Intro to Machine learning (ML)

• Machine learning is a subset of Artificial Intelligence (AI).
• Machine learning uses computer algorithms to analyze data and make intelligent, learning-based decisions.
• Machine learning builds models to classify and make predictions using data, rather than following rule-based algorithms.

Example of Machine Learning

• To determine heart failure, data such as BPM (beats per minute), BMI (body mass index), age, and sex is given.
• The result of whether or not the heart has failed along with the other data produces learning.

Traditional vs. ML

• Traditional algorithms use data and rules to develop an algorithm that provides an answer.
• Machine learning algorithms take data and answers to create the algorithm
• Machine learning models establish rules, determining what the machine learning model will be.
• Machine learning models determine if-then-else statements when they receive inputs.
• Machine learning models can be continuously trained and used to predict future values.
• Machine learning defines rules by examining and comparing large datasets to find common patterns.

Supervised Learning

• Algorithms are trained using human-labeled data.
• The precision of a supervised learning algorithm increases with the number of samples provided.

Unsupervised Learning

• The algorithm finds patterns in unlabeled data on its own.
• Unlabeled data is provided as input which allows the machine to infer qualities, draw inferences, and discover patterns.
• Useful for clustering data based on similarity to neighbors and dissimilarity to everything else.
• Various techniques can be applied to explore clustered data and to find patterns.

Reinforcement Learning

• Algorithms are given a set of rules and constraints, learning how to reach goals.
• The state, desired goal, allowed actions, and constraints are defined.
• The algorithm attempts to accomplish the goal by experimenting with different combinations of allowed actions.
• Depending on the decision's quality, the algorithm is either rewarded or punished.
• The algorithm seeks to maximize rewards, according to provided constraints.

Summary of Machine Learning

• Machine learning models are algorithms used to identify patterns in data without explicit programming.

Deep Learning

• Deep learning is a specialized subset of machine learning.
• Deep Learning layers algorithms are used to create a neural network
• Neural networks are artificial replications of the structure and functionality of the brain.
• Deep Learning enables AI systems to learn continuously and improve result’s quality and accuracy
• Deep Learning systems can learn from unstructured data like photos, videos, and audio files.

How Deep Learning Works

• Deep learning algorithms don't directly map input to output, instead relying on several layers of processing units.
• Layers pass their output to the next layer for further processing.
• Each layer transforms data received from the previous layer, extracting higher-level features from the input data.

Creating Deep Learning Algorithms

• Developers configure the number of layers and types of functions that connect layers' outputs to inputs of the next layer.
• Then a model is trained and provided with many annotated examples.
• The connections between nodes adjust during training on massive datasets.
• Adjustments allow the network to learn complex relationships and patterns within the data.

ML vs DL

• Efficiency and performance of machine learning algorithms plateau as datasets grow.
• Deep learning algorithms are able to consistently improve as more data is fed to them.

Key Advantages of Deep Learning

• Deep learning handles complex, unstructured data like images, videos, and audio well.
• Suitable tasks include image recognition, NLP (natural language processing) and speech recognition.
• Deep learning models continuously improve performance with exposure to new data.
• Models can adapt to evolving environments and refine their predictions over time.

Applications of Deep Learning

• Deep learning is efficient at image captioning, voice recognition, transcription, facial recognition, medical imaging, language translation, autonomous cars.

Summary of Deep Learning

• Deep learning is effective but has challenges like high computational needs and potential for bias in training data.
• Deep learning expands upon machine learning by utilizing neural networks for tasks involving complex data.

Artificial Neural Networks

• An artificial neural network is a collection of smaller computing units called neurons.
• The neural networks are based on how the human brain processes information.
• Artificial neural networks mimic aspects of biological brain neural networks to approximate results.

Backpropagation

• The units or neurons in a neural network take in data and learn to make decisions over time.
• Neural networks use backpropagation to learn.
• Backpropagation uses a set of training data that correlates known inputs with desired outputs.

Backpropagation Steps

• Inputs are plugged into the network, and outputs are determined.
• An error function calculates the difference between the actual and desired output.
• Adjustments are made to minimize errors.

