Kategorie: Google Cloud Platform

Catalyst Black: the next-generation mobile multiplayer action game built on Google Cloud

At Super Evil Megacorp our vision is to push the creative and technical boundaries of video games. This was the driving force behind Vainglory, released in 2014, which showcased AAA-quality games can be successful on mobile while filling a gap for multiplayer online battle arena games on mobile. Our vision continues with the development of our next free-to-play game, Catalyst Black, a AAA mobile multiplayer action game hosted on Google Cloud.Setting up the stage for a new kind of real-time mobile gameOur community enjoys developing complex skills and strategies whilst playing our games; they expect beautiful games that can keep up with their quick dexterity and creative thinking. Catalyst Black has been developed with this in mind. To match player expectations, we partnered with Google Cloud very early in our development process. We had a number of learnings from the development of Vainglory that Google Cloud was primed with solutions for. For starters, we wanted to roll our own deployment process and to automate certain aspects of our infrastructure management. Previously, we had five engineers rotating on-call so if a server went down someone could get it back up or spin up a new server manually in case of a sudden spike in players. Now, we use Google Kubernetes Engine to automatically scale our servers up and down as needed. By leveraging an infrastructure that takes care of itself, we can focus instead on creating an engrossing game with exciting features. This allows our team to push the boundaries of how multiplayer games on mobile can foster human connections, with new developments such as a drop-in/drop-out feature that allows players to join a friend’s match with minimal delay. The game is also cross-platform, which means that players on both Android and iOS can team up, no matter what device they use. Tailoring gaming experiences to players with data analyticsTo create great games we need to learn from our players, and for that we need a robust analytics pipeline with low latency. Data analytics enables us to recognize individual players and how they like to play so we can serve them a relevant, tailored gaming experience they’ll enjoy. When players log in, for example, we build a profile for them on our analytics pipeline to understand what kind of player they are. Based on their performance during the game tutorial and subsequent matches, we can understand their skill level, whether they’re interested in purchasing gear for loadouts in-play, what style they like to play, and so on. With this, we can then match a player with others that have similar skill levels and adjust our push notifications depending on how often they join the game.For our data analytics pipeline which informs all of our product decisions, we rely on BigQuery and Looker to scale our analytics frameworks and understand the players’ journey through our live operations, statistical modeling, and A/B testing. We also rely heavily on Looker for observability. That includes looking at crash rates, application responsiveness rates, uptime of all services, and latency across all regions so that players can interact with each other in near real time, which is a must for combat games. Capturing large amounts of detailed data is critical in helping us understand how players behave and how the game itself is performing. Thanks to Google Cloud, we have gained the scalability and compute power needed to support this kind of big data analysis that we rely on daily. With Catalyst Black in beta, with a small user base, we generated millions of events per day. That will go up several orders of magnitude once we launch. That’s a big operation to handle but relying on Google Cloud means we have no concerns that the technical side of things will keep up; we trust Google Cloud to scale.We also need data to make efficient decisions around marketing and products to sustain our business so that we can continue to offer free-to-play games. We use Google Analytics to maximize ROI on our paid media spend through traditional social media and advertising channels. We also use BigQuery ML models to estimate the lifetime of a user and predict ROS and ROI so we can optimize our ad spend.What we’ve noticed so far is that Google Cloud offers outstanding performance and speed for analyzing data and addressing potential issues because its analytics and infrastructure solutions are very well integrated in the Google Cloud environment. For example, by monitoring our player activities we can notice if there’s a sudden dip in latency in any region around the world, and simply spin up new servers to better serve the population there using Compute Engine. By migrating to Google Cloud we’re now spending only a fifth on a per DAU basis compared to what we spent with our previous provider while working on Vainglory. Democratizing great gaming experiencesThe launch of a new game is a big day for us, but it’s also the start of a long journey ahead. We expect to see a