Requisitos Mencionar Requerimientos técnicos: (20 min) Manual Ivan (README)
Crear equipos
Asignar cuentas de Azure por equipos
Crear instancias para participantes
Cuenta en Weights and Biases
Revisar el repo
Data Management and CI/CD in MLOps (30 min) –Alexis
Handling Data Drift –Gerardo — Demo
Pipelines (Alexis) – Demo
Hyperparameter Tuning and Model Verification (30 min)
Data Versioning and Validation
Ivan hablar sobre Selección de modelo DVC y DagsHub, breve ejemplo de código
Demo: DVC y DagsHUB (metashift? pendiente)
Monitoring Model Performance
Alexis hablar sobre Weights and Biases
PRACTICA: Weights and Biases
https://docs.ray.io/en/latest/tune/examples/tune-wandb.html
(Intercambiable) Automatic Hyperparameter Tunning Techniques and Optimization
Ivan optimización de hyperparámetros con raytune
PRACTICA: RayTune (with MLFlow) https://docs.ray.io/en/latest/tune/examples/tune-mlflow.html
Demo (30 min)
Ivan Demo MLFlow
(Intercambiable) Ivan Tuneo de hyperparámetros en MLFlow
Case Study: MLOps for Kidney Stones Image Analysis (20 min)
Application of MLOps in Medical Imaging
Gilberto hablar sobre cómo pudiera mejorar la ciencia con MLOps (reproducibilidad, escalabilidad)
Challenges and Solutions
Gilberto hablar sobre los retos en los datasets de CV-INSIDE
Course Wrap-up and Next Steps
Gilberto de tendencias
Expected Target Audience
Computer Science Students: Undergraduate or graduate students interested in concentrating in machine learning operations and computer vision who are studying computer science, data science, or a similar discipline.
Machine Learning Practitioners: Professionals in the machine learning industry who wish to improve their knowledge of and abilities in MLOps and its use in computer vision.
Data Scientists: Data scientists who want to comprehend the practical facets of implementing machine learning models, particularly in the context of computer vision.
Software Engineers: Software engineers who wish to comprehend the lifespan of machine learning models, from development to deployment and maintenance, and who are making the transfer into data-centric roles.
Medical Professionals: Healthcare industry experts that are curious about how MLOps might be used in medical imaging and other computer vision applications in the field.
Links and Materials
Lakshmanan, V., Robinson, S., & Munn, M. (2020). Machine learning design patterns. O'Reilly Media.
Made With ML https://madewithml.com/
MLOps Papers https://github.com/visenger/awesome-mlops/blob/master/papers.md
Microsoft MLOps Examples https://github.com/microsoft/MLOps
A curated list of references for MLOps https://github.com/visenger/awesome-mlops
How to avoid machine learning pitfalls: a guide for academic researchers: https://arxiv.org/abs/2108.02497
Hyperparameter Tunning by Google: https://github.com/google-research/tuning_playbook
Notes and Considerations
This tutorial was recently conducted at the Mexican Congress of Artificial Intelligence 2023 (COMIA). We had approximately 20 attendees from various universities, the majority of whom were PhD students.
Github Repo: https://github.com/Ivanrs297/micai-2023-mlops-tutorial
Github Repo COMIA 2023 (Past tutorial): https://github.com/Ivanrs297/patrones-mlops-comia
Tutorial Certificate: https://photos.app.goo.gl/jZJxK3JCtqicYwL6A