whoami
● Solutions Architect @ cnvrg.io
● = built by data scientists, for data scientists to help teams:
○ Get from data to models to production in the most efficient and fast way
○ Bridge science and engineering
○ Automate MLOps
def agenda(30 mins):
● Introduction to MLOps
● What can MLOps do?
● A world without MLOps
● Elements of an MLOps solution
● MLOps with cnvrg.io
MLOps.intro()
● Is it a practice? A job? A platform?
● Collaboration & communication between data science & engineering
professionals
● Stabilizes pipeline for entire ML lifecycle
● Create a real Machine Learning Cycle
● Ease friction at every point of the pipeline Research
Data ExplorationDeployment
Training
if not MLops:
+ Not all data scientists come from a DevOps background and are not
always following best practices
+ Enterprise ML development is slow & hard to standardize/scale
+ So many Tools & Frameworks
= Many delays in process
= Friction between teams
= Wasted resources
= Tough to iterate on models
= No standardization in industry
if MLOps == True:
● Deploy faster, easier and more often
● Centralize model tracking, versioning and monitoring
● Integrate different technologies together
● Reduce friction between science and engineering
○ Enhance collaboration
● Move towards standardization in the field
● Support continual learning with MLOps (CI/CD)
○ Webinar on CI/CD for Machine Learning
MLOps.elements()
1. Data management
2. Collaboration & communication
3. Model tracking, version and management
4. One-click experiment execution and model deployment
5. Unifies team no matter what language, framework, or provider
6. Features to assist every role in the team
a. Data Scientist
b. Engineer
c. Business Managers
7. Continual learning - the automatic deployment and retraining of models
IN production
MLOps.now()
● Many half-baked solutions
● Very early open-source technologies
● Requires deep know how to even implement
cnvrg.demo()
webinar.summary()
● Data Science is a quickly advancing field
● The lifecycle is inefficient within most firms
● MLOps is a still developing field that should be quickly adopted
● As a community we need to invest in standardizing MLOps
● You can incorporate MLOps with open source tools or full platform
solutions
Thanks!
https://cnvrg.io
info@cnvrg.io
+972506600186
Next webinar!

MLOps for production-level machine learning

  • 2.
    whoami ● Solutions Architect@ cnvrg.io ● = built by data scientists, for data scientists to help teams: ○ Get from data to models to production in the most efficient and fast way ○ Bridge science and engineering ○ Automate MLOps
  • 3.
    def agenda(30 mins): ●Introduction to MLOps ● What can MLOps do? ● A world without MLOps ● Elements of an MLOps solution ● MLOps with cnvrg.io
  • 4.
    MLOps.intro() ● Is ita practice? A job? A platform? ● Collaboration & communication between data science & engineering professionals ● Stabilizes pipeline for entire ML lifecycle ● Create a real Machine Learning Cycle ● Ease friction at every point of the pipeline Research Data ExplorationDeployment Training
  • 5.
    if not MLops: +Not all data scientists come from a DevOps background and are not always following best practices + Enterprise ML development is slow & hard to standardize/scale + So many Tools & Frameworks = Many delays in process = Friction between teams = Wasted resources = Tough to iterate on models = No standardization in industry
  • 6.
    if MLOps ==True: ● Deploy faster, easier and more often ● Centralize model tracking, versioning and monitoring ● Integrate different technologies together ● Reduce friction between science and engineering ○ Enhance collaboration ● Move towards standardization in the field ● Support continual learning with MLOps (CI/CD) ○ Webinar on CI/CD for Machine Learning
  • 7.
    MLOps.elements() 1. Data management 2.Collaboration & communication 3. Model tracking, version and management 4. One-click experiment execution and model deployment 5. Unifies team no matter what language, framework, or provider 6. Features to assist every role in the team a. Data Scientist b. Engineer c. Business Managers 7. Continual learning - the automatic deployment and retraining of models IN production
  • 8.
    MLOps.now() ● Many half-bakedsolutions ● Very early open-source technologies ● Requires deep know how to even implement
  • 9.
  • 10.
    webinar.summary() ● Data Scienceis a quickly advancing field ● The lifecycle is inefficient within most firms ● MLOps is a still developing field that should be quickly adopted ● As a community we need to invest in standardizing MLOps ● You can incorporate MLOps with open source tools or full platform solutions
  • 11.
  • 12.