Machine Learning
Perceptrons and Neural Networks
Image Processing and Hopfield Networks
Language
Reinforcement Learning
100
(2 answers).In supervised machine learning this is where the model is tested on data where the answer is not known, and this is where the model is predicting the answer without seeing the actual answer (but the human still has access to the answer). (Correct Order Only)
What are model application and model testing.
100
Neural networks are inspired by this biological structure.
What is a brain/ nervous system?
100
Conventional machine learning approaches often have to use handcrafting for this step.
What is feature extraction?
100
This factorization method for large matrices, was used to reduce the dimensionality of a word x document frequency matrix by identifying the key factors of the matrix when performing LSA (latent semantic analysis).
What is Singular Value Decomposition (SVD)?
100
This is a map from states in the environment to actions the agent should.
What is a policy?
200
This feature engineering method transforms a set of selected features into a new set of reduced variables.
What is feature reduction?
200
These are the considerations we should be making when setting up a neural network to solve a problem. (4 major categories expected)
What are: the learning rate, hidden node number, the configuration of the model(inputs/outputs + type of network) and the training + testing data? (at least 3/4)
200
These are some of the psychological paradigms Minerva 2 has been applied to. (Pick 3)
What are... (3 of):
- judgement of frequency and recognition
- category learning
-implicit learning
- associative and reinforcement learning in animals
- heuristic ans decision making
- hypothesis generation
- learning word meanings
- sentence production
200
This is a representation of word meanings as points and differences in meaning as the angle between words.
What is a semantic space?
200
(Two Questions): This is when a reinforcement agent selects the action with the maximum Q value/expected reward value. The other thing is when a reinforcement agent selects an action at random. (CORRECT ORDER ONLY)
What is exploitation? What is exploration?
300
This method has you partition the data into equal sized groups, numbering each group, then selecting the training data by randomly generate a number between 1 and the number of groups and then inserting the partition with that number into the training data.
What is the bagging method?
300
The w3 (bias node weight) that given the weights w1 = 0 and w2 = -1, produces the following outputs with the following inputs (Assume w3 = T):
What is w3 = 1?
300
If I am too big it will cause drastic changes to the outputs of your neural networks leading to divergent behaviours. If I am too small you may never reach the optimal solution.
What is the learning rate?
300
Given the Lexicon and Grammar rule NP = Det + Noun, these are the noun phrases of the sentence "The dog chased a cat".
What is "The dog" and "a cat"?
300
This function specifies what is good in the long term for a reinforcement learning agent.
What is a value function?
400
We transform the data in this way so that the blue and green points are linearly separable.
What if we transform the graph from using cartesian coordinates to using polar ones?
400
This process trains a multi-layered network, where using an input vector and a target output, the system feeds-forward the computation over the input, calculates a prediction error based on the the difference between then output and target then back-propagates that error to correct every layer.
What is back-propagation?
400
Given the hopfield network below, this is what the network produces given [-1, +1, -1, -1].
What is [-1, +1, -1, -1]? (Why is this the output?)
400
This is when a type of feature manipulation, where all of the features are used, but they transformed into a new, smaller set of more useful features that capture higher level relationships between the original features.
What is feature reduction?
400
With this approach the agent figures out the value function directly from interactions with the environment.
What is a model-free approach.
500
This type of learning discovers naturally occurring patterns in the data, can be thought of finding y, where f(x) = y.
What is unsupervised learning?
500
This common logical operation/problem, unsolvable by a single perceptron, was a reason why perceptron research did not advance as rapidly in the 50's and 60's.
What is the XOR problem?/ What is XOR?
500
This type of image-processing network uses a kernel to pass over the image with a filter, this passing over extracts all relevant features captured by the kernel's filter and passed to a pooling layer where it is further reduced. This type of model may combine convolutional and polling layers to the ith degree of depth.
What is a convolutional network?
500
Given the lexicon table below, and the grammar rule VP = NP Verb (with NP = det Noun particle / Noun particle), this is the verb phrase in "hug huul loog dumpsi dib jubba"
What is "(D)loog (N)dumpsi (P)dib (V)jubba"?
500
Your sketchy friend invites you to play a game. You flip 3 coins, if you get either all heads or 1 head and two tails you win 5$ otherwise you lose 6$.
This is the expected utility of playing this game.
What is expecting to lose 50 cents?