AI Fundamentals Flashcards
AI Trifecta
Algorithms
Data
Computing Power
Three Types of Deep Learning
Supervised
Unsupervised
Reinforcement
Definition of Machine Learning
The process of instructing computers to learn from data
What is Deep Supervised Learning
- A supervised learning algorithm takes a large amount of representative data, called training data
- For each piece of training data the right answer is given
- A neural network-based machine learning algorithm is deployed examine the data and develop a model
What is Deep Learning?
A learning system that uses a cascade of nodes that loosely mimics the neurons in the brain
Each layer of neurons connects to the layer before and after it
Combined, the layers of neurons can encode complex concepts
There are many types of deep learning algorithms
What is Unsupervised Learning?
An unsupervised learning process defers in that there is no “right answer” supplied with the training data
Instead, the algorithm must look within the data itself to find already existing patterns
For example, common uses of unsupervised learning algorithms include clustering data
When might we want to categorize a larger set of data into smaller clusters?
Reinforcement Learning
Reinforcement learning is based on a computer “agent” that assesses an environment, makes a decision, and then receives the consequences of that decision
Agents learn to make better decisions through trial and error in order to maximize their overall rewards
Reinforcement learning is most commonly used in games and robotics
A+1 = sigmoid(A*W + B)
In words, we multiply the activation for each neuron by the weights for each of its connections, then add the bias, then put the whole thing in a sigmoid function
What is “Bias”
A machine learning algorithm that exhibits high bias underfits training data
Note that this is mathematically related to the bias term in the layers, though the relationship might not be intuitive to you
The algorithm often learns quickly
What happens when an algorithm underfits training data?
What is high Variance
A machine learning algorithm that exhibits high variance overfits training data
This results in more mathematical complexity
The algorithm often learns more slowly
What happens when an algorithm overfits training data?
What is “Irreducible Error”
Inevitable error that cannon be reduced
Two Types of Adversarial Learning
Exploratory Attack
Causative Attack
What is an Exploratory Attack
discover weaknesses in the design or function of machine learning systems
Causative attack
create a weakness in the machine learning system that can later be exploited
What is a GAN?
Generative Adversarial Networks (GANs) are a way in which neural network technology can be applied to unsupervised learning problems (which go beyond just clustering)
A generator, tasked with coming up with a fake outputs
An evaluator or discriminator, tasked with differentiating real outputs from fake ones
The two networks “compete” against one another in a zero-sum game, in which the generator attempts to comes
