与异常检测方案时，需要同时核对模型在自身数据上的准确性、调用方式和治理要求。Early Access 不代表能力已适合所有生产场景，来源中的行业成效表述也不足以替代独立验证。","nextStep":"后续观察：应关注 Early Access 的实际开放范围、模型在 Confluent Cloud 中的使用反馈，以及 Confluent Platform 是否按来源所述跟进；同时需要观察来源提到的行业测试结果能否被更多公开证据支持。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-09-06T11:29:58.729Z","sourceHash":"4be8027740158d80","review":{"approved":true,"groundedness":95,"clarity":94,"duplicationRisk":10,"blockingIssues":[],"notes":[]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","editorial-labels","inference-boundary","low-source-overlap","no-html","independent-ai-review"]}},"tags":["行业动态","Hugging Face：Blog（RSS）"],"translations":{"zh-CN":{"title":"IBM 与 Confluent 发布 Granite 时间序列模型 Early Access，可在流数据上实时预测与检测异常","summary":"IBM 与 Confluent 宣布四款 IBM Granite 时间序列基础模型（PatchTST-FM、FlowState、TTM、TSPulse）以 Early Access 形式上线 Confluent Cloud。","category":"行业动态","source":"huggingface.co","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM 与 Confluent 发布 Granite 时间序列模型 Early Access，可在流数据上实时预测与检测异常 - Aioga AI资讯","description":"IBM 与 Confluent 宣布四款 IBM Granite 时间序列基础模型（PatchTST-FM、FlowState、TTM、TSPulse）以 Early Access 形式上线 Confluent Cloud。","url":"https://www.aioga.com/news/cmtk6zwl601osroqryptpbgai/","articleBody":["直到现在，那些决策都是基于过时的经济学：一次一个定制模型，每个模型都需要专家数月的工作。因此，团队只对几百个有资金的系列进行建模，其余的则通过安全边际、额外库存、额外空间、额外容差来覆盖，这些都是在时间窗口结束后采取的措施。这个边际成本是无法预测决策的代价，每个周期都要支付。","时间序列基础模型（TSFM）改变了这一点。它经过一次在庞大且多样的信号上训练，就能推广到从未见过的系列：提供一个测量窗口，它会告诉你接下来会发生什么，行为偏离正常的程度，历史中哪些情况类似，以及哪些设置最能实现目标。使用一个模型也不需要一大群数据科学家：需求规划员、欺诈分析师或流程工程师都可以在自己的数据流上使用这些模型。在这些模型周围，IBM正在构建能够向左移动工作的功能，使预测、异常检测、优化和语义智能成为你调用的能力，而不是需要你建设的项目。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/V5JlP6rnQZG8psqofXSf6.jpeg","想象一下巧克力工厂中的一条回火生产线，其温度、速度和产量每几秒钟采样一次，并与固定阈值进行监控。如果在该数据流中引入基础模型，它可以预测整晚班的产量，使规划员在还有时间采取措施时就能看到可能的短缺。它会根据生产线在黑巧克力上的正常表现对当天的运行进行评分，因此缓慢的漂移会在巧克力条开始膨胀前被发现。它会在工厂历史中找到最接近的匹配，使工程师知道之前类似运行的结果如何。它会根据工作人员可控制的设置进行条件处理，并在最后的准确点值得调整时进行微调。不需要数据科学团队，并且同一个模型可以应用到每条生产线的每个工厂。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/vTTaugPEu3DaLg_SwVfZP.png","IBM 在提供这些模型之前，先在自己的产品和运营中运行它们，然后与水泥、钢铁、纸浆和造纸、食品和电信行业的设计合作伙伴一起运行。数字说明了一切：每一个准确度点都价值数百万，生产力提升达到 5 到 10 倍，原本需要专家处理的工作现在由拥有决策权的领域专家直接操作。","现在，证明遇到了实时场景：IBM 提供前沿模型，可以理解信号的行为，背后有超过 4400 万次下载；Confluent 提供业务的实时状态，并触达每个执行系统。它们一起以流原生方式运行，托管在 Confluent Cloud 中，并从 Flink 调用。在 AWS 上的 Confluent Cloud 开放访问。随后是 Confluent 平台，将相同的模型和能力带到本地和混合环境中。","通常用于将模型接入生产的数月时间，现在被保留下来：Granite 读取信号，Confluent 提供上下文、治理以及对下游一切的交付。","信号的价值随时间衰减：今天发现漂移的泵是一个工单，下周同样的泵则变成停机。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/wlsMI5mra9TvaO7wA53CF.png","预测和检测是有状态的：下一个值只有在与近期历史对照下才有意义，异常只有在持续感知的正常状态下才存在。Flink 管理这些状态，每个序列都有键值，且容错，因此每个模型都能获取所需的历史数据，而无需单独的数据存储或每次调用访问数据库。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/7szxD-qQx3PDPT_KxWOiC.jpeg","这就是价值复合的地方。Confluent 的数据流平台让业务数据动起来，并使其可用于机器学习。该平台连续流动、连接、管理和处理实时数据，捕捉 IBM Granite 时间序列模型用于预测、异常检测、相似性搜索、分类、缺口填充和优化的实时业务信号。Confluent 提供实现流式用例所需的一切，快速、可靠、安全地实施，从而让你可以专注于开发实时机器学习应用，而不是管理数据基础设施。","Confluent Cloud 是 Confluent 数据流平台的云端部署，提供原生推理功能，使您能够直接在 Confluent 上的 Apache Flink® 中运行 IBM Granite 时间序列模型，为实时数据处理提供更高的灵活性、安全性和成本效率，同时统一数据和机器学习工作流。