Getting Started with TensorFlow on Google Cloud

Mariam Ali Aslam - GDE Firebase
TensorFlow
Google Cloud

Introduction
● Mariam Ali Aslam
● AI SDET Architect at RAKBank
● Google Developer Expert for Firebase
● Manager GDG Cloud Emirates
● Community Champion BrowserStack
● Ambassador Cypress.io
● Ambassador K6
● Community Leader WomenInTesting
● https://www.linkedin.com/in/mariam-aslam/

What is AI?
● AI stands for Artificial Intelligence.
● Teaching computers to think like us seemed like a good idea at the time.
Spoiler alert: They’re getting pretty good at it!
● The term "Artificial Intelligence" was coined in 1956 long before the
internet.

What is TensorFlow?
● Imagine teaching your computer to think (or at least pretend to).
● It's the secret sauce behind everything cool Google does (like suggesting
videos of cats).
● TensorFlow has over 160,000 stars on GitHub. That's more stars than
your last science project!

What is TensorFlow?
1. Open-source machine learning framework developed by Google.
2. Supports deep learning, neural networks, and a wide range of machine
learning tasks.
3. Extensive community and industry support.
4. Flexibility to run on various platforms (CPUs, GPUs, TPUs).
5. Comprehensive ecosystem with tools for model building, training, and
deployment.

Why Google Cloud for TensorFlow?
● Easily scale from a single machine to distributed systems.
● Seamless integration with Google Cloud services like BigQuery, cloud
storage, and AI platforms. (Lego Blocks)
● Pay-as-you-go model with options to reduce costs
● Fully managed services for TensorFlow, simplify deployment and scaling.

What does TensorFlow help with

TensorFlow Architecture for Model Building
1. Data Preprocessing: prepare data for the purpose of feeding it to the model
that you need to build.
2. Model Building: start model building, create your model by using various
algorithms.
3. Model training & evaluation: start training and evaluating it to check
whether it generates accurate results or not.

Using TensorFlow: Local vs. Cloud
Aspect TensorFlow Locally TensorFlow on Google Cloud
Setup &
Installation
Install via pip on your local
machine.
Set up using Google Cloud AI
Platform or Vertex AI.
Data Handling Load and preprocess datasets
locally.
Store and preprocess data using
Google Cloud Storage.
Model Training Train models using local
CPU/GPU resources.
Utilize Google Cloud GPUs/TPUs
for faster, scalable training.
Scalability Limited to local hardware
capabilities
Scalable with cloud resources,
including distributed training.
Flexibility Ideal for small projects or
development.
Best for large-scale, production-
level models.
Cost Free for local tasks; no cloud
costs.
Pay-as-you-go based on cloud
resource usage.

Setting Up TensorFlow on Google Cloud
● Create a Google Cloud Account.
● Set up a new project
● Install Google Cloud SDK.
● Launch a virtual machine (VM).
It can also be used with Python, and JavaScript.

Training Models on Google Cloud AI Platform
● AI Platform
○ It’s like a gym for your AI models. Let’s make them strong and smart!
● Steps to Train a Model
○ Upload your dataset—think of it as feeding your AI.
○ Write a training script—this is like planning your AI’s workout.
○ Configure the training job—select your machine (CPU? GPU? It’s like
choosing your AI’s treadmill).
○ Start the training job and cheer on your AI—go, little buddy, go!

Deploying TensorFlow Models with AI
Because keeping your AI to yourself is like making an amazing cake and not
sharing it. Deploy it so the world can enjoy!
● Steps to Deploy
○ Export your trained model—it’s like baking that cake.
○ Upload it to Google Cloud Storage—put it on display.
○ Deploy on AI Platform—now everyone can have a slice!
○ Use REST or gRPC APIs to interact with the deployed model.

Using TensorFlow Serving for Deployment
● TensorFlow Serving
○ A flexible, high-performance serving system for machine learning
models.
Imagine your AI model is a chef. TensorFlow Serving is the kitchen where
the chef works 24/7.
● Deploying on Compute Engine
○ Set up TensorFlow Serving on a Compute Engine VM - make sure your
kitchen is ready.
○ Serve the model via REST or gRPC APIs for real-time predictions —let the
chef cook and serve dishes to hungry customers!

Using TensorFlow Serving for Deployment

Integrating TensorFlow with Google Cloud Services
● BigQuery ML:
○ Train models directly in BigQuery on large datasets without data transfer.
○ Train models on huge datasets. It’s like taking your AI to a buffet—so much
data to chew on!
● Cloud Storage:
○ Store and manage training data and model artifacts in Google Cloud Storage.
○ Think of this as the fridge where you store all your ingredients (datasets and
models).
● TensorFlow Extended (TFX):
○ TFX is the conveyor belt that automates your AI kitchen—keeping everything
running smoothly.

Best Practices for TensorFlow on Google Cloud
● Optimize Model Performance:
○ Use TPUs for faster training on large models.
○ Implement hyperparameter tuning with AI Platform.
○ Use TPUs—they’re like the AI equivalent of a turbo boost!
● Cost Management:
○ Monitor resource usage and costs with Google Cloud Monitoring.
○ Use preemptible VMs for cost-effective training jobs.
○ Monitor usage to avoid a “Cloud Hangover” (aka bill shock).
● Security:
○ Keep your AI safe—like not letting your pet robot wander off on its own.

Best Practices for TensorFlow on Google Cloud

Real-World Use Cases
Healthcare:
AI doctors—because even doctors need a break (and a second opinion).
Finance:
AI money managers—ensuring you don’t spend all your money on AI puns.
Retail:
AI shopping assistants—because shopping is more fun with a helpful robot.

Getting Started with TensorFlow on Google Cloud

Recommended

More Related Content

Similar to Getting Started with TensorFlow on Google Cloud (20)

Recently uploaded (20)

Getting Started with TensorFlow on Google Cloud

Editor's Notes