significant spike in the number of players following the launch of Catalyst Black, and we also expect to see those numbers eventually flatten down to hundreds of thousands daily. By then, we’ll have a better understanding of who Catalyst Black players are and what their vision is for the future of the game, which we can then collaborate on together.Additionally, with our in-house E.V.I.L. game engine, we can perform on devices that are much older. We can run on Samsung S7s, a device originally released in 2016, with the same 30 frames per second that the latest devices can run. The idea is for anyone to be able to access great games, no matter where they are and what device they have. On 25 May 2022, we launched Catalyst Black. Collaborating with Google Cloud means that we’re prepared for that global scalability. Our games are live operations that organically change and grow, and by relying on GCP we can anticipate our infrastructure needs to support players anywhere in the world throughout that journey, so we’re excited to see where we’ll go from here.Related ArticleHow a top gaming company transformed its approach to security with Splunk and Google CloudAristocrat overhauled its security and compliance processes using Splunk and Google Cloud technologies.Read Article
Quelle: Google Cloud Platform

Getting started with ML: 25+ resources recommended by role and task

Wondering how to get started with Vertex AI? Below, we’ve collected a list of resources to help you build and hone your skills across data science, machine learning, and artificial intelligence on Google Cloud.We’ve broken down the resources by what we think a Data Analyst, Data Scientist, ML Engineer, or a Software Engineer might be most interested in. But we also recognize there’s a lot of overlap between these roles, so even if you identify as a Data Scientist, for example, you might find some of the resources for ML Engineers or Developers just as useful!Data Analyst From data to insights, and perhaps some modeling, data analysts look for ways to help their stakeholders understand the value of their data.Data exploration and Feature Engineering[Guide] Exploratory Data Analysis for Feature Selection in Machine Learning[Documentation] Feature preprocessing in BigQuery Data visualization[Guide] Visualizing BigQuery data using Data Studio[Blog] Go from Database to Dashboard with BigQuery and LookerData ScientistAs a data scientist, you might be interested in generating insights from data, primarily through extensive exploratory data analysis, visualization, feature engineering, and modeling. If you’d like one place to start, check out Best practices for implementing machine learning on Google Cloud. Model registry[Video] AI/ML Notebooks how to with Apache Spark, BigQuery ML and  Vertex AI Model RegistryModel training[Codelab] Train models with the Vertex AI Workbench notebook executor[Codelab] Use autopackaging to fine tune Bert with Hugging Face on Vertex AI Training[Blog] How To train and tune PyTorch models on Vertex AILarge scale model training[Codelab] Multi-Worker Training and Transfer Learning with TensorFlow[Blog] Optimize training performance with Reduction Server on Vertex AI[Video] Distributed training on Vertex AI Workbench Model tuning[Codelab] Hyperparameter tuning[Video] Faster model training and experimentation with Vertex AIModel serving[Blog] How to deploy PyTorch models on Vertex AI[Blog] 5 steps to go from a notebook to a deployed modelML EngineerBelow are resources for an ML Engineer, someone whose focus area is MLOps, or the operationalization of feature management, model serving and monitoring, and CI/CD with ML pipelines.Feature management[Blog] Kickstart your organization’s ML application development flywheel with the Vertex Feature Store[Video] Introduction to Vertex AI Feature StoreModel Monitoring[Blog] Monitoring feature attributions: How Google saved one of the largest ML services in troubleML Pipelines[Blog] Orchestrating PyTorch ML Workflows on Vertex AI Pipelines[Codelab] Intro to Vertex Pipelines[Codelab] Using Vertex ML Metadata with PipelinesMachine Learning Operations[Guide] MLOps: Continuous delivery and automation pipelines in machine learningSoftware Engineer with ML applicationsHere are some resources if you work more as a traditional software engineer who spends more time on using ML in applications and less time on data wrangling, model building, or MLOps.[Blog] Find anything blazingly fast with Google’s vector search technology[Blog] Using Vertex AI for rapid model prototyping and deployment[Video] Machine Learning for developers in a hurryLooking for resources?Are you looking for more information but you can’t seem to find them? Let us know! Reach out to us on Linkedin:Nikita NamjoshiPolong LinRelated ArticlePick your AI/ML Path on Google CloudYour ultimate AI/ML decision treeRead Article
Quelle: Google Cloud Platform

Accelerating ML with Vertex AI: From retail and finance to manufacturing and automotive

