PodcastMost Underrated MLOps TopicsThreads / The platform team / 2021
The platform team in 2021
40 sessions
PodcastMachine Learning Feature Store Panel DiscussionClaimSimarpal Khaira says Intuit moved toward a centralized feature platform because separate teams were creating their own storage mechanisms, which made production pipelines less robust and reliable.10:30
Meetup2 tools = 90% operational MLPushed backThe panel distinguished Kubernetes as a common infrastructure substrate from Kubeflow as the machine-learning platform built on it.33:05
PodcastMLOps Engineering Labs Recap, Part 2Pushed backLaszlo viewed direct Kubernetes use as too difficult for an average data scientist, while Artem argued that Kubernetes is highly useful and should be exposed through simpler platforms.39:23
MeetupHow to Avoid Suffering in MLOps/Data Engineering RoleClaimIgor Lushchyk advises people to inspect tools internally and understand how they work, why they work that way, and how data flows through them.9:19
MeetupOperationalizing Machine Learning at a Large Financial InstitutionClaimDaniel Stahl says he leads tooling, platforming, and practices for Regions Bank's data scientists, supported by engineering and data engineering teams.2:12
PodcastMLOps InvestmentsPushed backSarah Catanzaro does not favor end-to-end ML platforms as a general solution because they can be poorly suited to particular workflows, while point-product sprawl also creates maintenance and integration costs.30:24
MeetupMLOps Community 1 Year Anniversary!ClaimPeople should start with a small, feasible solution informed by user needs and iterate on it instead of adopting every available tool at once.17:06
PodcastMachine Learning at AtlassianClaimAtlassian moved its feature and MLOps platform to Tecton, which provided a more mature version of a streaming feature system and enabled data scientists to create historical training sets and online features with less engineering support.37:12
PodcastWar Stories Productionising MLClaimThe model's staging data had transformations that were not reproduced in production, and additional undocumented transformations occurred before data entered the pricing platform.19:38
MeetupLearnings from Live Coding: An MLOps Project on TwitchClaimFelipe Campos Penha started streaming on Twitch after seeing data scientists in Brazil use the platform, and he began live-coding the Greenhouse project there.16:10
PodcastMLOps: A Leader's PerspectiveClaimQuby used an internal Data Science 101 program to teach people across the business about data and machine learning.21:19
SGT Model Deployment Working Group May 19, 2021 MeetingClaimMany reasonably sized enterprises have built their own internal ML platform while also providing a unified experience for end users.2:42
MeetupBuilding an ML Platform from Scratch: Live Coding SessionClaimAlon Gubkin says an ML platform can include data collection and versioning, feature stores, training orchestration, experiment management, packaging, deployment, and model serving.4:40
PodcastMaturing Machine Learning in EnterprisePushed backKyle Gallatin rejects the idea that one platform can fit every machine learning use case and argues for integration-first platforms with room for customization.17:02
MeetupProject/Product Management for MLOpsPushed backKorri Jones rejects the idea that building an internal platform guarantees that users will adopt it.17:26
PodcastTour of Upcoming Features on the Hugging Face Model HubClaimJulien Chaumond says the Hugging Face platform lets users and organizations share and discover models and data sets, and increasingly train and deploy models.5:49
SGT Model Deployment Working GroupPushed backmlctl should provide neutral interconnection layers rather than become another horizontal platform that owns every service.40:05
PodcastAggressively Helpful Platform TeamsPushed backStefan Krawczyk said Stitch Fix did not need to replace Model Envelope with MLflow or TensorFlow Extended because its internal integration and model-management needs were different.32:12
PodcastCreating MLOps StandardsClaimAlex Chung says enterprises often need to combine existing internal systems with SageMaker and other tools instead of using one end-to-end platform.2:17
mlctl and Hydrosphere Open Source MLOps Libraries DemoClaimMlctl lets platform teams choose different infrastructure engines for different jobs through provider YAML files, while data scientists and ML engineers modify job YAML files.2:41
MeetupBuilding Machine Learning Models into Docker ImagesClaimModels can be containerized for batch prediction by sending multiple requests or by using a batch platform such as Modzy.22:20
PodcastVector Similarity Search at ScalePushed backThe host suggested vector search was new, while Dave Bergstein clarified that major companies such as Amazon and Google had already been using it and that the newer development was broader adoption by other companies.17:17
PodcastThe Future of ML and Data PlatformsPushed backMichael Del Balso distinguishes platform centralization from build-versus-buy decisions instead of treating them as one choice.18:37
MeetupDoing MLOpsClaimCloud platform tools such as SageMaker can handle much of the infrastructure needed to train and deploy machine learning models.11:23
MeetupMLOps at Volvo CarsClaimLeonard Aukea says Volvo Cars has not yet adopted Feast and is still considering how to connect feature versioning with LakeFS data versioning.21:04
PodcastPyTorch: Bridging AI Research and ProductionClaimDmytro Dzhulgakov said machine learning infrastructure had to shift from hands-on enablement to self-service platform building as the number of developers and use cases grew.14:05
PodcastThe Future of Data Science Platforms is AccessibilityPushed backSkylar Payne pushes back on the assumption that companies need to adopt Kubeflow or a large modern data stack simply because it is considered best practice.4:12
PodcastML Stepping Stones: Challenges & Opportunities for CompaniesPushed backJohn Crousse argues against relying on a single all-in-one machine learning platform because it may not interoperate with the rest of an enterprise system.24:50
PodcastBuilding for Small Data Science TeamsClaimSpotHero did not yet have a machine learning platform and was still designing its architecture and evaluating vendors.14:37
MeetupSetting up an ML Platform on GCP: Lessons LearnedClaimMefta Sadat's team builds the machine learning and data platform that data scientists use to run experiments, deploy models, and manage data ingestion and engineering.2:12