Layers in Neural Networks

• A layer is a collection of neurons that take in an input and provide an output.
• Any neural network will have one input layer and one output layer.
• Any neural network will have one or more hidden layers, simulating how the human brain works.

Deep Neural Networks

• Hidden layers receive a set of weighted inputs, production an output with an activation function.
• A neural network with more than one hidden layer is a deep neural network.

Perceptrons

• Perceptrons are the most basic and oldest types of neural networks.
• They are single-layered, connecting input nodes directly to an output node.

Bias

• Input layers multiple the input values by a weight and summing the results
• Hidden layers receive input from other nodes and forward their output to other nodes.
• Hidden and output nodes have a property called bias, a special type of weight applying to nodes after all inputs are considered.

Activation Functions

• An activation function determines how a node responds to its inputs.
• The function runs against the sum of the inputs and bias for an output.
• Activation functions can be various forms and critical to a neural network's success.

Types of Neural Networks

• Convolutional Neural Networks or CNNs
• Recurrent Neural Networks or RNNs

CNN Details

• CNNs are multilayer neural networks inspired by the animal visual cortex.
• Useful for applications like image/video recognition and natural language processing.
• Convolution is a mathematical operation applies a function to another, creating a blend of the two.
• Convolutions can find simple structures in images and combine them for more complex analyses.
• In CNNs, a process occurs over a series of layers, each conducting a convolution on the previous layers.
• CNNs excel at building complex features from less complex ones.

RNN Details

• RNNs perform the same task for every element of a sequence, with prior outputs feeding subsequent stage inputs.
• Neural networks process input through several layer with assumption that inputs are independent of each other.
• Scenarios requiring contextual awareness (like understanding spoken words) need RNNs to account for previous observations.
• RNNs use information in long sequences, with each layer representing a certain time of the observation.

Other Names for Neural Networks

• Artificial Neural Networks (ANNs)
• Simulated Neural Networks (SNNs)

Diving Deeper into Neural Networks

• Think of each node as a linear regression model.
• The model is composed of input data, weights, a bias/threshold, and an output.
• Formula: ∑wixi + bias = w1x1 + w2x2 + w3x3 + bias
• Output: f(x)
• Output is 1 if ∑w1x1 + b>= 0
• Output is 0 if ∑w1x1 + b < 0

Input in Neural Networks

• Weights are assigned to each input layer.
• Weights show the importance of a variable, while larger weights contribute more to the output.
• The inputs are multiplied by their given weights then summed.

Output in Neural Networks

• The output is passed to an activation function.
• The activation function’s output determines is the final output.
• If that output exceeds the threshold, the node activates, or "fires".
• The data passes to the next layer in the network.
• The output of one node becomes an input for the next node.
• The passing of data from one layer to another classifies the network as a feedforward network.

Neural Network Illustration: Case - Should I Surf?

• The decision if you go out to surf or not depends on 3 factors:
• If the waves are good (Yes: 1, No: 0)
• If the lineup is empty and not crowded (Yes: 1, No: 0)
• If there has been a recent shark attack (Yes: 0, No: 1)

Neural Network Illustration: The Inputs

• Are the waves are pumping: X1 = 1
• Crowds are around: X2 = 0
• No recent shark attacks: X3 = 1

Neural Network Illustration: The Weights

• Weights are assigned based on the importance
• Large swells don't comes around offen: W1 = 5
• You are used to crowds: W2 = 2
• Have a fear of sharks: W3 = 4

Neural Network Illustration: The Threshold

• Have a threshold of 3 will translate to a bias value of -3.
• ŷ = (15) + (02) + (1*4) – 3 = 6

Neural Network Illustration: Using The Activation Function

• We determine node will be 1, since 6 is greater than 0.
• Adjusting the weights or the threshold will allow for different outcomes from the model.

Neural Network Illustration: Rationale

• Through decisions can be seen how in neural networks progressively complex decisions can be made.
• Decisions can be made depending on prior decisions or layers.