Artificial Intelligence And Machine Learning Quiz

1969

1956

1959

1971

3

4 for even number of states, 3 for odd number

4

N (=number of states)

Limit the breadth (search only through interesting branches)

Limit the depth

Ignore low expected values

Only use branches with a high expectation of a positive outcome

Is faster by applying different techniques

Has 65536 cases

Uses heuristics to select an appropriate search

Looks up all possible cases

Knowledge that is helpful for solving a problem

Can also go wrong

Guarantees a result that satisfies our problem

Mostly a function that estimates the quality of a state

The (human) brain

Ants

Quantum Computers

A very old mathematical formula

A layer of perceptrons between the input layer and the output layer

A static layer of perceptrons that doesn’t change during the learning process

Another way of describing the input layer

The layer of perceptrons which provides biased weights to the network

Based on simple building blocks

Model the human brain

Behave complex

Have normally two layers (input and output)

For every step, there is a chance of making the problem worse by removing/adding states to the intial state

First Order Logic works with quantifiers which are superflous for planning

For every step, one has to declare all remaining states again

First Order Logic is not complete and sound, so it is possible to prove something wrong

Capable of capturing syntactic and semantic relationships between different words

Can have out-of-vocabulary words as well

Effort for humans to tag data was less, because it is a unsupervised technique

Vector size is not direct proportional to vocabular size

FTDC

TCDC

ENIAC

DCK1

Reproduction

Mutation

Cross-Over

Fitness-function

If null move search (sometimes at reduced depth) results in a cutoff, assume that making a move will do the same

Approaches a round where both players don’t make a move

Sometimes a move can make state worse (Zugzwang), which is a problem that has to be considered while null-move pruning

Add a null move to the search

It depends on the number of output links

It depends on the number of input links

The number changes throughout the learning process

Only one, independent of the number of input and output links

Handwritten text recognition

Approxing solutions for high polynomial formulas

Clustering of unlabeled data

Distinguishing between spam and normal mails

Uses MCTS

Uses semi-supervised learning

Uses Deep Learning

Uses reinforcement learning

Passed a restricted Turing test

Was able to give absurd responses

Was the predecessor of ELIZA

Attempted to behave like a paranoid schizophrenic

Can be unary, binary or higher-order constraints

Should be respected as much as possible

State how to solve the problem

Have to be implemented for solving CSP’s

R ≪ 1

The ratio does not affect the performance at all

R ≫ 1

R ≈ 1

Proving whether the first player will win, lose or draw from the initial position, given perfect play on both sides

Provide an algorithm for a guaranteed draw

Provide an algorithm which can produce perfect play from any position

Provide an algorithm which secures a win for one player, or a draw for either, against any possible moves by the opponent from the initial position only

Solving the puzzle by maximizing the heurisitics and minimize the variables

Maximizes the number of constraints and minimizes the number of variables or vice versa, s. t. R=number of constraints/number of variables is not in the near of the ratio 1

Solving the puzzle by maximizing the result and minimizing the amount of time needed to solve the puzzle

Minimizing the score of all moves if player 2 is to move and maximizing the score of all moves if it is player 1 to move.

Cutting of the search earlier if a heuristic evaluation function states that this branch is not promising

Using some random factor, which at some point reaches a certain value to stop further searching of this branch

Using special hardware instead of normal computer hardware (e.g. chess hardware)

Solving the formula tree bottom up, which can approx. halfen the processing time

Given a task T, a number of training data N and a performance measure P

Given a task T, a learning rule R and the number of training data N

Given a performance mesaure P, a learning rule R and the experience of the training data E

Given a task T, a performance mesure P and a experience with the task E

Sigmoid activation function

Threshold activiation function

Evaluation activation function

G(x):=1/(1+exp(-x))

Learned to play only from self-play

Uses null-move pruning

Lost to AlphaGo 100-0

Uses a lot more data than AlphaGo to become stronger

,

AND

:-

.