Description:

There is a usual notion that once the Machine Learning model is developed, the deployment and integration with the product will just be a small part of it. In reality, Model development is just a small piece of the entire process, the entire deployment pipeline requires brainstorming.

In this talk, we will discuss MLOps, the last & crucial step of the Machine Learning Lifecycle. We will discuss some Key Challenges, their importance, and how to make sure they are taken care of while deploying the model.

The concepts include:

1. Concept Drift & Data Drift
2. Importance of Documentation
3. Cross-Team Collaboration
4. Logging Data for Review
5. Model versioning and governance
6. Importance of Monitoring (Input Metrics like missing values, Software Metrics like memory consumed, server load, Output Metrics like null predictions)
7. Security & Privacy of Data
8. Deciding on Realtime Vs Batch Prediction

After discussing the key concepts, we will take a look at the different modes in which a Machine Learning Model can be deployed depending on acceptable downtime, operation cost, human involvement, ease of rollback & need for testing in production.

Deployment strategies:

1. Recreate Deployment
2. Shadow Deployment
3. Canary Deployment
4. A/B Testing Deployment
5. Blue-Green Deployment

Lastly, we will discuss some of the learnings I have learned in the process of Deploying ML Models to production.

Key Takeaways:

1. Model development is just a fraction of the overall process - MLOps encompasses the crucial step of deploying and integrating Machine Learning models with the backend.
2. Understand the key concepts in MLOps, such as concept drift, data drift, monitoring (input, software, and output metrics), logging data, handling data bias, and ensuring data security and privacy.
3. Consider the decision between real-time and batch prediction, and recognize the importance of documentation and cross-team collaboration in successful model deployment.
4. Explore different deployment strategies based on factors like downtime tolerance, operation cost, human involvement, rollback ease, and need for testing in production.
5. Gain insights from practical experience in deploying ML models to production, learning from challenges and best practices for successful deployment.

Brief Outline:

1. Introduction to MLOps & its significance [3 minutes]
2. Key Challenges, Importance & How to Tackle Them [13 mins]
3. Deployment Strategies [10 mins]
4. Best Practices [2 mins]
5. Q&A [2 minutes]

Prerequisites:

Familiarity with Machine Learning basics will be helpful but not mandatory. Attendees should have a basic understanding of the Machine Learning Lifecycle.

Content URLs:

Experience Deploying ML Model to Production

• Blog: https://blog.shlokashah.com/experience-deploying-first-ml-model-to-production
• HackerNoon: https://hackernoon.com/my-first-experience-deploying-ml-model-to-production

ML Model Deployment Strategies

• Blog: https://blog.shlokashah.com/ml-model-deployment-strategies
• HackerNoon: https://hackernoon.com/5-effective-deployment-strategies-for-machine-learning-models

Speaker Info:

Shloka works at HackerRank as a Software Development Engineer II, demonstrating her passion for Problem-Solving. She is a part of the HackerRank Labs team where she focuses on building new products and finding their product market fit. Her main areas of interest revolve around Software Development, Backend Development, and Machine Learning. Over the past 2.5 years, she has gained valuable experience building scalable Web Crawlers & Scrapers using Python, scalable applications using Ruby on Rails, and gained hands-on experience in developing and productionizing various Machine Learning models to solve complex problems. She takes great pleasure in developing her own solutions using a data-driven methodology.

Speaker Links:

• Personal Blog: https://blog.shlokashah.com/
• HackerNoon: https://hackernoon.com/u/shlokashah
• Linkedin: https://www.linkedin.com/in/shloka-shah-98625015a/
• Twitter: https://twitter.com/shlokashah0603
• Github: https://github.com/shlokashah
• Youtube: https://www.youtube.com/@shlokashah

Shloka shares her experiences on her personal blog, earning her the HackerNoon Contributor of the Year award. In addition to her writing achievements, she mentors aspiring software developers on various topics related to Software Development. Shloka has spoken at events such as the Ruby on Rails Global Summit by Geekle, the Pune FOSS Conference, and Mumbai FOSS meetup. She has served as a judge and mentor in multiple Hackathons. Furthermore, she actively contributes to the community as a mentor with Rails Girls Bangalore and volunteers with FOSS United Bangalore and FOSS United Mumbai. Shloka also contributes her expertise as a member of the CFP review team at FOSS United. She is also a volunteer in the Content Team for PyCon India 2023.