Definitions and Concepts
- AI Artificial Intelligence
- ML Machine Learning
- DL Deep Learning
- Data labeling
Human Learning
Decisions are based on calculated weighted averages, based on historic data and guided rules, it is calculated based on mathematical equation result in ranking the selection, for example buying a car, weighting the attributes of a car gives a rank that determines the possible purchase, attributes such as model for example attribute 1 car style (sedan=3, hatchback=2, station wagon=3, truck=0.5..etc.) and weighted based on history of sales (80%, 35%, 40%,10%) respectively, another attribute Color (Red=1, Blue=1, Green=0,…) and weighted based on history of sales (10%, 30%,…) and other attributes such as number of doors, brand, customer rating, market review etc.
f(x)= a0w0 + a1w1 + a3w3 … + anwn
rank = f(x)=3 x 0.80 + 1 x 0.10 + … = 3.06 if greater than 3 then it shows as an option
Comparison between ML and DL
Generative AI (Gen-AI)
Generative AI is sometimes called gen AI, is artificial intelligence (AI) that can create original content—such as text, images, video, audio or software code—in response to a user’s prompt or request
Is a branch of AI Machine Learning that employees Deep Learning (DL) with Neural Network based on neurons and synapses, to predict images or text. by pre-train a model (Foundation Model) called LLM (Large Language Models)
Generative AI uses advanced techniques to optimize performance or natural language literacy with techniques such as Prompt Engineering or Fine-Tuning, or Human Feedback
Gen-AI Use cases
- Content Generation
- Code Generation
- Natural Language processing (Translation, Summarization)
- Creativity and Ideation
- Scientific Discovery exploration and discovery
Types of Foundation Models
Text to Text (T2T)
Called LLM Large Language Models specialized in text summarization, extraction, respond to questions, or create contents.
Techniques
NLP Natural Language Processing
[Tokenization, Stemming, Lemmatization, stop & Removal, Speech Tagging, Speech Recognition, Sentiment Analysis]
RNN Recurrent Neural Networks
Based on memory mechanism, sequential Data/tasks
Slow, complex, no parallelization
Text to Image (T2I)
Examples, Open AI-DALL-E2, Google Imagen, Stability AI , Midjourney based on Diffusion,
The diffusing process works
Machine Learning
Supervised
A human provides input and outputs samples, like teacher show answers, led by example and guiding a blind person on specific routes over and over, repeatedly until a pattern is developed, relies on 3 methods to learn:
Binary Classification
Data is classified according 2 poles, such as true or false, yes or no, red or blue , small and large…etc, labeling data is done by a human
Multiclass Classification
Similar to Binary classification except it goes to more than 3 poles, such as company size small, medium, large , and xlarge, or directing help desk calls to specific destination such as support, sales, marketing, billing and so on, there is a discrete list of options or (data labels) that the data can be classified.
Regression
In regression model, data labeled and output of the grid maps with the input, not related, such as the weather temperature today, the stock exchange value, the winning team in a match and so on.
Unsupervised
Auto creating labels, by automatically classify information into clusters, to identify “norm” patterns and then detect the abnormal pattern, such as traffic patterns during the day with factors such as rush hours, peak periods, weekends, weather conditions and so on.
Reinforcement
This pattern is self learning, by continuously improving by applying patterns randomly and measure the feedback with reward (positive outcome) or penalty (negative outcome), trial and error model for actions performed in the behavior pattern and rank the overall outcome, for example in chess game, measures the patterns of all the possibilities and rank them.
Deep Learning
DL is a subset of ML based on ANN Artificial Neural Network, just like human brain, initially provided with basic human guided rules, then iterate with complex approach through (100s) of layers, each layer summarizes and produces input to the next layer, to recognize patterns that are way more complex than regular ML.
Generative AI
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.
Embedding
The challenge: is to answer or “generate” new answers or artifacts such as audio, video, images, or text based on a trained large language models called “Foundation Models”
