MeetupOur 1st MLOps MeetupClaimModel monitoring differs from regular software monitoring because production data is unlabeled, so a model can have normal latency and error rates while producing incorrect results.23:00
Threads / Serving and deployment / 2020
Serving and deployment in 2020
32 sessions
MeetupOur 1st MLOps Meetup
MeetupBuilding an ML Platform at SurveyMonkeyPushed backShubhi Jain said SurveyMonkey's inference architecture did not run on Kubernetes.34:48
MeetupTrueLayer's MLOps PipelinePushed backAlex Spanos favors simple interpretable algorithms in regulated financial services, rather than moving to more complex models when interpretability is reduced.19:45
MeetupOptimizing Your ML Workflow with Kubeflow 1.0ClaimKubeflow's mission is to make it easy for everyone to develop, deploy, and manage portable, distributed, and scalable machine learning systems on Kubernetes.7:07
Meetup10 Years Deploying ML in the Enterprise: The Inside Scoop!Pushed backCharles Martin disputes the idea that running a machine learning model inside a database such as SQL Server is a simple solution.27:20
MeetupMachine Learning at Scale in Mercado LibrePushed backCarlos de la Torre argued that Fury Data Apps should not automate model deployment because responsibility for production deployment must remain explicit.45:30
MeetupFury Platform and Fury Data Apps at Mercado LibreClaimDifferent teams initially used separate approaches, including IBM Watson, cloud services, and independently built pipelines.0:50
PodcastServing Models with KubeflowClaimServing a model means taking a trained machine learning model and making its predictions available to users when they need them.2:29
MeetupRunning a Fintech on Machine LearningClaimCaique Lima works on tools, model deployment, and related infrastructure in a horizontal team at Nubank.1:23
PodcastDifferent Ways of Serving ML ModelsPushed backThe author says the microservice model-serving approach should be an aspiration, but Byron Allen says it is not suitable for everyone and that simpler approaches can be better.1:00:00
MeetupDeep Dive on Paperspace ToolingClaimPaperspace provides notebooks, job and experiment execution, model tracking, artifact management, and inference services.14:41
MeetupHow to Become a Better Data Scientist: The Definitive GuidePushed backKubernetes should be learned sufficiently for a data scientist to use a service and ask for help, rather than requiring data scientists to become Kubernetes experts.13:20
MeetupHow to Leverage ML Tooling EcosystemClaimThe production workflow uses Airflow for pipelines, Seldon for deployment, and Grafana for monitoring.23:09
PodcastMLOps: Isn't That Just DevOps?ClaimMLOps is what is necessary to make the machine-learning build, deploy, and monitor lifecycle as smooth and safe as possible.3:28
Introducing MLOps Engineering Labs - Join us to better your skillsClaimThe labs will explore batch, real-time, and streaming system architectures while promoting reproducibility, automation, and code modularity.10:21
MeetupScaling ML Capabilities in Large OrganizationsPushed backJoe argues that organizations should avoid buying a model serving platform when a simpler, opinionated solution is sufficient.57:25
MLOps Cubonacci workshopClaimJan says deploying a model is the point at which an organization can integrate it with other systems and begin capturing the value created by its development.6:30
PodcastMLOps and DevOps, Parallels and DeviationsClaimMLOps differs from DevOps mainly in the processes used to produce and deploy machine-learning models.5:34
MeetupBuilding Say Less: An AI-Powered Summarization AppPushed backDemetrios Brinkmann initially interpreted the 'What's next' slide as discussing advanced deployment techniques for Say Less, while Yoav Zimmerman clarified that those techniques were mainly relevant to Model Zoo.39:56
PodcastContinuous Delivery and Automation Pipelines in Machine Learning, Part 1ClaimGoogle treats machine learning system development as experimental and separates training from serving while also including data extraction, feature engineering, and validation.13:01
MeetupStreaming Machine Learning with Apache Kafka and Tiered StoragePushed backKai Waehner said Kafka is not the strongest system for MapReduce-style batch processing.41:43
MeetupBring Your On-Prem ML Use Cases to Production on Google Cloud using KubeflowClaimThe pipeline blocks cover data preparation, hyperparameter tuning, model training, model deployment, model prediction, and model explainability.6:57
PodcastAnalyzing "Continuous Delivery and Automation Pipelines in ML", Part 3Pushed backThe speakers question the paper's broad implication that deployed models generally break and that frequent retraining is always needed.23:54
PodcastMLOps Coffee Sessions #14 Conversation with the Creators of DaskClaimHugo Bowne-Anderson founded Coiled to build scalable products and hosted deployments for data scientists and machine-learning practitioners.1:40
PodcastMLOps + Machine LearningClaimJames Sutton explained that embeddings are compact internal representations of information that preserve useful conceptual relationships.9:08
MeetupScalable Python for Everyone, EverywhereClaimDask breaks high-level computations into smaller tasks and executes them across parallel resources while preserving the result of the original computation.10:48
MeetupOperationalize Open Source Models with SAS Open Model ManagerClaimIvan Nardini says operationalizing analytics requires an end-to-end cycle covering governance, deployment, monitoring, retraining, and decisioning.7:38
PodcastLuigi in ProductionClaimLuigi Patruno says the algorithms took almost no time compared with the time required for data structures, data pipelines, deployment, and monitoring.5:48
PodcastMonzo Bank - An MLOps Case StudyClaimMonzo focuses machine learning on several banking areas, including customer service, lending, credit scoring, financial crime detection, and customer experience.5:57
PodcastA Conversation with Seattle Data GuyClaimBenjamin Rogojan is writing a series that covers the data lifecycle from ETLs and data warehouses through model deployment.2:40
PodcastSRE for ML InfraClaimTodd Underwood strongly agrees that teams can start with a serving system before the rest of the machine learning pipeline is complete.19:36