Artificial intelligence (AI) and machine learning (ML) are transforming industries around the world, from trailblazing new frontiers in conversational human-computer interactions and speech-based analysis, to improving product discovery in retail,to unlocking medical research with advancements like AlphaFold. But underpinning all ML advancements is a common challenge: fast-tracking the building and deployment of ML models into production, and abstracting the most technically complex processes into unified platforms that open ML to more users. Our mission is to remove every barrier in the way of deploying useful and predictable ML at scale. This is why, in May 2021, we announced the general availability of Vertex AI, a managed ML platform designed specifically to accelerate the deployment and maintenance of ML models. Leveraging Vertex AI, data scientists can speed up ML development and experimentation by 5x, with 80% fewer lines of code required.In the year since the launch, customers across diverse industries have successfully accelerated the deployment of machine learning models in production with Vertex AI. In fact, through Vertex AI and BigQuery, we have seen 2.5 times more machine learning predictions generated in 2021 compared to the previous year. Additionally, customers are seeing great value in Vertex AI’s unified data and AI story. This is best represented by the 25x growth in active customers we have seen for Vertex AI Workbench over the last six months.Let’s take a look at how some of these organizations are using Vertex AI today. Accelerating ML in retail: ML at Wayfair, Etsy, Lowe’s and Magalu Our research of over 100 global retail executives identified that AI and ML-powered applications have the potential to drive $230-515 billion in business value.  Whether the use cases involve optimizing inventory or bettering customer experience, retail is among the industries where ML adoption has been strongest. For example, online furniture and home goods retailer Wayfair has been able to run large model training jobs 5-10x faster by leveraging Vertex AI. “We’re doing ML at a massive scale, and we want to make that easy. That means accelerating time-to-value for new models, increasing reliability and speed of very large regular re-training jobs, and reducing the friction to build and deploy models at scale,” said Matt Ferrari, Head of Ad Tech, Customer Intelligence, and Machine Learning at Wayfair, in a Forbes article. Vertex AI helps the company to “weave ML into the  fabric of how we make decisions,” he added. Elsewhere, Etsy estimates it  has reduced the time it takes to go from ideation to a live ML experiment by about 50%. “Our training and prototyping platform largely relies on Google Cloud services like Vertex AI and Dataflow, where customers can experiment freely with the ML framework of their choice,” the company notes in a blog post. “These services let customers easily leverage complex ML infrastructure (such as GPUs) through comfortable interfaces like Jupyter Notebooks. Massive extract transform load (ETL) jobs can be run through Dataflow while complex training jobs of any form can be submitted to Vertex AI for optimization.”Forecasting in particular is a major retail use case that can be significantly bettered with the power of ML. Vertex AI Forecast is already helping Lowe’s with a range of models at the company’s more than 1,700 stores, according to Amaresh Siva, senior vice president for Innovation, Data and Supply Chain Technology at Lowe’s.“Using Vertex AI Forecast, Lowe’s has been able to create accurate hierarchical models that balance between SKU and store-level forecasts. These models take into account our store-level, SKU-level, and region-level inventory, promotions data and multiple other signals, and are yielding more accurate forecasts,” said Siva.Brazilian retailer Magalu has similarly deployed Vertex AI to reduce inventory prediction errors. With Vertex AI, “four-week live forecasting showed significant improvements in error (WAPE) compared to our previous models,” said Fernando Nagano, director of Analytics and Strategic Planning at Magalu. “This high accuracy insight has helped us to plan our inventory allocation and replenishment more efficiently to ensure that the right items are in the right locations at the right time to meet customer demand and manage costs appropriately.”From memory to manufacturing to mobile payments: ML at Seagate, Coca Cola Bottlers Japan, and Cash App Retail is not the only industry leveraging the power of AI and ML. According to our research, 66% of manufacturers who use AI in their day-to-day operations report that their reliance on AI is increasing.Google joined forces with Seagate, our HDD original equipment manufacturer (OEM) partner for Google’s data centers, to leverage ML for improved prediction of frequent HDD problems, such as disk failure. The Vertex AI AutoML model generated for the effort achieved a precision of 98% with a recall of 35%, compared to precision for 70-80% and recall of 20-25% for the competing custom ML model. Coca Cola Bottlers Japan (CCBJ) is also ramping up its ML efforts, using Vertex AI and BigQuery to process billions of data records from 700,000 vending machines, helping the company to make strategic decisions about when and where to locate products. “We have created a prediction model of where to place vending machines, what products are lined up in the machines and at what price, how much they will sell, and implemented a mechanism that can be analyzed on a map,” said Minori Matsuda,Data Science Manager / Google Developer Expert at CCBJ, in a blog post.  “We were able to realize it in a short period of time with a sense of speed, from platform examination to introduction, prediction model training, on-site proof of concept to rollout.”Turning to finance, Cash App, a platform from the U.S.-based financial services company Square, is leveraging products from Google Cloud and NVIDIA to achieve a roughly 66% improvement in completion time for core ML processing workflows. “Google Cloud gave us critical control over our processes,” said Kyle De Freitas, a senior software engineer at Dessa, which was acquired by Cash App in 2020. “We recognized that Compute Engine A2 VMs, powered by the NVIDIA A100 Tensor Core GPUs, could dramatically reduce processing times and allow us to experiment much faster. Running NVIDIA A100 GPUs on Google Cloud’s Vertex AI gives us the foundation we need to continue innovating and turning ideas into impactful realities for our customers.”Driving toward an ML-fueled future: ML at Cruise and SUBARUIn the automotive space, manufacturers throughout the world have invested billions to digitize operations and invest in AI to both optimize design and enable new features. For instance, self-driving car service Cruise has millions of miles of autonomous travel under its belt, with Vertex AI helping the company to quickly train and update ML models that power crucial functions like image recognition and scene understanding. “After we ingest and analyze that data, it’s fed back into our dynamic ML Brain, a continuous learning machine that actively mines from the collected data to automatically train new models that exceed the performance of the older models,” explained Mo Elshenawy, Executive Vice President of Engineering at Cruise, in a blog post. “This is done with the help of Vertex AI, where we are able to train hundreds of models simultaneously, using  hundreds of GPU years every month!”Meanwhile, SUBARU is turning to ML to eliminate fatal accidents caused by its cars. SUBARU Lab uses Google Cloud to analyze images from the company’s EyeSight  stereo cameras, for example. The team uses a combination of NVIDIA A100 GPUs and Compute Engine for processing muscle, with data scientists and data engineers using Vertex AI to build models. “I chose Google Cloud from many platforms because it had multiple managed services such as Verex AI, the managed notebooks option, and Vertex AI Training that were useful for AI development. It was also fascinating to have high-performance hardware that could handle large-scale machine learning operations,” said Thossimi Okubo, Senior Engineer of AI R&D at SUBARU. Working together to accelerate ML deploymentWe are very encouraged by the adoption of Vertex AI, and we are excited to continue working with key customers and partners to expand our thinking around the challenges data scientists face in accelerating deployment of ML models in production. Watch our Google Cloud Applied ML Summit session with Smi-tha Shyam, Director of Engineering for Uber AI, and Bryan Goodman, Director of AI and Cloud at Ford, to get a sense of how we’re working with partners and customers in this journey. To learn more, check out additional expert commentary at our Applied ML Summit, peruse our latest Vertex AI updates, or visit our Data Science on Google Cloud page to learn more about our unified data and AI story.Related ArticleWhat is Vertex AI? Developer advocates share moreDeveloper Advocates Priyanka Vergadia and Sara Robinson explain how Vertex AI supports your entire ML workflow—from data management all t…Read Article
Quelle: Google Cloud Platform

Accelerate the deployment of ML in production with Vertex AI

As part of today’s Google Cloud Applied ML Summit, we’re announcing a variety of product features and technology partnerships to help you more quickly and efficiently build, deploy, manage, and maintain machine learning (ML) models in production. Our performance tests found a 2.5x increase in the number of ML predictions generated through Vertex AI and BigQuery in 2021, and a 25x increase in active customers for Vertex AI Workbench in just the last six months. Customers have made clear that managed and integrated ML platforms are crucial to accelerating the deployment of ML in production. For example, Wayfair accelerated large model training jobs by 5-10x with Vertex AI, enabling increased experimentation, reduced coding, and more models making it to production. Likewise, Seagate used AutoML to build a ML model with 98% precision, compared to only 70-80% from their earlier custom models.Bryan Goodman, Director of AI and Cloud at Ford, said, “Vertex AI is an integral part of the Ford machine learning development platform, including accelerating our efforts to scale AI for non-software experts.