# clustering Published articles for clustering. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Multi-Cluster v2 and Stateless Mode now in Preview DevFeed: [Multi-Cluster v2 and Stateless Mode now in Preview]() Original publisher: [Read original article]() Author: Alexander Schwartz Published: 2026-07-17T00:00:00Z Content type: release Language: en Sources: [Keycloak Blog]() Topics: [Keycloak](), [Authentication](), [Database](), [Availability](), [Replication](), [Kubernetes](), [Amazon Web Services]() Tags: [authentication](), [availability](), [clustering](), [database](), [distributed](), [failover](), [idm](), [kerberos](), [ldap](), [monitoring](), [openid-connect](), [preview](), [replication](), [saml](), [sso]() ### AI overview Keycloak 26.7 introduces Stateless Mode as a preview. It moves volatile data, including authentication sessions, action tokens, and brute-force counters, from Infinispan caches to the database, supporting multi-cluster deployments without an external Infinispan cluster. ### Source excerpt Starting with Keycloak 26.7, the stateless feature is available as a preview. It fundamentally simplifies how Keycloak handles volatile data -- authentication sessions, action tokens, and brute-force counters move from embedded or external Infinispan caches to the database. The result: Keycloak deployments that are easier to operate, more resilient, and ready for multi-cluster setups without an external Infinispan cluster. Read on to learn more, and ask questions in our upcoming live Q&A event on Jul 21! The problem with the current approach A multi-cluster setup becomes necessary when no transparent networking between two sites is available for Keycloak's embedded Infinispan to form a single cluster -- for example, when sites are in different availability zones or data centers without a shared network layer. Running Keycloak in such a multi-cluster setup today requires significant infrastructure beyond Keycloak itself. Without the stateless feature, the architecture depends on an external Infinispan cluster with cross-datacenter replication channels, dedicated monitoring to detect and handle Infinispan failover, and operational procedures for failback. It was tied to specific environments like Kubernetes and AWS, and using it in other environments would require significant work to adapt it. Even in single-cluster deployments, Infinispan clustering introduces operational overhead: rebalancing data in distributed caches when a node is lost or restarted, and transient failures in login flows if a node unexpectedly disappears. In large installations, the login failure cache can grow significantly, consuming substantial memory and causing long rebalancing times when nodes join or leave the cluster. While distributed caches replicate data across at least two nodes -- so individual node failures do not lose data -- a full cluster restart during minor upgrades resets volatile state such as in-progress authentication sessions and brute-force counters. What changes with stateles ## BIY: Preparing a Dataset and Benchmarking AI Models for Scatterplot-Related Tasks DevFeed: [BIY: Preparing a Dataset and Benchmarking AI Models for Scatterplot-Related Tasks]() Original publisher: [Read original article]() Author: João Palmeiro Published: 2026-01-19T14:52:55Z Content type: article Language: en Sources: [Feedzai]() Topics: [Accessibility](), [LLM evaluation / benchmarking](), [multimodal-ai](), [Canvas](), [AI Models](), [dataset](), [SVG]() Tags: [accessibility](), [ai-models](), [artificial-intelligence](), [benchmarking](), [canvas](), [clustering](), [data-visualization](), [dataset](), [large-language-models](), [multimodal](), [multimodal-ai](), [outlier-detection](), [svg]() ### AI overview The article introduces Benchmark It Yourself (BIY), an effort to prepare a dataset and benchmark multimodal AI models for scatterplot-related tasks. It examines generating distribution-aware descriptions from scatterplot images to improve the accessibility of canvas charts, including direct description generation and structured-data extraction for predefined templates. Initial results were mixed. ### Source excerpt Benchmark It Yourself (BIY): Preparing a Dataset and Benchmarking AI Models for Scatterplot-Related Tasks When we need to visualize and interact with millions, or even just thousands, of individual points while analyzing data, we typically resort to rendering them in the browser using a canvas. The other common approach for the web, SVG, doesn't scale when the number of individual elements increases to such quantities. However, while solving one problem, canvas charts introduce a new challenge: accessibility. Although SVG charts are not accessible by default, they can be by design. Each