Terms Flashcards
(39 cards)
Machine Learning (ML)
Enabling computers to learn on their own using data.
Artificial Intelligence (AI)
An overarching term for enabling computers to mimic human intelligence.
Deep Learning (DL)
Using neural networks and deep layers to learn without intervention.
Generative AI (Gen-AI)
Generating new content to expand on the input data (Training data).
Supervised Learning
The Model is trained on labeled data.
Use cases:
Image classification
spam detection
Unsupervised Learning
The Model is trained on unlabeled data, tries to find patterns on its own.
Use cases:
Customer segmentation
Anomaly detection
Self-Supervised Learning
The model is trained on unlabeled data, it creates its own labels to predict and infer missing information.
Use cases:
Natural language processing (NLP) models
Semi-Supervised Learning
The model is trained on a mix of data. From a small amount of partially labeled and a large amount of unlabeled data.
Use Cases:
Speech recognition with some transcribed audio
Reinforcement learning
The model learns through interacting with the environment. Receiving feedback in the form of rewards or penalties.
Use Cases:
Recommendation systems
Self-driving cars
Model Fit Patterns:
Underfitting & Overfitting
Underfitting - A model doesn’t learn enough from the training data, so it performs poorly on the training set.
Overfitting - A model learns too much detail from the training data, but performs poorly on new, unseen data.
Balanced - The model learns the right amount from the training data. It can generalize and perform well on new data.
Foundation Models
Foundation models are large models that can be adapted to perform a variety of tasks across various types of data.
Fine-tuning a model
You can adapt foundation models through further training using smaller task-specific datasets via fine-tuning.
Large language models (LLMs)
LLMs are a subset of foundation models that can understand and generate human language.
Machine Learning Process Stages
Generating data
Training the model
Deploying the model
Exploratory Data Analysis (EDA)
The process of examining and understanding a dataset before diving into modeling.
Correlation Matrix
A correlation matrix allows you to quantify relationships between variables.
Feature Engineering
The process of transforming the raw data into meaningful features.
Hyperparameters
Definition: Settings that you choose before training a model.
Purpose: Control the learning process and model behavior.
Examples: Learning rate, batch size, number of epochs.
Parameters
Definition: Values that the model learns during training.
Purpose: Directly impact model predictions.
Examples: Weights and biases in a neural network.
Amazon Rekognition
Analyses images and videos to identify text, objects and people.
Can be enhanced by using Content Moderation and Content Labels.
Use cases:
Content moderation to identify harmful or offensive images
Identify verification
Identify objects and text in images
Textract
Uses machine learning as to extract information from any kind of document
Can handle printed or handwritten using Optical Character Recognition (OCR)
Use cases:
Automated ID Processing
Analyzing Invoices
Translate
Uses machine learning as a language translation service
Quickly translates large volumes of HTML or text content
Supports 70+ languages
Can customize it to recognize your own brand names, products, and terminology
Amazon Polly
Generates realistic, natural sounding speech, from text that you provide.
You can provide the text in a variety of languages
The resulting audio can be streamed, saved, or downloaded
Can be used to add natural sounding speech to your applications
Supports a variety of languages and voices
Transcribe
Uses machine learning as a speech to text service
Can handle streamed audio or audio files
use case:
subtitles or meeting notes
