Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
30 Cards in this Set
- Front
- Back
|
Entropy equation |
|
|
|
3 types of function for neural nets |
|
|
|
Main idea of deep learning |
Transform the input space into higher level abstractions with lower dimensions (unsupervised learning) |
|
|
Offline Vs online learning |
Offline: all inputs available from the beginning Online: inputs come in as a stream |
|
|
K-fold cross validation |
Splitting the data into k partitions, each one taking it in turns to be the testing partition |
|
|
Bayes' Theorem |
|
|
|
Simulated annealing |
Escape local maxima by allowing some "bad" moves, but gradually decrease their likelihood |
|
|
Local beam search |
Start with k randomly generated states. Each iteration, generate all the successors for all states. If one is the goal state stop, else pick the k highest states |
|
|
Safely explorable |
Can reach the goal state from any valid state |
|
|
Four things to define a problem |
Initial state Successor function Goal test Path cost |
|
|
Completeness |
Does it always find a solution if one exists? |
|
|
Optimality |
Does it always find a least-cost solution? |
|
|
Bidirectional search |
Do two searches - one from the initial state, one from the goal state |
|
|
Greedy best-first search |
Expands the node that appears to best closest to the goal. |
|
|
Admissible heuristic |
Never overestimates |
|
|
Consistent heuristic |
Once a node is expanded, the cost by which it was reached is the lowest possible. |
|
|
Weak AI |
Focussed on one narrow task |
|
|
Artificial general intelligence |
Ability to apply intelligence to any problem rather than just one specific problem |
|
|
Lazy learning |
System does not maintain a generic model. When there's an evaluation request on a new data point, the system will only provide a (local) model for that particular data, eg distance metrics in k-NN.
Nearest neighbour is an example. |
|
|
Markov decision process |
Process in which the state transitions are stochastic |
|
|
Markov property |
The probability of moving into state j as the next state depends only on the current state and the current action; the past does not influence the near future |
|
|
Partially observable MDP |
Decision process where the system dynamics are determined by an MDP, but the agent can't directly observe the underlying state. Instead it maintains a probability distribution over the set of possible states |
|
|
Independence of variables |
Their joint probability = product of their probabilities |
|
|
Backtracking search |
Like DFS but we just apply one action to produce one successor node, rather than working out all successors at the same time |
|
|
Map colouring - what order to consider attributes? |
Most constrained variable Most constraining variable Least constraining value (leave other variables open) |
|
|
Partial-order planning |
Maintains a partial ordering between actions and only commits ordering between actions when it needs to. |
|
|
Temporal difference learning |
Maintaining a value that combines current reward, future reward and progressive learning (learning rate) |
|
|
Q-learning |
Also including the expected value of taking each action |
|
|
Lazy Vs eager |
Lazy: generalisation beyond the training data is delayed until a query is made Eager: the system tries to construct a general, input-independent target function during training of the system |
|
|
Reinforcement learning |
Repeatedly interact with the environment, receive feedback. Learning problem: to get a good policy based on the feedback |