其优势包括：","通过桥接运营和分析资产，Confluent 帮助团队将实时业务事件转化为可执行的洞察，把 IBM Granite 时间序列模型引入数据流中。并且，由于单一模型无法同时适用于洗发水产品线、信用卡网络和零售目录，IBM 和 Confluent 提供的是一个组合模型，而不是单一模型。","每个决策都向未来提出不同的问题。一个计划周期需要一系列结果，交易台需要从数据中获得各个利率下最准确的数值，一支拥有十万辆车队的车队需要保持合理成本，安全团队则需要在数据流行为异常时及时获知，并了解上次出现问题的情况。该组合包含四个互补的时间序列基础模型，均处于早期访问阶段，并可通过 Confluent 现有的 AI_FORECAST 和 AI_DETECT_ANOMALIES Flink SQL 函数调用。通过一个 SQL 参数即可切换模型，无需重新设计数据管道。","更改模型值，同样的调用即可运行四个模型中的任意一个，无需构建或操作单独的机器学习堆栈。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/cPDlF0B1m3D9wB2cgGMh5.png","没有单一最佳模型，因此几个问题会引导选择。一条系列还是成千上万？一个变量还是多个？训练时间还是开箱即用？预测多久？预测还是异常检测？PatchTST-FM 读取序列的方式就像语言模型读取文本一样，逐块处理，每个变量在自己的通道中，这样一个嘈杂的信号不会拖累其他信号，并返回完整的分布，因此计划者可以根据第90百分位设置补货点。FlowState 会保持一个随每个点更新的运行摘要，并且由于其动态在时间上是连续的，它可以读取秒级 SCADA 数据和小时级市场数据。TTM 在时间和变量之间使用微小的混合网络来替代注意力机制，因此一个拥有百万参数的模型可以在CPU上夜间覆盖十万条系列。而 TSPulse 将时间和频率视图结合在一个小型多任务模型中，用于异常检测、分类、缺口填充以及每个操作员都会问的问题：我们之前见过这个吗。","“小型”是一个决定，而不是妥协：推理原生运行在 Confluent Cloud 中，或者在你自己的 CPU 上使用 Hugging Face Hub 提供的开放权重运行，并且无需云输入或输出，使架构保持简单并降低成本。IBM Granite 还带来了 IBM 的企业 AI 治理框架，提供模型来源和许可透明度，并在模型上提供日益丰富的功能，使每个用例开箱即用时更有用。这正是投资组合和平台汇聚的地方，模型不再只是过去的图书管理员，而是正在进行决策的优化器。","所有四种模型都压缩了事件发生与知晓之间的时间。在流数据中，这一延迟从几天缩短到几秒，并且信号成为其他 AI 系统、代理和流程的触发器，用于调查、分流重要事项，并在需要人工决策时调入人类，且上下文已被收集。每个模型也都被打包为功能模块，使更多工作在平台中完成，而使用它的团队承担的工作更少。","：https://cdn-uploads.huggingface.co/production/uploads/64e8143f6de557454220921e/OrlAwl1BVrDRC49ALrGCF.jpeg","大多数预测仍然是统计模型和直觉的包装，打扮成决策。定制的机器学习并没有弥补这个差距，每个系列一个模型，由人工重新拟合，并随着预测期的延长而漂移，因此规划只覆盖那些值得付出努力的少数系列，其余系列依赖安全库存运行。","跟随一家杂货零售商的需求计划员，规划只覆盖目录的顶部，从不触及底部，底部库存作为流动资金堆在货架上。她给整个目录使用一个共享模型：它可以立即在未见过的系列上工作，考虑天气和促销等因素，并返回一个分布而不是单一预测。没有按SKU构建模型，因此一个模型就是一个模型工厂：同一任务中处理一个运行了两年的SKU和一个只有三个月历史的SKU，用类似SKU启动的新产品，每晚CPU处理10万个SKU，在不同类别和地区使用相同的方法。","该分布将服务水平转化为她可以口头表述的策略。由于每个预测都落在一个主题上，它是一个触发器而不是报告：补货由此触发，分配和定价使用相同的数字，降价在库存老化前发生，补货在货架清空前完成。结果在业务记分的地方显现：缺货和降价减少，客户找到他们需要的商品，货架上的收入得到保护，释放的流动资金通常是商业案例中最大的一项。","异常检测是最广的赛道，而一个被漏掉的异常很少是小事：在欺诈中，它是客户的钱，在安全中，它是一次泄露，在IT运营中，它是客户首先遇到的故障，每一个都会影响品牌，就像影响财务一样。在金融服务中，基于规则的检测是可枚举的，因此对手也可以枚举它，而收紧规则会导致更多诚实客户流失，收入减少，客户流失一半。定制机器学习需要标签，而标签稀缺且过时，且误报的成本高于实际犯罪。","想象一下零售银行的欺诈负责人：模型对每张卡片保持一种正常状态感知，并在支付进行中通过同一个 AI_DETECT_ANOMALIES 调用对每笔支付进行评分。因为它还能进行预测，它会在事件发生之前标记出潜在问题的漂移。一张在过去两年内在相同三个邮政编码购买日用品的卡片，在凌晨 3 点向海外钱包转账时，警报会在资金移动之前响起，而同一客户在诚实的假期中则会顺利通过。","保护从第一天就开始，因为模型在其他地方学到的东西可以转移到没有标记案例的新产品、通道和资产类型上。它也必须持续前进，因为对手不会停：欺诈模式和攻击特征每月变化，因此模型会在银行自己的数据流上定制，根据确认的案例重新拟合，持续改进，而不是每年重建。随着银行和商业变得具有代理性，代理以机器速度发起支付，正常节奏发生变化，交易量上升，因此实时上下文和在途评分更为重要。每一次评分都推动下一步动作：阻止支付，将其升级到附带最接近过往案例的分析师，或者交给代理。","同样的技术可以重新应用于任何具有节奏的实体：IT 延迟、基站 KPI、开线校正线。","每个工厂都依赖于自身模型运行，而保持其准确性很难：统计模型会漂移，基于规则的控制维持设定点但从不改进它，而定制化的机器学习无法解释任何东西，因此优化仍停留在试点阶段，工厂依赖利润运行。","Andrés 在一家洗发水工厂运营工艺，其混合生产线将温度、搅拌器速度、投料速率和粘度流入 Confluent。他在数据流上应用了一个基础模型，并且开箱即可使用：数月的定制建模缩短为几天，生产力提升可达 10 倍，他仅在生产线需要时进行定制。基于他可控的条件，预测变成了一个模拟器：不同混合速度的能耗，不同温度和投料速率下的产量，粘度是否符合规格。优化器在他命名的 KPI 上搜索该空间，尊重他的约束，并解释其建议，因为他无法质询的建议不会被采纳。Andrés 是一名工艺工程师，而不是建模专家，了解生产线的人才掌控它。","优化不会在上线时冻结，而是随着输入变化而重新优化：界面活性剂供应商更换，香氛批次行为不同，需求从400ml瓶转为旅行容量，本季度目标是吞吐量而非能源。重新设定目标和约束，生产线从最近的过去运行及其修正开始，采用下一个最佳设定点而非去年的。收益体现在他CFO追踪的货币中：一个业务点数亿美元的运营，一个点成了七位数的收益，一家食品制造商起步时只有一个工艺，后面有400家工厂。","生产优化支撑着更大的领域：质量预测和设备状况紧随其后，随着人工智能进入制造业、机器人和物理系统，它成为一个庞大且颠覆性市场中的第一个。","上述每条通道都以同一个问题结束：我们以前见过这种情况吗？