https://www.youtube.com/watch?v=JMUxmLyrhSk

AI application

google reccomendation
analyze legal documents J.P Morgan
IBM Watson medical technology
self driving
tweet emotional recognition
...

Types of AI

Artifical Narrow intelligence (only specific problems)
Artifical General Intelligence (perform any intellectual task)
Artificial Super Intelligence (a time when AI will surpass human) doesn't exist yet

Programming languages for AI

Python
R
Java
Lisp
Prolog
C++

Introduction to machine learnign

Machine learning is used in AI

increase in data generation
improve decision making
uncover patterns & trends in data
solve complex problem

What is Machine Learning

Machine learning (ML) is the scientific study of algorithms and statistical models that computer systems use to perform a specific task without using explicit instructions, relying on patterns and inference instead.

Algorithm: a set of rules and statistical techniques used to learn patterns from data

Model: A model is trained by using a ML Algorithm

Predictor Variable:it is a feature(s) of the data that can be used to predict output

Response Variable: it is the feature or the output variable that need to be predicted by using the predictor variable(s)

Training Data: The machine learning model is built using the training data.

Testing Data: The machine learning model is evaluated using the testing data

Machine learning process

Problem model

Define Object
Data Gathering
Preparing Data
Data exploration
Model
Model evaluation
Predictions

Sample

Data Gathering weather conditions, humidity level, temperature, pressure, etc ...
Preparing Data
- Transform data
- Data cleaning
Exploratory Data Analysis Involves undestand the pattern and trends (es temperature low possible rain)
Building a machine model split data in train, test always use training data linear regression, decision tree
Model evaluation and optimization
Predictions the final output is predicted after performing tuning and improving the accurancy

Types of machine learnign

Supervised learning we teach or train the machine using data which is well labelled
Unsupervised learning unlabeled data and allowing the algorithm to act on that information without guidance
Reinforcement learning an agent is put in an environment and he learns to behave in this environment by performing certain actions and observing the rewards

Types of problem solved with ai

regression
classification
clustering

predict the house pricing index

build ML module to predict the price Linear regression algorithm supervised learning regression problem

approve or reject loan

KNN algorithm classification problem supervised learning

popular or not movies

k-means algorithm unsupervised learnign classiffication problem

Supervised learnign algorithm

Linear regression
Logistic regression
Decision Tree
Random Forest
Naive Bayes Classifier
K nearest neighbour
Support vector Machines

Linear regression

Linear Regression is a method to predict dependent variable(Y) based on values of indipendent variables (X). It can be used for the cases where we want to predict some continuos quantity

$Y=\beta o + \beta 1X + \appartiene$

Logistic regression

Logistic regression is a method used to predict a dependent variable, given a set of indipendent variables, such that the dependent variable is categorical

per predire quantità non continue ma per quantità categoriche

classification algorithm

yes/no 0 or 1

$P(x) = e^{\beta o + \beta 1 x} / e^{\beta o + \beta x} +1$

Decision Tree

A decisison tree is a Supervised ML algorithm which look like an inverted tree, wherein each node representes a predictore variable, the link between the nodes represents a decision and aeach leaf node represents an outcome.

root node starting point

internal nodes decision point

leaf/terminal the outcome

branches connection between node

ID3 the most useful algporithm to make a decision tree

choose the best attribute as the root

How to find the best attribute ? that one that splits the data in two differetn classes

information gain and entropy

entropy measures the impurity or uncertainty present in the data

information gain IG indicates how much "information" a partticular feature/values give us about the final node

is necessary calculate the entroy and IG of each term ti choose the right one

random forest

classification algorithm Random forest builds multiple decision trees and glues them together to get a more accurate and stable precision

more accurancy
avoid overfitting
bagging

we use a bootstrap dataset

naive bayes

classification algorithm supervised algorithm follow probabilistic h approach

Naive bayes is based on the Bayes theorem that is used to solve classification problems by following a probabilistch approach

it is based on the idea that the predictor variables ina machine learning are indipendent of each other

k nearest neighbour (knn)

K nearest neighbour is a supervised learning algorithm that classifies a new data point into the target class, depending on the features of it's neghbouring

no way about correlation of the variables

lazy algorithm , learn from training data

k stand for nearest neighbors

measures Euclidean distance and manatthan distance

support vector machines svm

regression and classification algorithm separate data using hyperplanes can also be used to classify non linear data draw a decision boundary to classify data

unsuoervised learnign

k-means clustering

the process by which objects are classified into a predefined number of groups so that they are as much dissimilar as possible from one group to another group, but as much similar as possible.

start
  |
  ▼
Number of   ----> centroid ---> distance object ---> grouping based 
clusters K     ▲                to centroid          on min. distance
               |                                            |
               |                                            |
               | false                                      |
              --------  centrioid has converged ? <---------
                                    | true
                                    |
                                    ▼
                                    end

decide number of clusters
then we provide centroid of all clusters
then the algorithm calculate the euclidian distance from each centroid
nex centroid are recalculated
reassigned points
and then again
these steps are repeated until we have a repetiiton in centroids

the elbow method

reinforcement learnign

an agent is put inside unknow environment and with some function it understand where it is and what it can do.

agent
environment
action
state

markov's decision process

the mathematical approach for mapping solution in reinforcement learning is called markov decision process

parameters :

set of actions a
set of states s
reward r
policy
value

q learnign

AI vs ML vs DL

AI is a technique which enables machines to mimic human action

ML Subset of AI techniques which use statistical methods to enable machines to improve with experience

DL subset of ML which make the computation of multi-layer neural network feasible

limitiation of ML

is not capable enough to handle the high dimensional data

is difficult the feature extraction

introduction deep learning

Deep learning models are capable to focus on the right features by themselves, requiring little guidance from the programmer. these models also partially solvwe the dimensional problem.

how deep learning work ?

Deep learnign is a form of machine learning that uses a model of computing that's very much inspired by the structure of the brain.

Neuron:

Dendrite. Receives signals from other neurons Cell Body: Sums all the inputs Axon: it is used to transmit signal to other cells

What is deep learning

Deep learning is a collection of statistical machine learnign techniques used to learn feature hierarchies based on the concept of artificial neural networks.

single layer perceptron

An artificial neuron or a perceptron is a linear model used for binary classificatiopn. It model a neuron whihc has a set of inputs, each of which is given a specific weight. the neuron computes some fuction on these weighted inputs and gives the output.

is linear or binary classifier

somma pesata

activation function, is a treshold the neuron is activated

multi layer perceptron

a multi layer perceptron has the same strucutre of a single layer perceptron but with one or more hidden layers and is thus considered a deep neural network

feed forward connected network is a full layer connected with other

backpropagation

Backpropagation algorithm is a supervised learning method for multilayer perceptron

classsify lead on the basis of priority

to reduce error

Training a neural network

propagate backwards values and update the weights to reduce error

gradient desce

calculate the error

calculate the rate of change of error wrt change in the weights

based on change in weight update the values

limitiation of feed network

A trained feed forward network can be exposed to any random collection of photgraphs, and the first photograph it is exposed to will not necessarily alter how to it classifies the second one.

recurrent neural network

Recurrent networks are a type of officila neural network designed to recognize patterns in a sequence of data, such as text, genomes, handwriting, the spoken word, or numerical time series data emanating from sensor stock markets and government agencies.

convolutional neural network

a particular neuron is connected only with few neuron in a region

Arkango/appunti.md