part of an SVG chart has a corresponding element on the web page, allowing for a programmable, accessible experience for screen reader users. We can simply think of SVG as HTML. On the other hand, a canvas chart is just like a PNG image. If a screen reader user tries to learn more about a canvas chart, unless the developer has prepared a detailed description of it, they will just hear the word "image". There's no way to get an idea of what one of these charts represents, let alone extract any insights. For static charts, the solution can be as simple as preparing a description and integrating it into the rendered chart. However, for platforms leveraging dynamic, large datasets, automatically generating these descriptions is not a simple task, especially for charts like scatterplots where data distributions can assume countless forms. At Feedzai, we started exploring ways to generate data distribution-aware descriptions for scatterplots from their respective images using recent multimodal AI models. When the raw data is not available, or the datasets are composed of several thousand or million instances, relying on chart images and these models becomes tempting. This combination has the potential to generate such descriptions and serve them alongside their respective charts, significantly improving the accessibility of canvas charts. That said, we focused on two main directions: using ## A differentially private framework for gaining insights into AI chatbot use DevFeed: [A differentially private framework for gaining insights into AI chatbot use]() Original publisher: [Read original article]() Published: 2025-12-10T21:59:41Z Content type: article Language: en Sources: [The latest research from Google]() Topics: [Artificial Intelligence](), [AI Chat](), [Large Language Model](), [Google](), [Algorithm](), [pii]() Tags: [ai](), [chatbots](), [clustering](), [code](), [data](), [data-protection](), [generative-ai](), [google](), [insights](), [large-language-model](), [llm](), [pii](), [pipeline](), [research](), [responsible-ai](), [security-privacy-and-abuse-prevention]() ### AI overview Google Research introduces Urania, a framework for generating high-level insights into AI chatbot usage while protecting user conversation privacy. Its pipeline combines differentially private clustering, keyword extraction, and LLM summarization to provide formal, end-to-end differential privacy guarantees. ### Source excerpt Generative AI ## Engineering a Scalable Topic Pipeline: A BERTopic and GenAI Case Study DevFeed: [Engineering a Scalable Topic Pipeline: A BERTopic and GenAI Case Study]() Original publisher: [Read original article]() Author: Saniya Naphade Published: 2025-08-28T06:25:05Z Content type: tutorial Language: en Sources: [GumGum]() Topics: [genai](), [Artificial Intelligence](), [Algorithm](), [data](), [GPU](), [datasets]() Tags: [ai](), [algorithm](), [case-study](), [clustering](), [data](), [datasets](), [engineering](), [genai](), [gpu](), [llm-applications](), [topic-modeling](), [workflow]() ### AI overview GumGum describes a scalable hybrid topic-modeling pipeline that combines deterministic BERTopic clustering with generative AI post-processing. The workflow processes impression data weekly, uses cuML for GPU acceleration, and addresses BERTopic's high outlier rate to recover more topic-related insights. ### Source excerpt Introduction: In a world where trends change in an instant, relying on static taxonomies to target ad campaigns is no longer effective. At GumGum, we face a constant torrent of digital content, ingesting millions of web pages every day. This massive flow of information must be turned into actionable insights, but because trends are dynamic, traditional methods like hard-coded taxonomies and slow hand-labeling are simply not viable at this scale. While large language models (LLMs) can offer some insight, using them at this volume is often cost-prohibitive and can lead to inconsistent results. To solve this, we've developed a lightweight and repeatable hybrid pipeline. This approach combines a deterministic clustering method -- specifically BERTopic¹ -- with a strategic post-processing layer using generative AI. The result is a more efficient and powerful topic modeling workflow that gives us the stability needed for large datasets and provides valuable, human-centric insights. Let's now walk through the key stages of this pipeline, beginning with data preparation. Data preparation The pipeline's first step is a critical data preparation phase built to handle immense scale. We process a comprehensive collection of impression data on a weekly basis to ensure we can capture not only major trends like the Summer Box Office but also the smaller, more nuanced shifts within that