模型用嵌入来回答这个问题，这些紧致的向量捕捉窗口在时间和频率上的形状，使得两个看似相似的事件无论规模或偏移如何，都相近地落在附近。在流中，每个窗口在到达时被嵌入，并与过去的事件及其结果进行匹配，因此回传的是一个先例，而非评分：那些向此方向漂移的运行及其修复因素，新SKU最相似的需求曲线，以及本会谈中押韵的已确认欺诈案件。同样的嵌入驱动分类和填补缺口，并索引代理在行动前获取的上下文。","这仅仅是时间序列的开始。与语言模型一样，节奏正在加快：新的架构、新的数据源、基于模型构建的代理和用户体验，并且更多内容在首日即可使用。IBM和Confluent将继续创新，继续与设计合作伙伴和客户合作，在实际企业场景中提升生产力、准确性和响应性，涵盖各领域。","Confluent Cloud 现已在抢先体验中提供：可直接在您的数据流上进行预测和异常检测，无需进行模型训练、特征工程或 AI/ML 专业知识。Confluent Cloud 是起点，Confluent Platform 是下一步，因此相同的模型和功能可以覆盖本地和混合环境。反馈循环是关键：您在自己的流中发现的内容会教会模型下一步该如何发展。"]},"en":{"title":"IBM and Confluent have released the Granite time series model Early Access, which can predict and detect anomalies in real time on streaming data","summary":"IBM and Confluent have announced that four IBM Granite time series foundational models (PatchTST-FM, FlowState, TTM, TSPups) are now available in early access format on Confluent Cloud.","category":"Industry","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM and Confluent have released the Granite time series model Early Access, which can predict and detect anomalies in real time on streaming data - Aioga AI News","description":"IBM and Confluent have announced that four IBM Granite time series foundational models (PatchTST-FM, FlowState, TTM, TSPups) are now available in early access format on Confluent C...","url":"https://www.aioga.com/en/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:07:47.254Z"},"ja":{"title":"IBMとConfluentは、ストリーミングデータの異常をリアルタイムで予測・検出できるGranite時系列モデル「Early Access」をリリースしました","summary":"IBMとConfluentは、4つのIBM Granite時系列モデル(PatchTST-FM、FlowState、TTM、TSPups)がConfluent Cloudで早期アクセス形式で利用可能になったと発表しました。","category":"業界動向","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBMとConfluentは、ストリーミングデータの異常をリアルタイムで予測・検出できるGranite時系列モデル「Early Access」をリリースしました - Aioga AIニュース","description":"IBMとConfluentは、4つのIBM Granite時系列モデル(PatchTST-FM、FlowState、TTM、TSPups)がConfluent Cloudで早期アクセス形式で利用可能になったと発表しました。","url":"https://www.aioga.com/ja/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:07:47.676Z"},"ko":{"title":"IBM과 Confluent는 스트리밍 데이터에서 실시간으로 이상 현상을 예측하고 탐지할 수 있는 Granite 시계열 모델 Early Access를 출시했습니다","summary":"IBM과 Confluent는 네 가지 IBM Granite 시계열 기초 모델(PatchTST-FM, FlowState, TTM, TSPups)이 이제 Confluent Cloud에서 얼리 액세스 형식으로 제공된다고 발표했습니다.","category":"업계 동향","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM과 Confluent는 스트리밍 데이터에서 실시간으로 이상 현상을 예측하고 탐지할 수 있는 Granite 시계열 모델 Early Access를 출시했습니다 - Aioga AI 뉴스","description":"IBM과 Confluent는 네 가지 IBM Granite 시계열 기초 모델(PatchTST-FM, FlowState, TTM, TSPups)이 이제 Confluent Cloud에서 얼리 액세스 형식으로 제공된다고 발표했습니다.","url":"https://www.aioga.com/ko/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:07:56.990Z"},"es":{"title":"IBM y Confluent han lanzado el modelo de serie temporal Granite Early Access, que puede predecir y detectar anomalías en tiempo real en datos en streaming","summary":"IBM y Confluent han anunciado que cuatro modelos fundacionales de la serie temporal IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) ya están disponibles en formato de acceso anticipado en Confluent Cloud.","category":"Industria","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM y Confluent han lanzado el modelo de serie temporal Granite Early Access, que puede predecir y detectar anomalías en tiempo real en datos en streaming - Aioga Noticias de IA","description":"IBM y Confluent han anunciado que cuatro modelos fundacionales de la serie temporal IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) ya están disponibles en formato de acceso anti...","url":"https://www.aioga.com/es/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:07:57.166Z"},"fr":{"title":"IBM et Confluent ont publié le modèle de série temporelle Granite Early Access, qui peut prédire et détecter des anomalies en temps réel sur des données en flux","summary":"IBM et Confluent ont annoncé que quatre modèles fondateurs de la série temporelle IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) sont désormais disponibles en format accès anticipé sur Confluent Cloud.","category":"Industrie","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM et Confluent ont publié le modèle de série temporelle Granite Early Access, qui peut prédire et détecter des anomalies en temps réel sur des données en flux - Aioga Actualités IA","description":"IBM et Confluent ont annoncé que quatre modèles fondateurs de la série temporelle IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) sont désormais disponibles en format accès antic...","url":"https://www.aioga.com/fr/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:06.491Z"},"de":{"title":"IBM und Confluent haben das Granite-Zeitreihenmodell Early Access veröffentlicht, das Anomalien in Echtzeit auf stromenden Daten vorhersagen und erkennen kann.","summary":"IBM und Confluent haben angekündigt, dass vier grundlegende IBM Granite-Zeitreihen-Modelle (PatchTST-FM, FlowState, TTM, TSPups) nun im Early-Access-Format auf Confluent Cloud verfügbar sind.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM und Confluent haben das Granite-Zeitreihenmodell Early Access veröffentlicht, das Anomalien in Echtzeit auf stromenden Daten vorhersagen und erkennen kann. - Aioga KI-News","description":"IBM und Confluent haben angekündigt, dass vier grundlegende IBM Granite-Zeitreihen-Modelle (PatchTST-FM, FlowState, TTM, TSPups) nun im Early-Access-Format auf Confluent Cloud verf...","url":"https://www.aioga.com/de/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:06.582Z"},"pt-BR":{"title":"IBM e Confluent lançaram o modelo de série temporal Granite Early Access, que pode prever e detectar anomalias em tempo real em dados em streaming","summary":"IBM e Confluent anunciaram que quatro modelos fundamentais de séries temporais IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) estão agora disponíveis em formato de acesso antecipado no Confluent Cloud.