True ML Talks #3 | Machine Learning Workflow @ Facebook
True ML Talks #3 | Machine Learning Workflow @ Facebook
TrueMLTalks is a video series in which we interview machine learning leaders and explore their companies' ML stacks. We'll get an in-depth look at handling ML pipelines and developing best practices along the way. In our segment, we talk with ML-first organisations like Gong, StichFix, SalesForce, Facebook, Simpl, and many more.
Aditya Kalro is currently a Senior Engineering Manager at Google in the Identity team, and prior to this, Aditya was at Facebook, where he led the build-out of the entire ML workflow management platform at Facebook called FBLearner Flow. And we'll talk about it in more detail in today's call.
Our conversations with Aditya will cover below aspects:
✅Overview of FBLearner Flow, a pipeline management system that facilitates experimentation, training, and deployment of models.
✅A/B Testing and Shadow Testing in large-scale systems.
✅Bridging the gap between Research and Production.
✅Optimizing cost and latency in AI Inference.
✅The architecture of FBLearner Flow.
✅Bridging the Gap Between Software and ML Deployment Platforms.
✅Importance of Monitoring and Distributed training.
✅Core principles for building an ML system for scale.
Don't miss this opportunity to learn from one of the brightest minds in the ML industry. Tune in to the next episode of TrueMLTalks now!
About Our Guest
Aditya is presently working as a member of Google's identity team, where he is attempting to reduce abuse prevention in Google Cloud using AI tools. Prior to this, he worked at Facebook for 7.5 years on AI infrastructure, including FBLearner flow as well as inference, feature engineering, labeling infrastructure, and a high-level platform to tie all of these pieces together to make it easier for ML practitioners to use.
⚡Reach out to Aditya at below link -
/ adityakalro
About TrueFoundry
TrueFoundry is a cross-cloud ML deployment PaaS that helps organizations accelerate developer workflows for model testing and deployment while maintaining full security and control for the Infra/DevSecOps team. We enable machine learning teams to deploy and monitor models in 15 minutes with 100% reliability and scalability, allowing them to save money and release models to production faster, enabling actual business value to be realized. We deploy on the customer's infrastructure, ensuring data privacy and other security concerns are addressed.
⚡Concept Video: • TrueFoundry Explainer Video
⚡ A short ~4 min demo of the platform: https://t.ly/0fk1
Read our blog: https://blog.truefoundry.com/
Subscribe to our newsletter: https://t.ly/oORp
00:00 start
02:16 Insight on the scale that FBLearner was supporting, along with the kind and number of models.
04:15 The evolution of FBLearner, starting with the workflow, its core pieces within the system, and how did it tie into the inferencing engine?
08:51 Workflow and its operators
11:27 Resource management
14:59 Bin packing of machines and the kind of trade-offs done to be able to support the running of the operators
17:03 How was such a huge library of different models and operators maintained?
19:15 How did the evaluation system come about, and what were the components that had to be built around it?
22:51 Canary or shadow testing set up and the right A/B testing framework
27:09 Architectural design to tackle the problem of cost optimization versus latency
34:09 Commonalities between a software engineer platform and an ML development platform
39:45 Distributed training
44:12 Other major parts surrounding FBlearner Flow
44:48 Major principles to keep in mind while building an ML system for scale
47:08 The security aspect involved in the entire journey of a model
48:10 An ideal description of an ML platform and its infrastructure pieces according to Aditya
Try our platform now - https://app.truefoundry.com/company-r...
Book a personalised demo - https://www.truefoundry.com/book-demo...
#machinelearning #datascience #technology
Watch video True ML Talks #3 | Machine Learning Workflow @ Facebook online without registration, duration hours minute second in high quality. This video was added by user TrueFoundry 30 March 2023, don't forget to share it with your friends and acquaintances, it has been viewed on our site 737 once and liked it 14 people.