What is a Foundation Model?
Broad and wide modal that can do a lot (trained on a lot of data)
What is a LLM?
Large Language model designed to generate human like text
Are LLM Deterministic or NonDeterministic?
NonDeterministic - they generate different responses each tim
What is a Diffusion Model used for?
Generative AI for images from text
How does a Diffusion Model work?
It creates noise from an image and uses reverse diffusion to create an image from noise.
What is the name of the high performing FM from AWS?
AWS Titan
What is RAG?
Retrieval Augmented Generation - allows FM to reference a data source outside of its training data
What is Multi-model?
It can take multiple inputs (text, img, etc) and can generate variety of outputs (e.g cat photo that has audio to it)
What is GPT?
Generative Pre-trained Transformer - generates human text or computer code based on input propmpts
What is BERT often used for?
Bidirectional Encoder Representations for Transformers often used for text analysis (sentiment analysis, understanding documents)
What is RNN used for?
Recurrent Neural Network - Meant for sequential data such as time-series or text
Often used for Speech Recognition and
Time Series Prediction
What is ResNet used for?
Residual Network - Deep Convolutional Neural Network (CNN) used for image recognition tasks, object detection and facial recognition
What is SVM used for?
Support Vector Machine - ML algorithm used for classification and regression
What is WaveNet used for
Model to generate raw audio waveform often used for speech synthesis
What is GAN used for?
Generative Adversarial Network - generate sythetic data such as images, videos that resemble traiing data
Often used for Data augmentation.
What is XGBoost used for?
Extreme Gradient Boosting - implement gradient boosting for data augmentation
Type of ML learning that needs labeled data for training
Supervised Learning
Downside of Supervised Larning
OFten difficult to perform on millions of data points because you need labeled data
Type of ML Supervised Learning that predicts numeric values based on input. Output is continuous meaning it can take any value within range
Regression (e.g predict house prices, weather forcast)
Type of ML Supervised Learning that predicts categorial label of input data
Classification (e.g spam not spam, animals in zoo)
Type of ML Learning with Unlabled Input Data (and humans label output data)
Unsupervised Learning
Unsupervised Learning Technique that groups similar data points together based on their features
Clustering Technique
Example of Clustering Technique (K means clustering)
Customer Segmentation - customer wants to segment customers to understand different purchasing behavior (looking at customer purchase history)
Technique that associates one item with another
Association Rule Learning Technique
