Machine Learning

What is Machine Learning?

Machine Learning is about building systems that can learn from data, improving performance at a task based on some measure.

What is a labeled training set?

A labeled training set contains the desired solution (a.k.a. a label) for each instance.

What are the two most common supervised tasks in Machine Learning?

Regression and classification.

Can you name four common unsupervised tasks?

Clustering, visualization, dimensionality reduction, and association rule learning.

What type of algorithm would you use to allow a robot to walk in unknown terrains?

Reinforcement Learning is likely to perform best for learning to walk in various unknown terrains.

What is an online learning system?

An online learning system can learn incrementally, adapting rapidly to changing data and training on large quantities of data.

What is out-of-core learning?

Out-of-core algorithms can handle vast quantities of data that cannot fit in memory, using mini-batches and online learning techniques.

What do model-based learning algorithms search for?

They search for optimal values for model parameters to generalize well to new instances.

What is a test set used for?

A test set is used to estimate the generalization error that a model will make on new instances.

What is the difference between a model parameter and a hyperparameter?

A model parameter determines predictions for new instances, while a hyperparameter governs the learning algorithm itself.

Can you name four challenges in Machine Learning?

Lack of data, poor data quality, nonrepresentative data, and uninformative features.

What should you do if your model performs great on training data but poorly on new instances?

Possible solutions include getting more data, simplifying the model, or reducing noise in the training data.

What is the purpose of a validation set?

A validation set is used to compare models and select the best model while tuning hyperparameters.

What happens when tuning hyperparameters using the test set?

You risk overfitting the test set, leading to an optimistic generalization error measurement.

Which Linear Regression training algorithm can you use for millions of features?

Stochastic Gradient Descent or Mini-batch Gradient Descent.

What occurs when the features in a training set have very different scales?

Gradient Descent may take a long time to converge; scaling the data can improve this.

Can Gradient Descent get stuck in a local minimum with a Logistic Regression model?

No, Gradient Descent cannot get stuck in a local minimum with a Logistic Regression model because the cost function is convex.

If validation error goes up during training, what could this indicate?

It may indicate that the learning rate is too high or that the model is overfitting.

What are the advantages of a CNN over a fully connected DNN for image classification?

Fewer parameters, ability to generalize across image features, and leveraging spatial information.

What is the main technical difficulty of semantic segmentation?

The loss of spatial information in CNNs needs to be restored to predict pixel classes accurately.

What is reinforcement learning?

An area of Machine Learning focused on creating agents that maximize rewards through interactions with an environment.