1: Intro to ML

Question 1

What is meant by Machine Learning?

Answer 1

A mathematical model to define the relations between the inputs and the outputs, and utilize it to predict the outputs for new scenarios or generate insights about new data scenarios.

Question 2

What are parameters?

Answer 2

Constants that define specific characteristics of the system (e.g., growth rate, decay constant).

Question 3

What does ML use?

Answer 3

Machine learning uses a computing device to process data and integrates many distinct mathematical tools such as probability and statistics, optimization, and control theory.

Question 4

What are the possible outcomes in any dataset (generally speaking)?

Answer 4

Either Continous or Discrete.

Question 5

What kind of model is predicted for Continous outputs?

Answer 5

Regression.

Question 6

What kind of model is predicted for Discrete (finite, categorical) outputs?

Answer 6

Classification.

Question 7

What ML techniques are used to discover the hidden patterns within the data (i.e., there are no target outputs provided)?

Answer 7

Clustering Analysis.

Question 8

What are the four major machine learning paradigms?

Answer 8

Supervised L, Unsupervised L, Semi-supervised L, Reinforcement L.

Question 9

What is Unsupervised L about?

Answer 9

Concerned with discovering the hidden patterns in the data inputs and includes clustering as an important sub-domain.

Question 10

What is Supervised L about?

Answer 10

Denotes the learning tasks when data inputs and corresponding target outputs are provided, and includes classification and regression approaches.

Question 11

What is Semisupervised L about?

Answer 11

Covers problems where only partial label information exists. A basic classification model is designed on the few labeled data instances, which is called the semi-supervised classification step. A semi-supervised clustering step is then performed, where the model is tuned up to operate without supervision on the remaining large unlabeled data instances, and assigns them to the classes from the first step.

Question 12

What is Reinforcement L about?

Answer 12

Denotes the learning setup where the goal is to find an action policy that achieves a given goal. Follows the“cause and effect” method. A reward function that acts as a feedback to the agent.

Question 13

What are the two datasets used for building the model?

Answer 13

The training set - Develop the classification model. The testing set Evaluates the accuracy of the developed model.

Question 14

What questions do precision and recall answer?

Answer 14

Precision: Of all the predicted positive cases, how many are actually positive?
Recall: Of all the actual positive cases, how many did the model identify correctly?

Question 15

How to validate the best classificaion model?

Answer 15

Several techniques may be tested in parallel, and the technique that returns the highest evaluation performance is selected.

Question 16

How to validate a regression model?

Answer 16

The evaluation metric that is routinely calculated to judge the model’s performance is the mean squared error (MSE).

Question 17

Please define classification and give an example.

Answer 17

In classification, one assigns objects (instances) to one of a set of predefined classes designated by class labels. This is done based on information extracted from the training data that has already been classified. Classification is a form of supervised machine learning. Example: Determine if a given e-mail is a spam.

Question 18

Please list the evaluation metrics in a classification algorithm.

Answer 18

The confusion matrix is used to evaluate the model. Some of the metrics that are based on the confusion matrix are: accuracy, precision, recall, and f-score.

Question 19

Describe the dfference between classification and clustering.

Answer 19

Classification: Classifying data according to pre-defined categories

Clustering: Partitioning data into groups

Question 20

What are the ingridients in Reinforcement Learning?

Answer 20

Agent - performs action, Action - the possible moves, Environment - scenario the agent faces, State - current situation returned by E, Reward - the immediate return, Policy the strategy that agent employs, Value - the long-term return

Question 21

What are the applications of Reinforcement Learning?

Answer 21

Well-known applications of reinforcement learning are game playing and robotic movement control.

Question 22

How does RL work?

Answer 22

In particular, if the agent starts at state S1 and will take action A1, they will reach state S2 and gain a reward R2 . By moving further, taking an action A2, the output will be S3 and R3.

Question 23

Please list the reinforcement learning approaches.

Answer 23

There are three major approaches to perform a reinforcement learning, which are model-based (and model-free), value-based, and policy-based approaches.

Model-based: Learns a model of the environment’s dynamics (how actions affect states) to plan the best actions.
Value-based: Focuses on learning the value (expected return) of actions in states, e.g., using Q-learning.
Policy-based: Directly learns a policy (mapping from states to actions) to maximize rewards, e.g., using policy gradient methods.