“This momentum is tremendous, but we know there is more work to be done to help enterprises across the globe fast-track the digitization of operations with AI. According to Gartner*, “Only 10% of organizations have 50% or more of their software engineers trained on machine learning skills.” [Source: Gartner: Survey Analysis: AI Adoption Spans Software Engineering and Organizational Boundaries – Van Baker, Benoit Lheureux – November 25, 2021] Similarly, Gartner states that “on average, 53% of [ML] projects make it to production.” [Source: Gartner: 4 Machine Learning Best Practices to Achieve Project Success – Afraz Jaffri, Carlie Idoine, Erick Brethenoux – December 7, 2021].These findings speak to the primary challenge of not only gaining ML skills or abstracting technology dependencies so more people can participate in the process of ML deployment, but also to applying those skills to deploy models in production, continuously monitor, and drive business impact.Let’s take a look at how our announcements will help you remove the barriers to deploying useful and predictable ML at scale. Four pillars for accelerating ML deployment in productionThe features we’re announcing today fit into the following four-part framework that we’ve developed in discussions with customers, partners, and other industry thought leaders.Providing freedom of choiceData scientists work most effectively when they have the freedom to choose the ML frameworks, deployment instances, and compute processors they’ll work with. To this end, we partnered with NVIDIA earlier this year to launch One Click Deploy of NVIDIA AI software solutions to Vertex AI Workbench. NVIDIA’s NGC catalog lets data scientists start their model development on Google Cloud, speeding the path to building and deploying state-of-the-art AI. The feature simplifies the deployment of Jupyter Notebooks from over 12 complex steps to a single click, abstracting away routine tasks to help data science teams focus on accelerating ML deployment in production.We also believe this power to choose should not come at a cost. With this in mind, we are thrilled to announce the availability of Vertex AI Training Reduction Server, which supports both Tensorflow and PyTorch. Training Reduction Server is built to optimize bandwidth and latency of multi-node distributed training on NVIDIA GPUs. This significantly reduces the training time required for large language workloads, like BERT, and further enables cost parity across different approaches. In many mission-critical business scenarios, a shortened training cycle allows data scientists to train a model with higher predictive performance within the constraints of a deployment window. Meeting users where they are Whether ML tasks involve pre-trained APIs, AutoML, or custom models built from the ground up, skills proficiency should not be the gating criteria for participation in an enterprise-wide strategy. This is the only way to get your data engineers, data analysts, ML researchers, MLOps engineers, and data scientists to participate in the process of ML acceleration across the organization. To this end, we’re announcing the preview of Vertex AI Tabular Workflows, which includes a glassbox and managed AutoML pipeline that lets you see and interpret each step in the model building and deployment process. Now, you can comfortably train datasets of over a terabyte, without sacrificing accuracy, by picking and choosing which parts of the process you want AutoML to handle versus which parts you want to engineer yourself. Elements of Tabular Workflows can also be integrated into your existing Vertex AI pipelines. We’ve added new managed algorithms including advanced research models like TabNet, new algorithms for feature selection, model distillation and much more. Future noteworthy components will include implementation of Google proprietary models such as Temporal Fusion Transformers, and Open Source models like XGboost and Wide & Deep. Uniting data and AITo fast track the deployment of ML models into production, your organization needs a unified data and AI strategy. To further integrate data engineering capabilities directly into the data science environment, we’re announcing features to address all data types: structured data, graph data, and unstructured data. First up, for structured data, we are announcing the preview of Serverless Spark on Vertex AI Workbench. This allows data scientists to launch a serverless spark session within their notebooks and interactively develop code. In the space of graph data, we are excited to introduce a data partnership with Neo4j that unlocks the power of graph-based ML models, letting data scientists explore, analyze, and engineer features from connected data in Neo4j and then deploy models with Vertex AI, all within a single unified platform. With Neo4j Graph Data Science and Vertex AI, data scientists can extract more predictive power from models using graph-based inputs, and get to production faster across use cases such as fraud and anomaly