AI REVISION 1 Flashcards
What is the turing test? and examples
The turing test is a test created by Alan turing which was a test that consisted of an interrogator being connected to a person and a machine via a terminal and cannot see either of them, if the machine could trick the interrogator 30% of the time when answering questions the machine was considered intelligent.
Examples -> Siri, chatbot
What is combinatorial explosion?
Occurs in numeric problems where the complexity of the problem rapidly increases caused by the increasing number of possible combinations of inputs
What is machine learning?
Machine learning is a field of study that gives computers ability to learn without being explicitly programmed.
What is supervised learning vs unsupervised learning
Supervised learning relies on labelled input and output training data, whereas unsupervised learning processes unlabelled or raw data.
2 Examples of supervised and unsupervised learning
Supervised learning.
1. HandWriting Recognition
2. Traffic Prediction
Unsupervised Learning
1. AD Targeting (e.g. group similar costumers together)
2. Crime Hotspots (e.g. identify areas with high crime rate and report it to police management.)
What is clustering?
Clustering is an unsupervised technique about grouping similar objects, the input is unlabelled data as mentioned before, and the goal is to find a natural partitioning or groups of similar data points.
What is Association rules?
Association rules is an unsupervised technique that involves the input being a set of transaction records containing items and the goal is to produce dependency rules to predict occurrence of one variable based on occurrences of another variables.
What is data mining?
Is the analysis of large quantities of data to discover, valid, non-obvious, useful, and ultimately humans should be able to understand pattern in data.
How is the data split?
The data is usually split into two groups, training set and test set (used as estimate of how well the model works)
What is model training?
Learn the model parameters
Decision Tree nodes and goal
Internal nodes: Decision Rules on features
Leaf nodes: Predicted class label
Ultimate Goal -> Prediction
Decision tree pros and cons
Pros - 1.Easy to interpret and explain decisions, 2. Reasonable training time, 3. Feature scaling not necessary
Cons - 1. Small variations in data can lead to very different trees, 2. over complex trees can lead to overfitting
What is linear regression?
Supervised learning technique to estimate a relationship between an input variable X, and output variable X.
aim -> find a line of best fit.
Linear regression pros and cons ->
Pros -> 1. easy to interpret
2. easy to train
3. works well even with limited data
Cons ->
1. only applicable if relationship is linear
2. relies on assumptions about the data
What is model evaluation?
How well does the model do on the validation set.
What is model tuning?
adjust model hyperparameters
What is overfitting
the model fits too well to the training data which negatively impacts performance on unseen data.
Why overfitting happens?
- Training data is limited
- Model is too complex
How to fix overfitting?
- Train with more data
- Simplify the model.
Explain first artificial Neurons
Combine binary inputs with some weights and compare it with a fixed threshold t to decide if neuron fires 1 or 0, that is done by multiplying and adding both inputs, e.g. if weight coeff was w1= 2 and w2 = 2 then x1 and x2 would be -> w1*x1 + w2+x2 = … and that results would create a logic gate.
What is perceptron
Is a learning algorithm that learn the weights to help detect certain input data in business intelligence.
Explain the learning algorithm of perceptron
The weights are adjusted automatically, as more instances are checked if they are misclassified then the weights are adjusted and eventually the algorithm will converge the set of weights that separates the two classes if the function is linearly separable.
What is epoch?
Is the training of the neural network with all the training data for one cycle.
What is learning rate?
is a hyperparameter of the model that controls how much we adjust the weights by each time an error is produced
What are issues if learning rate is too high?
May jump over a good combination of weights.
What are issues if learning rate is too low?
Training will take a very long time
Pros and cons of neural networks
Pros -> 1. Can learn more complicated decision boundaries, 2. Can handle a large number of features.
Cons -> 1. Difficult to interpret
2. Difficult to design neural networks
3. Slow training time.
What is the concept of search space?
is to find the set of all possible solutions to a problem. e.g. input a problem
return solutions to the problem.
Construct a search tree , states, search space, operators, neighbourhood, goal test, path cost.
States: Possible states of the problem
Search space: all possible states reachable from the initial state
Operators - a set of actions that move one state to another.
Neighbourhood: All possible states reachable from a given state
Goal test: test if the search reached a state that solves the problem
Path cost: how much it costs to take a certain path.
Problems with search tree
- Tree size it can be combinatorial explosion
- Computational expensive.
What is a blind search? and their searches, e.g. BFS
Is a search where no information about the search space is given.
- Breadth first search (BFS)
- Depth First Search (DFS)
- Uniform Cost Search (UCS)
What are the three nodes during search in tree search?
- Fringe nodes (open nodes which have not been tested but have been discovered)
- Visited nodes (Closed nodes which have been explored and tested if they match a goal)
- Undiscovered nodes (have not yet been discovered)
What are the three nodes during search in tree search?
- Fringe nodes (open nodes which have not been tested but have been discovered)
- Visited nodes (Closed nodes which have been explored and tested if they match a goal)
- Undiscovered nodes (have not yet been discovered)
What is BFS?
Expand root node first and then expand all nodes in level 1 before expanding nodes on level 2
(ENQUEUE-AT-END)
completeness and optimal
What is DFS?
Expand root node first and explore one branch before exploring another branch.
(ENQUEUE-AT-FRONT)
not completeness and not optimal
What is UCS?
Expand root node first
always remove the smallest cost node first in the queue
(ENQUEUE-BY-COST)
completeness and optimal
What is meant by admissible heuristic?
heuristic is only admissible if it never over-estimates the actual cost, e.g by always being either equal or below the actual cost
What is A* Search?
A* Search is a heuristic search which the aim is to estimate the actual cost to reach the goal state in a search space
(ENQUEUE-BY-ESTIMATED-COST)
Optimal if cost is not over-estimated | Completeness
What is the minimax search?
The minimax search is a game search done in a search tree which finds the most optimal move for a player.
What is the utility function in minimax search? and when do you use it.
A method of measuring how good a game state is.
And you propagate utility function values from leaf node to root node.
How to find the utility values of a move?
For a leaf node is just the utility value of its outcome (of the game).
And for a root node is based on the value of their child nodes.
What is alpha-beta pruning?
Alpha-Beta pruning is a technique that helps the minimax search to be more efficient, by decreasing the number of nodes that are evaluated by the minimax search algorithm.
What does Alpa and Beta mean in the Alpha-Beta Pruning
Alpha: the minimum score that MAX is assured of
Beta: The maximum score that MIN is assured of.