topic. For instance, in a given week, the pipeline might detect a specific surge of interest around a new movie's opening weekend, which would be lost in a less frequent processing schedule. This raw data is then thoroughly cleaned and prepared for the next stage, laying the groundwork for accurate and timely topic modeling. Dynamic Topic Modeling With this clean, prepared data, the real work of topic modeling begins. The pipeline then leverages BERTopic¹, a topic modeling algorithm, in conjunction with cuML³ for GPU acceleration, that is both powerful and efficient, and serves as our deterministic co ## Keycloak 26.3.2 released DevFeed: [Keycloak 26.3.2 released]() Original publisher: [Read original article]() Author: Keycloak Team Published: 2025-07-24T00:00:00Z Content type: release Language: en Sources: [Keycloak Blog]() Topics: [Keycloak](), [OpenID connect (OIDC)](), [Databases](), [Security](), [migration]() Tags: [clustering](), [dependencies](), [idm](), [kerberos](), [keycloak](), [keycloak-release](), [ldap](), [logging](), [mariadb](), [migration](), [oidc](), [openid-connect](), [postgresql](), [release](), [saml](), [sso](), [upgrade](), [version](), [vulnerability]() ### AI overview Keycloak 26.3.2 was released on July 24, 2025. The release adds an OIDC identity-provider authentication option, improves clustering and client-session logging, upgrades dependencies, and lists resolved bugs and vulnerabilities. ### Source excerpt To download the release go to Keycloak downloads. Upgrading Before upgrading refer to the migration guide for a complete list of changes. All resolved issues New features #40237 Add option "Requires short state parameter" to OIDC IDP authentication Enhancements #40970 Run clustering compatibility tests on release/x.y branches #41034 Improve logging for client sessions load #41257 Upgrade to Infinispan 15.0.18.Final infinispan Bugs #39091 Flaky test: org.keycloak.testsuite.cluster.JGroupsCertificateRotationClusterTest#testCoordinatorHasScheduleTask ci #39634 Update MariaDB connector to 3.5.3 dist/quarkus #39854 Flaky test: org.keycloak.testsuite.cluster.PermissionTicketInvalidationClusterTest#crudWithFailover ci #40553 Upgrade org.postgresql:postgresql to version 42.7.7 to address CVE-2025-49146 dependencies #40736 CVE-2025-49574 - Exposure of Resource to Wrong Sphere vulnerability in io.vertx:vertx-core dependencies #40782 Flaky test: org.keycloak.testsuite.cluster.RealmInvalidationClusterTest#crudWithFailover ci #40784 Default jdbc-ping cluster setup for distributed caches fails in Oracle infinispan #40977 Loglevel recorded from build phase dist/quarkus #40980 Can't update security-admin-console via admin UI with volatile sessions infinispan #40995 LDAP / ModelException: At least one condition should be provided to OR query core #41018 Flaky test: org.keycloak.testsuite.cluster.ClientInvalidationClusterTest#crudWithFailover ci #41038 FIPS errors in CI #41082 Multiple primary key defined when attempting to upgrade after 26.3.0 core #41103 Service Account users now showing in the User List admin/ui #41105 Unknown relation when removing realm role with --db-schema configured storage #41152 Docs use em-dashes instead of double dashes for SPI options in regular text docs #41204 UpdateTest CI failures ci #41370 [26.3] MariaDB connector dependency is not properly overriden dist/quarkus ## Tutorial: Building a Highly Available PostgreSQL Cluster with etcd, Patroni, HAProxy, and keepalived DevFeed: [Tutorial: Building a Highly Available PostgreSQL Cluster with etcd, Patroni, HAProxy, and keepalived]() Original publisher: [Read original article]() Author: Techno Tim Published: 2024-12-07T13:00:00Z Content type: tutorial Language: en Sources: [Techno Tim]() Topics: [postgresql clusters](), [PostgreSQL](), [Tutorial](), [Database]() Tags: [clustering](), [ha](), [patroni](), [postgres](), [postgresql](), [production](), [self-hosted](), [tutorial]() ### AI overview A tutorial for building a highly available and fault-tolerant PostgreSQL cluster with etcd, Patroni, HAProxy, and keepalived. It covers PostgreSQL node setup, certificate-based communication, etcd configuration, and cluster verification. ### Source excerpt Today is the day that you start running PostgresSQL in a cluster! In this tutorial we'll be setting up a production ready Postgres cluster that's highly available and fault tolerant using PostgreSQL, etcd, Patroni, HA Proxy, and keepalived. This resilient combination will ensure that you can always reach your database even when a node in the cluster goes down! 