generated queestions

In the engineering machine learning workflow, which step immediately follows data collection and involves normalization and feature extraction

Preprocessing

What is the fundamental difference between supervised and unsupervised learning

Supervised learning requires input-output pairs (labeled data), whereas unsupervised learning discovers hidden structures in unlabeled data.

Why does Mean Squared Error (MSE) square the individual error terms when evaluating a regression model

To heavily penalize larger errors/outliers and ensure the function is differentiable.

If the learning rate ($\alpha$) in a gradient descent algorithm is set too large, what is the most likely outcome during training

The algorithm may overshoot the minimum and fail to converge.

How does L1 (LASSO) regularization differ from L2 (Ridge) regularization regarding feature selection

L1 can drive weights completely to zero, performing intrinsic feature selection, while L2 shrinks weights close to zero but keeps them non-zero.

When interpreting a logistic regression model output for predicting circuit component failures, what does a prediction value of 0.85 represent

There is an 85% probability that the component belongs to the positive class (Class 1 / Faulty).

In Support Vector Machine (SVM) classification, what characterizes "Support Vectors"

They are the critical boundary points that lie directly on or closest to the margin lines.

What happens to an SVM's decision boundary if you significantly increase the soft-margin penalty parameter $C$

The model forces all feature weights to become zero.

Traditional Recurrent Neural Networks (RNNs) suffer from vanishing and exploding gradients when processing long time-series data. How does a Long Short-Term Memory (LSTM) network mathematically address this

By introducing a dedicated "Cell State" information highway regulated by linear gating mechanisms.

Which unsupervised learning technique is non-linear and strictly optimized for local structure visualization rather than general data compression

t-SNE (t-Distributed Stochastic Neighbor Embedding)