detection, recommendation engines, customer 360 , logistics, and more.In the space of unstructured data, our partnership with Labelboxis all about helping data scientists leverage the power of unstructured data to build more effective ML models on Vertex AI. Labelbox’s native integration with Vertex AI reduces the time required to label unstructured image, text, audio, and video data, which helps accelerate model development for image classification, object detection, entity recognition, and various other tasks. With the integration only available on Google Cloud, Labelbox and Vertex AI create a flywheel for accelerated model development. Managing and maintaining ML modelsFinally, our customers demand tools to easily manage and maintain ML models. Data scientists shouldn’t need to be infrastructure engineers or operations engineers to keep models accurate, explainable, scaled, disaster resistant, and secure, all in an ever-changing environment. To address this need, we’re announcing the preview of Vertex AI Example-based Explanations. This novel Explainable AI technique helps data scientists identify mislabeled examples in their training data or discover what data to collect to improve model accuracy. Using example-based explanations to quickly diagnose and treat issues, data scientists can now maintain a high bar on model quality.Ford and Vertex AIAs mentioned, we’ve seen our customers achieve great results with our AI and ML solutions. Ford, for example, is leveraging Vertex AI across many use cases and user types.“We’re using Vertex AI pipelines to build generic and reusable modular machine learning workflows. These are useful as people build on the work of others and to accelerate their own work,” explained Goodman. “For low code and no code users, AutoML models are useful for transcribing speech and basic object detection, and we like that there is integrated deployment for trained models. It really helps people get things into use, which is important. For power users, we are extensively leveraging Vertex AI’s custom model deployment for our in-house models. It’s ideal for data scientists and data engineers not to have to master skills in infrastructure and software. This is critical for growing the community of AI builders at Ford, and we’re seeing really good success.”Customer stories and enthusiasm propel our efforts to continue creating better products that make AI and ML more accessible, sustainable, and powerful. We’re thrilled to have been on this journey with you so far, and we can’t wait to see what you do with our new announcements. To learn more, check out additional expert commentary at our Applied ML Summit, and visit our Data Science on Google Cloud page to learn more about how Google Cloud is helping you fast-track the deployment of ML in production. *GARTNER is a registered trademark and service of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved.Related ArticleAccelerating ML with Vertex AI: From retail and finance to manufacturing and automotiveHow businesses across industries are accelerating deployment of machine learning models into production with VertexAI.Read Article
Quelle: Google Cloud Platform

Google Cloud supports higher education with Cloud Digital Leader program

College and university faculty can now easily teach cloud literacy and digital transformation with the Cloud Digital Leader track, part of the Google Cloud career readiness program. The new track is available for eligible faculty who are preparing their students for a cloud-first workforce. As part of the track, students will build their cloud literacy and learn the value of Google Cloud in driving digital transformation, while also preparing for the Cloud Digital Leader certification exam. Apply today!Cloud Digital Leader career readiness trackThe Cloud Digital Leader career readiness track is designed to equip eligible faculty with the resources needed to prepare their students for the Cloud Digital Leader certification. This Google Cloud certification requires no previous cloud computing knowledge or hands-on experience. The training path enables students to build cloud literacy and learn how to evaluate the capabilities of Google Cloud in preparation for future job roles. The curriculumFaculty members can access this curriculum as part of the Google Cloud Career Readiness program. Faculty from eligible institutions can apply to lead students through the no-cost  program which provides access to the four-course on-demand training, hands-on practice to supplement the learning, and additional exam prep resources. Students who complete the entire program are eligible to apply for a certification exam discount. The Cloud Digital Leader track is the third program available for classroom use, joining the Associate Cloud Engineer and Data Analyst tracks. Cloud resources for your classroomReady to get started? Apply today to access the Cloud Digital Leader career readiness track for your classroom. Read the eligibility criteria for faculty. You can preview the course content at no cost.Related ArticleRead Article
Quelle: Google Cloud Platform