📺 Watch Video Nodes # ha pr... ## Carnival of Mathematics #233 DevFeed: [Carnival of Mathematics #233]() Original publisher: [Read original article]() Published: 2024-11-01T07:00:00Z Content type: article Language: en Sources: [Jeremy Kun]() Topics: [math](), [GPU](), [Nvidia](), [clustering](), [Regular expression](), [Python]() Tags: [carnival](), [clustering](), [fibonacci](), [gpu](), [mathematics](), [nvidia](), [python](), [regex]() ### AI overview A roundup of mathematics news and media from October 2022, including a newly reported largest known prime, mathematical research, discussions of political geography and clustering, and videos about voting paradoxes, polylinks, and prime-recognizing regular expressions. ### Source excerpt Welcome to the 233rd Carnival of Mathematics! Who can forget 233, the 6th Fibonacci prime? Hey, not all numbers are interesting. Don't ask me about the smallest positive uninteresting number. You can't make it interesting with your feeble mind tricks! Anyway, on to the fun. Provers and Shakers The big discovery this month was a new largest known prime number, $2^{136279841} - 1$, as reported by the Great Internet Mersenne Prime Search. ## Semantically Compress Text to Save On LLM Costs DevFeed: [Semantically Compress Text to Save On LLM Costs]() Original publisher: [Read original article]() Author: Lou Kratz Published: 2024-10-28T13:34:46Z Content type: article Language: en Sources: [Bazaarvoice]() Topics: [Large Language Model](), [long-context](), [Compression](), [clustering](), [data]() Tags: [bazaarvoice](), [clustering](), [compression](), [context-window](), [llm](), [uncategorized]() ### AI overview Bazaarvoice describes a multi-pass hierarchical clustering approach for semantically compressing large collections of user reviews so its AI Review Summaries feature can operate within LLM context-window limits and reduce token-related costs. The article says the technique allows adjustment of detail loss and helped make the feature financially feasible. ### Source excerpt Introduction Large language models are fantastic tools for unstructured text, but what if your text doesn't fit in the context window? Bazaarvoice faced exactly this challenge when building our AI Review Summaries feature: millions of user reviews simply won't fit into the context window of even newer LLMs and, even if they did, it would [...] ## Comparison between Redis and DragonflyDB's data stores DevFeed: [Comparison between Redis and DragonflyDB's data stores]() Original publisher: [Read original article]() Published: 2024-08-27T00:00:00Z Content type: article Language: en Sources: [Gaurav Sarma's Blog]() Topics: [Redis](), [data](), [hashing]() Tags: [architecture](), [clustering](), [comparison](), [hash](), [memory](), [redis](), [single-threaded](), [vs]() ### AI overview This article compares how Redis and DragonflyDB store data in memory. It explains Redis's single-threaded architecture and DragonflyDB's multithreaded, shared-nothing design, then introduces separate chaining, open addressing, and extendible hashing. ### Source excerpt . [Redis vs DragonflyDB Data Store Comparison](redis-vs-dragonflydbs-data-stores-cover... ## Fully Homomorphic Encryption in Production Systems DevFeed: [Fully Homomorphic Encryption in Production Systems]() Original publisher: [Read original article]() Published: 2024-07-31T07:00:00Z Content type: article Language: en Sources: [Jeremy Kun]() Topics: [homomorphic encryption](), [FHE](), [Encryption](), [Cryptography](), [Edge](), [Microsoft](), [iOS](), [Differential Privacy](), [Embeddings](), [clustering](), [Library]() Tags: [apple](), [clustering](), [cryptography](), [differential-privacy](), [embeddings](), [encryption](), [fhe](), [homomorphic-encryption](), [ios](), [library](), [mathematics](), [microsoft-edge](), [programming](), [systems]() ### AI overview A living document catalogs production systems known to use fully or somewhat homomorphic encryption. It describes deployments and techniques involving Microsoft Edge password checking, Apple's Live Caller ID Lookup, and Apple's private image search, while distinguishing FHE from SHE. ### Source excerpt In this living document, I will list all production systems I'm aware of that use fully homomorphic encryption (FHE). For background on FHE, see my overview of the field. If you have any information about production FHE systems not in this list, or corrections to information in this list, please send me an email with sufficient detail allow the claim to be publicly verified. For all production deployments, I will distinguish between cases where the deployed system does "fully" homomorphic encryption (with bootstrapping), aka FHE, and "somewhat" homomorphic encryption, aka SHE (avoiding bootstrapping). ## A Better Mastodon Client DevFeed: [A Better Mastodon Client]() Original publisher: [Read original article]() Published: 2023-12-19T00:00:00Z Content type: opinion Language: en Sources: [Tim Kellogg]() Topics: [Mastodon](), [client](), [Streamlit](), [Embeddings](), [SQLite](), [Algorithm](), [Vibe coding](), [Artificial Intelligence](), [scikit-learn](), [tailscale](), [OpenAI]() Tags: [client](), [clustering](), [embeddings](), [github](), [github-copilot](), [mastodon](), [sqlite](), [streamlit](), [tailscale]() ### AI overview A developer describes building a Mastodon timeline dashboard that caches posts in SQLite, creates embeddings, clusters posts by topic, summarizes the clusters with an LLM, and makes the dashboard accessible on a phone through Tailscale. ### Source excerpt Last night I had an idea and went ahead and built it. I'd like to tell you about it. Find the source code here. ## A clustering-based approach to create deep learning datasets in a day DevFeed: [A clustering-based approach to create deep learning datasets in a day]() Original publisher: [Read original article]() Author: Roland Meertens Published: 2023-11-28T17:33:30Z Content type: article Language: en Sources: [Bumble Tech]() Topics: [dataset](), [Computer vision](), [Deep learning](), [foundation-models](), [Fine-tuning](), [Inference]() Tags: [clustering](), [computer-vision](), [data-science](), [dataset](), [deep-learning](), [fine-tuning](), [inference](), [machine-learning](), [models](), [retrieval](), [zero-shot]() ### AI overview The article discusses the cost and time involved in creating labeled computer vision datasets. It considers zero-shot learning and foundational models such as GPT-3 and CLIP for data retrieval, while noting that some use cases still require fine-tuning on task-specific data. ### Source excerpt A clustering-based approach to create deep learning datasets in a day Introduction Understanding what's happening in an image is both an important task, as well as a costly one. In the last few years, the field of computer vision has greatly accelerated due to the advances in neural networks. At Bumble Inc., we see potential value in computer vision for a variety of use cases, such as improving the safety of our platform and providing our members with a better user experience. The most common way to train these neural networks is by showing it many images with the corresponding label. Unfortunately, this can be a costly task. Not only does one need to build and train the model, one also wants to do hyperparameter search over multiple configurations of possible networks, and -- of course -- one needs to find or build a dataset suitable for the task at hand. Building the dataset is both the most important task, as well as a very time consuming one. Gathering data, setting up labelling requirements, and of course the labelling itself all take a lot of time and money. This normally leads to trade-offs, by choosing either to build only a small dataset, or by trying to fit existing datasets into your specific use-case. One alternative is of course to not build a dataset at all, to instead use zero-shot learning for your use case. I argued in the past that this is unreasonably effective, and allows you to test your use-case before even training a model. When using zero-shot learning one predicts labels without explicitly training on the classes you are trying to learn. One example of this can be achieved by using the CLIP model, which is trained to have a strong association between text and images. By looking at the distance between the description of your class and the image you can run inference without training anything. However, there are some use cases where we need the strongest possible model by fine-tuning it to our specific data. Using foundational models for data s ## Advent of Kotlin Solutions DevFeed: [Advent of Kotlin Solutions]() Original publisher: [Read original article]() Published: 2022-01-02T00:00:00Z Content type: tutorial Language: en Sources: [Kt. Academy]() Topics: [Kotlin](), [Programming](), [clustering](), [JSON]() Tags: [algorithms](), [clustering](), [json](), [kotlin](), [pattern-matching](), [recursion](), [sealed-class](), [workshop-learning-programming]() ### AI overview This article presents Kotlin solutions for Advent of Kotlin 2021 exercises, including JSON stringification, generating well-formed parentheses, tree algorithms, and k-means clustering. It discusses recursion, sealed classes, smart casting, and Kotlin functions used in the solutions. ### Source excerpt Solutions to the Advent of Kotlin! ## Advent of Kotlin: Week 3 DevFeed: [Advent of Kotlin: Week 3]() Original publisher: [Read original article]() Published: 2021-12-14T00:00:00Z Content type: article Language: en Sources: [Kt. Academy]() Topics: [clustering](), [Kotlin]() Tags: [clustering](), [kotlin](), [workshop-learning-programming]() ### AI overview This article covers k-means clustering