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM e Confluent lançaram o modelo de série temporal Granite Early Access, que pode prever e detectar anomalias em tempo real em dados em streaming - Aioga Notícias de IA","description":"IBM e Confluent anunciaram que quatro modelos fundamentais de séries temporais IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) estão agora disponíveis em formato de acesso anteci...","url":"https://www.aioga.com/pt-BR/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:15.877Z"},"ru":{"title":"IBM и Confluent выпустили модель Granite Time Series Early Access, которая может предсказывать и обнаруживать аномалии в реальном времени на потоковых данных","summary":"IBM и Confluent объявили, что четыре базовые модели временных рядов IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) теперь доступны в формате раннего доступа на Confluent Cloud.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM и Confluent выпустили модель Granite Time Series Early Access, которая может предсказывать и обнаруживать аномалии в реальном времени на потоковых данных - Aioga Новости ИИ","description":"IBM и Confluent объявили, что четыре базовые модели временных рядов IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) теперь доступны в формате раннего доступа на Confluent Cloud.","url":"https://www.aioga.com/ru/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:15.977Z"},"ar":{"title":"أصدرت آي بي إم وكونفلوينت نموذج السلاسل الزمنية جرانيت Early Access، الذي يمكنه التنبؤ والكشف عن الشذوذات في الوقت الحقيقي على بيانات البث","summary":"أعلنت IBM وConfluent أن أربعة نماذج تأسيسية لسلسلة زمنية IBM Granite (PatchTST-FM، FlowState، TTM، TSPups) متاحة الآن بصيغة الوصول المبكر على سحابة Confluent.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"أصدرت آي بي إم وكونفلوينت نموذج السلاسل الزمنية جرانيت Early Access، الذي يمكنه التنبؤ والكشف عن الشذوذات في الوقت الحقيقي على بيانات البث - Aioga أخبار الذكاء الاصطناعي","description":"أعلنت IBM وConfluent أن أربعة نماذج تأسيسية لسلسلة زمنية IBM Granite (PatchTST-FM، FlowState، TTM، TSPups) متاحة الآن بصيغة الوصول المبكر على سحابة Confluent.","url":"https://www.aioga.com/ar/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:25.136Z"},"hi":{"title":"आईबीएम और कॉन्फ्लुएंट ने ग्रेनाइट टाइम सीरीज़ मॉडल अर्ली एक्सेस जारी किया है, जो स्ट्रीमिंग डेटा पर वास्तविक समय में विसंगतियों की भविष्यवाणी और पता लगा सकता है","summary":"आईबीएम और कॉन्फ्लुएंट ने घोषणा की है कि चार आईबीएम ग्रेनाइट समय श्रृंखला मूलभूत मॉडल (पैचटीएसटी-एफएम, फ्लोस्टेट, टीटीएम, टीएसपीअप) अब कॉन्फ्लुएंट क्लाउड पर प्रारंभिक पहुंच प्रारूप में उपलब्ध हैं।","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"आईबीएम और कॉन्फ्लुएंट ने ग्रेनाइट टाइम सीरीज़ मॉडल अर्ली एक्सेस जारी किया है, जो स्ट्रीमिंग डेटा पर वास्तविक समय में विसंगतियों की भविष्यवाणी और पता लगा सकता है - Aioga AI समाचार","description":"आईबीएम और कॉन्फ्लुएंट ने घोषणा की है कि चार आईबीएम ग्रेनाइट समय श्रृंखला मूलभूत मॉडल (पैचटीएसटी-एफएम, फ्लोस्टेट, टीटीएम, टीएसपीअप) अब कॉन्फ्लुएंट क्लाउड पर प्रारंभिक पहुंच प्रारूप...","url":"https://www.aioga.com/hi/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:25.719Z"},"it":{"title":"IBM e Confluent hanno rilasciato il modello della serie temporale Granite Early Access, che può prevedere e rilevare anomalie in tempo reale sui dati in streaming","summary":"IBM e Confluent hanno annunciato che quattro modelli fondamentali della serie temporale IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) sono ora disponibili in formato accesso anticipato su Confluent Cloud.