Google Machine Learning Flashcards
AI
The ability of a digital computer to perform tasks that intelligent human beings perform
ML
Used by machines to make decisions based on data without getting specific instructions
Training Data
The data that creates the model for ML predictions
Model Training
The process of developing the a model for training data, with the goal of answering questions with the highest degree of accuracy
What are the seven steps of Machine Learning?
1) Gather the Data
2) Prepare the Data
3) Choose a Model
4) Training
5) Evaluation
6) Hyperparameter Training
7) Prediction
Gathering Data
Ensure you haven’t collected too much of any particular kind of data, split data into training (80%) and evaluation (20%), you may need to normalize or deduplicate data
Choosing a Model
Ensure you a picking a model that is suitable to the data that you want to collect
Training
Use X + W * Y + b where W is weight and b is bias. These values are manipulated to determine if predictions are accurate.
Evaluation
Once training is complete, model is evaluating based off test data against training
Hyperparameter Tuning
Fine tune your assumed parameters or hyperparameters to get higher accuracy
Prediction
Use your model for evaluation of data
Supervised Learning
Most common model, used when the training data and validation data is labeled and the task is learning how to set a label to input data.
Classification
A subclass of supervised learning, occurs when output data is a category (ex. apple, pear, orange)
Regression
A subclass of supervised learning, occurs when the output data is a value, such as cost and temperature
Unsupervised Learning
When training data is not labeled and model attemps to learn the structure of the data and export information or features that might be useful for classification. Accuracy can’t be measure but data can be moved into groups.
Clustering
A subclass of unsupervised learning, occurs when you want to group the data
Association
A subclass of unsupervised learning, occurs when you want to link two different actions or behaviors
Semi-Supported Machine Learning
When part of the data is labelled but part of it isn’t resulting in a mix of both methods
TensorFlow
For Data Scientists. An option for those who want to work with ML from scratch, and is a software library that is developed and maintained by Google.
ML Engine
For Data Scientists. An option for those who want to train their own models, but use Google for training and predicitions. Managed TensorFlow service that offloads all infrastructure and software bits from users. Integrates with other GCP services.
Pretrained ML Models
For Developers. An option for those who want to leverage ML without having knowledge of it. Uses Google Developed models to perform predictions.
Auto ML
For Developers. An option for those who want to leverage ML without having any knowledge or it and where the pretrained models are not for a fit purpose. Allows models to be trained by supporting labelled data.
Cloud Tensor Processing Units (TPUs)
Google custom developed application specific integrated circuts (ASICs) used to speed up ML workloads. Enhance linear algebra computation for use in models using matrix computations, without custom TensorFlow operations inside the main traning loop, that take a long time to train, or have very large batch sizes.
Cloud Speech to Text API
Empowers developers with the ability to turn speech into text, API accepts audio and returns text transcription
Cloud Text to Speech API
Gives developers the ability to transform text into a form of Synthesis Markup Language (SSML) input into audio data of natural human speech
Cloud Translation API
Enables translation into thousands of languages, and detect language if it is unknown. Can be used directly in code with REST API.
Cloud Natural Language API
Allows leverage of deep learning models that Google uses for it’s search engine to analyze text and Google Assistant (extract information on entities, categorize entities, perform sentiment analysis, and perform syntax analysis)
Cloud Vision API
Provides vision detection features such as Image Labeling, face and landmark detection, optical character recognition (OCR) and tagging explicit content
Cloud Video Intelligence
Allows you to analyze video that’s been uploaded to Cloud Storage, using labels, shots, and tagging explicit content
DialogFlow
Tool outside of GCP (originates in api.ai) developed to perform human to computer interaction using natural language processing. Can produce auto ansers in skype, google assistant, slack and fb messenger.
Auto ML Vision
Classifies your images according to your own defined labels
Auto ML Translation
Performs translation queries, returns specific results to your domain
Auto ML Natural Language
Classifies English language content into a custom set of categories
Auto ML Tables
Turns structured data into predictive insights
Auto ML Video Intelligence
Classifies segments of videos