as Week 3 of the Advent of Kotlin series. ### Source excerpt Week 3 of the Advent of Kotlin: k-means clustering ## An overview of end-to-end entity resolution for big data DevFeed: [An overview of end-to-end entity resolution for big data]() Original publisher: [Read original article]() Author: adriancolyer Published: 2020-12-14T14:37:00Z Content type: article Language: en Sources: [Adrian Colyer]() Topics: [Entity resolution](), [big-data](), [data]() Tags: [algorithms-and-data-structures](), [article](), [big-data](), [blocking](), [clustering](), [comparisons](), [overview](), [uncategorized]() ### AI overview This article summarizes an ACM Computing Surveys survey on end-to-end entity resolution for big data. It explains the main pipeline stages: blocking to reduce comparisons, block processing to remove redundant and superfluous comparisons, matching entity-description pairs, and clustering matches into resolved entities. It also outlines classification dimensions including schema awareness, matching process, and batch or incremental processing. ### Source excerpt An overview of end-to-end entity resolution for big data, Christophides et al., ACM Computing Surveys, Dec. 2020, Article No. 127 The ACM Computing Surveys are always a great way to get a quick orientation in a new subject area, and hot off the press is this survey on the entity resolution (aka record linking) problem. It's an ... Continue reading An overview of end-to-end entity resolution for big data ## Preparing a Proxmox Server for Production Use DevFeed: [Preparing a Proxmox Server for Production Use]() Original publisher: [Read original article]() Author: Techno Tim Published: 2020-11-28T14:00:00Z Content type: tutorial Language: en Sources: [Techno Tim]() Topics: [Proxmox](), [networking](), [ceph](), [Linux]() Tags: [ceph](), [clustering](), [hardware](), [iommu](), [linux](), [networking](), [proxmox](), [ssh](), [storage](), [tutorials]() ### AI overview A tutorial covering the author's preparation steps for newly installed Proxmox servers, including updates, storage, networking and VLANs, ISO uploads, clustering, PCI passthrough, cloned Linux machine cleanup, and alerts. ### Source excerpt After setting up my Proxmox servers, there are a few things I do before I use them for their intended purpose.This ranges from updates, to storage, to networking and VLANS, to uploading ISOs, to clustering, and more.Join me as we pick up where the rest of the proxmox tutorials stop, and that's everything you need to do to make these production ready (and maybe a bonus item too). 📺 Watch Vide... ## High availability with nats-streaming-server (clustering) DevFeed: [High availability with nats-streaming-server (clustering)]() Original publisher: [Read original article]() Author: Carlos Alexandro Becker Published: 2019-05-16T00:00:00Z Content type: tutorial Language: en Sources: [Carlos Becker]() Topics: [Streaming](), [clustering](), [Availability](), [Raft]() Tags: [clustering](), [fault-tolerance](), [high-availability](), [leader-election](), [raft](), [streaming]() ### AI overview A practical guide to configuring a high-availability nats-streaming-server cluster. It compares fault-tolerance and clustering approaches, then demonstrates clustering NATS and nats-streaming-server with three nodes, RAFT leader election, replicated writes, and peer configuration. ### Source excerpt I wanted to set up a high available nats-streaming-server cluster, but couldn't find a "quick" guide on how to do it. ## Topic Modeling: Optimizing for Human Interpretability DevFeed: [Topic Modeling: Optimizing for Human Interpretability]() Original publisher: [Read original article]() Author: Alyssa Wisdom Published: 2017-12-20T22:38:24Z Content type: tutorial Language: en Sources: [Square Corner Blog](), [Square Corner Blog RSS Feed]() Topics: [Machine learning](), [data](), [information retrieval](), [tokenization]() Tags: [article](), [clustering](), [data](), [data-science](), [machine-learning](), [tokenization]() ### AI overview This article explains topic modeling as an unsupervised machine-learning method for identifying latent topics in large text collections. It discusses evaluating topic models, improving interpretability, preprocessing text, choosing the number of topics, and using document-term matrices. ### Source excerpt Optimizing for Human Interpretability ## So You Have Some Clusters, Now What? DevFeed: [So You Have Some Clusters, Now What?]