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM e Confluent hanno rilasciato il modello della serie temporale Granite Early Access, che può prevedere e rilevare anomalie in tempo reale sui dati in streaming - Aioga Notizie IA","description":"IBM e Confluent hanno annunciato che quattro modelli fondamentali della serie temporale IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) sono ora disponibili in formato accesso an...","url":"https://www.aioga.com/it/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:35.135Z"},"nl":{"title":"IBM en Confluent hebben het Granite tijdreeksmodel Early Access uitgebracht, dat afwijkingen in realtime kan voorspellen en detecteren op stromende data","summary":"IBM en Confluent hebben aangekondigd dat vier IBM Granite tijdreeks-basismodellen (PatchTST-FM, FlowState, TTM, TSPups) nu beschikbaar zijn in early access-formaat op Confluent Cloud.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM en Confluent hebben het Granite tijdreeksmodel Early Access uitgebracht, dat afwijkingen in realtime kan voorspellen en detecteren op stromende data - Aioga AI-nieuws","description":"IBM en Confluent hebben aangekondigd dat vier IBM Granite tijdreeks-basismodellen (PatchTST-FM, FlowState, TTM, TSPups) nu beschikbaar zijn in early access-formaat op Confluent Clo...","url":"https://www.aioga.com/nl/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:34.646Z"},"tr":{"title":"IBM ve Confluent, akış verilerinde anormallikleri gerçek zamanlı olarak tahmin edip tespit edebilen Granite zaman serisi modeli Early Access'i yayımladı","summary":"IBM ve Confluent, dört IBM Granite zaman serisi temel modelinin (PatchTST-FM, FlowState, TTM, TSPups) artık Confluent Cloud'da erken erişim formatında mevcut olduğunu duyurdu.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM ve Confluent, akış verilerinde anormallikleri gerçek zamanlı olarak tahmin edip tespit edebilen Granite zaman serisi modeli Early Access'i yayımladı - Aioga AI Haberleri","description":"IBM ve Confluent, dört IBM Granite zaman serisi temel modelinin (PatchTST-FM, FlowState, TTM, TSPups) artık Confluent Cloud'da erken erişim formatında mevcut olduğunu duyurdu.","url":"https://www.aioga.com/tr/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:44.476Z"},"vi":{"title":"IBM và Confluent đã phát hành mô hình chuỗi thời gian Granite Early Access, có thể dự đoán và phát hiện bất thường theo thời gian thực trên dữ liệu truyền trực tuyến","summary":"IBM và Confluent đã công bố rằng bốn mô hình nền tảng chuỗi thời gian IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) hiện đã có sẵn dưới dạng truy cập sớm trên Confluent Cloud.