1. What is Generative AI?
Ans:
AI that creates new content like text, images, or audio by learning patterns in data. Helps automate creative tasks and simulate scenarios.
2. How do GANs work?
Ans:
Two networks—generator (creates data) and discriminator (checks real vs fake)—compete to improve output realism.
3. How to measure ML performance?
Ans:
- Classification: Accuracy, Precision, Recall, F1-score, ROC-AUC
- Regression: MSE, MAE, R-squared
4. How attention helps Transformers?
Ans:
- Focus on important input
- Capture token relationships
- Improve context & output quality
5. Generative vs Discriminative models?
Ans:
Generative: creates new examples from data. Discriminative: predicts or classifies labels. Use depends on the task.
6. What is a Variational Autoencoder (VAE)?
Ans:
Encodes data into a compact form and reconstructs it; latent space allows generating new, similar outputs.
7. Why is Transfer Learning important?
Ans:
Reuses pretrained models for new tasks, saving time, improving accuracy, and reducing data needs.
8. How does the Attention Mechanism improve Transformer architectures?
Ans:
Models can concentrate on particular segments of an input sequence that are most essential for prediction thanks to the attention mechanism. Self-attention enables each token in transformers to examine how it relates to every other token, leading to a deep understanding of context. As a result, processes like language production, translation and comprehension become more accurate.
9. What ethical concerns arise with the use of Generative AI?
Ans:
Generative AI introduces challenges such as biased outputs, lack of transparency and risks related to privacy and misinformation. Ensuring responsible use requires models to be explainable, safe and protected from misuse, especially in sensitive applications. Maintaining confidence and guaranteeing ethical deployment depend on addressing these issues.
10. How can machine learning model performance be assessed?
Ans:
The evaluation of model performance varies by problem type. Regression tasks rely on measurements like mean squared error, mean absolute error and R-squared, whereas classification tasks require metrics like accuracy, precision, recall, F1-score and ROC-AUC. These indicators show how well a model predicts outcomes and how effectively it generalizes to new data.
Fees Starts From