() Original publisher: [Read original article]() Author: Inna Kaler Published: 2017-11-09T00:11:19Z Content type: tutorial Language: en Sources: [Square Corner Blog](), [Square Corner Blog RSS Feed]() Topics: [data](), [Algorithms](), [datasets](), [systems]() Tags: [algorithm](), [analysis](), [blog-post](), [clustering](), [data-science](), [dataset](), [features](), [insights]() ### AI overview This article explains how Square used unsupervised clustering to analyze seller behavior. It describes how manually selecting familiar signals embedded existing assumptions into the data, limiting new insights, and then outlines rebuilding clusters with thousands of behavioral signals generated by Square's risk systems to reduce that bias. ### Source excerpt How to Add Value to Your Clusters ## Faster Snowflake Queries through Clustering DevFeed: [Faster Snowflake Queries through Clustering]() Original publisher: [Read original article]() Author: Michal Klos Published: 2017-03-13T14:17:01Z Content type: article Language: en Sources: [Localytics]() Topics: [data-processing](), [Databases](), [Query (disambiguation)](), [Amazon Redshift]() Tags: [clustering](), [data-processing](), [database](), [databases](), [join](), [mpp](), [mysql](), [partitioning](), [performance](), [redshift](), [s3](), [scale](), [snowflake](), [time-series]() ### AI overview This article explains how Snowflake clustering and re-clustering can improve query performance by helping the query engine prune unneeded data. It discusses choosing cluster keys, comparisons with MPP database techniques, and the caveat that naturally ordered time-series data may not need clustering. ### Source excerpt At Localytics we have petabytes of data that needs to be served at low latencies and we use Snowflake in our mix of data processing technologies. Snowflake, like many other MPP databases, has a way of partitioning data to optimize read-time performance by allowing the query engine to prune ## Cluster Analysis Using K-means Explained DevFeed: [Cluster Analysis Using K-means Explained]() Original publisher: [Read original article]() Author: umer Published: 2017-02-19T00:00:00Z Content type: tutorial Language: en Sources: [Code Ahoy - Articles]() Topics: [data](), [Machine learning](), [Algorithm](), [datasets]() Tags: [algorithm](), [blog-post](), [clustering](), [data](), [dataset](), [machine-learning]() ### AI overview This tutorial explains clustering and k-means, including how k-means assigns data points to clusters, updates cluster centroids, and repeats the process to minimize the total distance within clusters. It also discusses applications and the importance of choosing a clustering method based on requirements and the dataset. ### Source excerpt Clustering or cluster analysis is the process of dividing data into groups (clusters) in such a way that objects in the same cluster are more similar to each other than those in other clusters. It is used in data mining, machine learning, pattern recognition, data compression and in many other fields. In machine learning, it is often a starting point. In a machine learning application I built couple of years ago, we used clustering to divide six million prepaid subscribers into five clusters and then built a model for each cluster using linear regression. The goal of the application was to predict future recharges by subscribers so operators can make intelligent decisions like whether to grant or deny emergency credit. Another (trivial) application of clustering is for dividing customers into groups based on spending habits or brand loyalty for further analysis or to determine the best promotional strategy. There are various models and techniques for cluster analysis. When I first started, I was mistakenly searching for 'the best clustering model or technique.' I wasn't aware that there is no universal best algorithm and the choice depends on your requirements and the dataset. There are density-based, graph based or centroid based clustering models. We finally settled on a clustering technique called k-means. This blog post is a brain-dump of everything I've learned about clustering and k-means so far. K-means K-means is a very simple and widely used clustering technique. It divides a dataset into 'k' clusters. The 'k' must be supplied by the users, hence the name k-means. It is general purpose and the algorithm is straight-forward: We call the process k-means clustering because we assume that there are k clusters, and each cluster is defined by its center point -- its mean. To find these clusters, we use Lloyd's Algorithm: we start out with k random centroids. A centroid is simply a datapoint around which we form a cluster. For each centroid, we find the datapoints ## Clustering Similar Stories Using LDA DevFeed: [Clustering Similar Stories Using LDA]() Original publisher: [Read original article]() Author: https://www.linkedin.com/in/arnab-bhadury-a6304768 (Arnab Bhadury) Published: 2017-02-08T00:00:00Z Content type: tutorial Language: en Sources: [Flipboard]() Topics: [Algorithms, Complexity](), [Algorithms](), [data]() Tags: [algorithm](), [algorithms](), [clustering](), [trimming]() ### AI overview This article explains how Flipboard uses a clustering algorithm and Latent Dirichlet Allocation (LDA) to group related stories into multi-source story roundups. It discusses challenges including high-dimensional text representations, differing word usage, unknown cluster counts, and the need for fast updates. ### Source excerpt There is more to a story than meets the eye, and some stories deserve to be presented from more than just one perspective. With Flipboard 4.0, we have released story roundups, a new feature that adds coverage from multiple sources to a story and provides you with a fuller picture of an event. Here's how it looks: With our scale of millions of articles and constant stream of documents, it's impossible to generate these roundups manually. So, we have developed a clustering algorithm that's both fast and scalable, and in this blog post, I will explain how we create these roundups on Flipboard. Why is this difficult? Although there are many sophisticated automatic clustering algorithms, such as K-means or Agglomerative clustering, story clustering is a non-trivial problem. Because each text document can contain any word from our vocabulary, most text document representations are extremely high-dimensional. In high-dimensional spaces, even basic clustering or similarity measures fail or are very slow. Additionally, two very similar documents often have very different word usages. For example, one article may use the term kitten and another may use feline, but both articles could be referring to the same cat. Furthermore, we don't know the number of roundups that we expect to see beforehand. This makes it difficult for us to directly use parameteric algorithms such as K-means. Our clustering algorithm also needs to be fast and easy to update, because there is a constant stream of documents coming into our system. Overview Since even the most basic distance measures fail in high dimensions, the first thing we do is lower the problem's dimensionality. We represent each of our text documents as a bag-of-words, and remove stop-words and rare words from our vocabulary. Even after an aggressive trimming, the documents are still very high-dimensional. We then we use Latent Dirichlet Allocation (LDA) to further lower the documents' dimensionality. We use LDA because this algorith ## Big Dimensions, and What You Can Do About It DevFeed: [Big Dimensions, and What You Can Do About It]() Original publisher: [Read original article]() Published: 2016-02-08T10:00:14Z Content type: article Language: en Sources: [Jeremy Kun]() Topics: [data](), [math](), [clustering](), [datasets](), [Algorithms, Complexity]() Tags: [algorithm](), [clustering](), [data](), [dataset](), [math](), [techniques]() ### AI overview The article uses movie-title data to illustrate the curse of dimensionality: as vector dimensions grow, geometric intuition and clustering algorithms can perform poorly. It introduces dimension reduction as a way to condense high-dimensional data while retaining important geometric properties, and discusses embeddings and their mathematical guarantees. ### Source excerpt Data is abundant, data is big, and big is a problem. Let me start with an example. Let's say you have a list of movie titles and you want to learn their genre: romance, action, drama, etc. And maybe in this scenario IMDB doesn't exist so you can't scrape the answer. Well, the title alone is almost never enough information. One nice way to get more data is to do the following: ## CHAR(14): Conference on Clustering, High Availability, Replication, and Distributed Architectures DevFeed: [CHAR(14): Conference on Clustering, High Availability, Replication, and Distributed Architectures]() Original publisher: [Read original article]() Published: 2014-07-07T22:00:00Z Content type: article Language: en Sources: [Dimitri Fontaine]() Topics: [migration](), [PostgreSQL](), [Availability](), [clustering](), [Replication](), [MySQL](), [parallel](), [CSV](), [Structured-data]() Tags: [availability](), [clustering](), [csv](), [database-migration](), [distributed](), [high-availability](), [migration](), [mysql](), [parallel](), [postgresql](), [replication]() ### AI overview CHAR(14) is an international conference focused on clustering, high availability, replication, and parallel, distributed, and grid architectures. The page also describes a lightning talk about pgLoader and its support for input formats and database migration to PostgreSQL. ### Source excerpt A lightning talk about pgLoader and the wide range of input formats it accepts, from CSV to advanced database migration from MS SQL or MySQL to PostgreSQL. [Next page]()