🎉 Use coupon LEARN40 and get 40% OFF on all courses! Limited time — don’t miss out! - Use code:

LEANR40

Top AI Interview Questions: Part-3

5 min read 5 views 0 comments
Top AI Interview Questions: Part-3
Top AI Interview Questions: Part-3

🤖 Top AI Interview Questions: Part-3

Continue your AI interview prep with advanced ML & Data Science concepts.

  • Generative Models: Learn joint probability P(x, y) and can generate new data. Examples: Naive Bayes, GANs, HMM.
  • Discriminative Models: Learn conditional probability P(y|x) and focus on classification. Examples: Logistic Regression, SVM, Neural Networks.

PCA is a dimensionality reduction technique. Transforms features into principal components keeping only the most important ones that explain the variance. Benefits:
  • Reduces overfitting
  • Improves visualization
  • Speeds up training

Feature selection chooses the most relevant features for a model. Benefits:
  • Reduces overfitting
  • Improves model accuracy
  • Speeds up training
Methods:
  • Filter (correlation)
  • Wrapper (RFE)
  • Embedded (Lasso)

Converts categorical data into numerical format. Each category becomes a binary column (0 or 1). Example: Color = Red, Green, Blue → Red=[1,0,0], Green=[0,1,0].

Reducing the number of input variables while retaining important info. Techniques:
  • PCA (unsupervised)
  • LDA (supervised)
Benefits: Simplifies models, avoids curse of dimensionality.

Regularization prevents overfitting by penalizing large weights.
  • L1 (Lasso): Adds absolute values, can shrink some weights to zero (feature selection)
  • L2 (Ridge): Adds squared values, reduces weight magnitudes but keeps all features

As dimensions increase, data becomes sparse and harder to model. Distance-based algorithms like KNN become less effective. Solution: Use dimensionality reduction or feature selection.

Unsupervised algorithm that groups data into k clusters. Steps:
  1. Choose k centroids
  2. Assign each point to the nearest centroid
  3. Recalculate centroids
  4. Repeat until convergence
Used in market segmentation, image compression.

  • KNN (K-Nearest Neighbors): Supervised, classification/regression, uses labeled data
  • K-Means: Unsupervised, clustering, does not use labels

A probabilistic classifier based on Bayes' Theorem. Assumes features are independent (naive assumption). Works well for text classification like spam detection and is very fast.

Comments (0)

No comments yet

Be the first to share your thoughts!

Leave a Comment

Your email address will not be published.