Machine learning on Google Cloud

Key Takeaways

hi my name is Ryan and i' like to tell you about machine learning on Google cloud in the next few minutes we'll talk about several of the different machine learning options on Google cloud from pre-trained apis to manag platforms ready let's get [Music] started Google cloud has many different machine learning products for developers to use these span a wide spectrum depending on your experience level and project needs on one side we have machine learning apis which you can use even if you have little machine learning experience yourself on the other side we have vertex Ai and AI infrastructure products which allow you to build and deploy your own custom machine learning models let's take a closer look in the first episode of Google Cloud Essentials we talked about how cloud computing is all about getting things done using someone else's computers in the machine learning world Google Cloud shares its machine learning capabilities with developers through the machine learning API I with these apis you can gain insights from data using Google Cloud's pre-trained machine learning models and what's awesome about these apis is that they require zero prior knowledge of machine learning basically Google Cloud handles the training aspect of machine learning Gathering data and building a predictive model allowing you to jump straight to the prediction aspect where you give the API data and get back information about that data the vision API lets you gain insights about your image using Google Cloud's pre-trained Vision models one feature is face detection detecting different faces in an image along with the likelihood that each face has emotions like Joy sorrow and anger another feature is object detection detecting different objects in the image with a confidence score for each one the vision API also makes it easy to detect text logos and landmarks another machine learning API to check out is the natural language API which helps you analyze text you can detect entity keywords in the text perform sentiment analysis analyze syntax and categorize the text based on its topic there also several other machine learning apis including the video intelligence API translation API speech to text API text to speech API and the cloud inference API as a developer you can call all of these machine learning apis from your code using client libraries in programming languages like python node.js Java go C SHP PHP and Ruby the machine learning apis are easy to use since Google handles data collection model training and maintenance but what if you want to train a machine learning model using your own custom data set this is where automl comes into play with autom ML you provide training data and Google Cloud builds a machine learning model for you that comes with a prediction endpoint there are several different automl products including automl Vision automl natural language automl translation automl video intelligence and automl tables these are all accessible via the vertex AI section of the Google Cloud console let's take a look at an example automl use case perhaps you're a meteorologist looking to use machine learning to identify different types of clouds like Cirrus cumulo Nimbus or Stratus with the pre-trained vision API you might be able to identify that there are clouds in an image but not what type of clouds they are this more specific task requires a custom data set with automl vision you can upload your own custom image data set of clouds each one labeled as Cirrus cumulon Nimbus or Stratus then Google Cloud will use this custom Training data set to train a machine learning model finally you'll be provided with a prediction endpoint that you can use to classify new images of clouds you'll also be able to see evaluation metrics for the model like Precision recall and the confusion Matrix the machine learning apis and automl are great if you want to use or build upon a pre-trained model that Google has already created but what if your needs are beyond that this brings us to vertex AI a managed machine learning platform that lets you build deploy and scale ml models faster in fact it requires about 80% fewer lines of code to train a custom model on vertex AI versus other platforms vertex AI integrates with why they use open source machine learning framework Frameworks such as tensorflow pytorch andit learn and can also support all ml Frameworks via custom contats for training and prediction vertex AI is a single platform with all the tools you need to manage all the steps in a custom machine learning workflow this includes creating a data set and uploading data training an ml model on your data uploading and storing your model in vertex AI deploying your train model to an endpoint for online predictions running batch pred jobs and managing your models and endpoints vertex AI also features ml Ops tools to easily manage your data and models with confidence and repeat at scale these tools include vertex AI model monitoring for monitoring the quality of your deployed models vertex AI pipelines for orchestrating repeatable training and serving pipelines and vertex AI feature store for organizing storing and serving features to interact with vertex AI you can use notebooks prepackaged with jupyter lab and deep learning packages the Google Cloud console to manage your ml resources and get access to monitoring and logging in the cloud client libraries and rest apis to call vertex AI from your code while vertex AI is a great managed solution for building and deploying machine learning models to the cloud perhaps you want to handle even more of the process yourself this brings us to AI infrastructure tools these give you the raw machines and tools that you can use to build and host machine learning models deep learning VM images are Google compute engine instances that come pre-installed with the latest versions of machine learning Frameworks like tensorflow pytorch and pyit learn there are also Cloud gpus which are great for speeding up compute jobs like machine learning scientific Computing and 3D visualization and finally there are Cloud tpus which help you train and run machine learning models faster than before these topics can fill a whole video on their own so we willon go into dep about them here so there you have it an overview of several different Google Cloud products you can use for machine learning ranging from pre-trained apis to a managed platform for developing and deploying your own machine learning models make sure to check out other episodes of Google Cloud Essentials where we cover topics like storing data on Google Cloud the Google Cloud console the Google Cloud key use cases and more I can't wait to see what you build with Google Cloud