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM và Confluent đã phát hành mô hình chuỗi thời gian Granite Early Access, có thể dự đoán và phát hiện bất thường theo thời gian thực trên dữ liệu truyền trực tuyến - Tin tức AI Aioga","description":"IBM và Confluent đã công bố rằng bốn mô hình nền tảng chuỗi thời gian IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) hiện đã có sẵn dưới dạng truy cập sớm trên Confluent Cloud.","url":"https://www.aioga.com/vi/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:44.393Z"},"id":{"title":"IBM dan Confluent telah merilis model seri waktu Granite Early Access, yang dapat memprediksi dan mendeteksi anomali secara real time pada data streaming","summary":"IBM dan Confluent telah mengumumkan bahwa empat model fondasi seri waktu IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) kini tersedia dalam format akses awal di Confluent Cloud.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM dan Confluent telah merilis model seri waktu Granite Early Access, yang dapat memprediksi dan mendeteksi anomali secara real time pada data streaming - Berita AI Aioga","description":"IBM dan Confluent telah mengumumkan bahwa empat model fondasi seri waktu IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) kini tersedia dalam format akses awal di Confluent Cloud.","url":"https://www.aioga.com/id/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:53.827Z"},"th":{"title":"IBM และ Confluent ได้เปิดตัวโมเดลชุดข้อมูลเวลา Granite ชื่อ Early Access ซึ่งสามารถทํานายและตรวจจับความผิดปกติแบบเรียลไทม์บนข้อมูลสตรีมมิ่ง","summary":"IBM และ Confluent ได้ประกาศว่าโมเดลพื้นฐานของชุดข้อมูลเวลา IBM Granite สี่แบบ (PatchTST-FM, FlowState, TTM, TSPups) พร้อมให้ใช้งานในรูปแบบ early access บน Confluent Cloud แล้ว","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM และ Confluent ได้เปิดตัวโมเดลชุดข้อมูลเวลา Granite ชื่อ Early Access ซึ่งสามารถทํานายและตรวจจับความผิดปกติแบบเรียลไทม์บนข้อมูลสตรีมมิ่ง - ข่าว AI Aioga","description":"IBM และ Confluent ได้ประกาศว่าโมเดลพื้นฐานของชุดข้อมูลเวลา IBM Granite สี่แบบ (PatchTST-FM, FlowState, TTM, TSPups) พร้อมให้ใช้งานในรูปแบบ early access บน Confluent Cloud แล้ว","url":"https://www.aioga.com/th/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:08:53.468Z"},"pl":{"title":"IBM i Confluent wypuściły model szeregów czasowych Granite Early Access, który potrafi przewidywać i wykrywać anomalie w czasie rzeczywistym na podstawie danych strumieniowych","summary":"IBM i Confluent ogłosiły, że cztery podstawowe modele serii czasowych IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) są już dostępne w formacie wczesnego dostępu na Confluent Cloud.","category":"行业动态","source":"Hugging Face：Blog（RSS）","aggregationSource":"Hugging Face：Blog（RSS）","pageTitle":"IBM i Confluent wypuściły model szeregów czasowych Granite Early Access, który potrafi przewidywać i wykrywać anomalie w czasie rzeczywistym na podstawie danych strumieniowych - Aioga Wiadomości AI","description":"IBM i Confluent ogłosiły, że cztery podstawowe modele serii czasowych IBM Granite (PatchTST-FM, FlowState, TTM, TSPups) są już dostępne w formacie wczesnego dostępu na Confluent Cl...","url":"https://www.aioga.com/pl/news/cmtk6zwl601osroqryptpbgai/","contentTranslated":true,"sourceHash":"230b4517060c9e0d","translatedAt":"2026-09-02T15:09:03.198Z"}},"evidenceTier":"verified-news","reviewStatus":"automated-ingest","indexable":true,"editorialCover":""}}