Home › AKTU PYQ › Deep Learning › Predicted Paper

Deep Learning (BAI701) AKTU predicted paper 2026-27

AI-predicted paper for Deep Learning based on 5 past papers (2023, 2025, 2026). NOT an official AKTU paper. Use as a revision tool only.

Paper pattern

  • Section A: Attempt all questions. (14 marks)
  • Section B: Attempt any THREE questions out of five. (21 marks)
  • Section C: Attempt ONE question from each pair (a or b). (35 marks)

Why these questions

  • Unit 2 (Deep Networks) dominates — 25% avg weightage
  • Asked every year: Linear Models (SVM, Perceptron, Logistic Regression), Neural Networks & Shallow Networks
  • Most repeated topic: Training a Network (Loss, Backprop, SGD)

Section A – 2-mark questions

  • What is the difference between linear regression and logistic regression?
  • Discuss convolutional neural network (CNN)?
  • Write a short note on ConvNet Architectures (AlexNet, VGG, ResNet).
  • Provide examples of optimization algorithms used for deep networks.
  • Describe the architecture and training of WaveNet for audio generation.
  • Trace the historical development of deep learning that enabled deep networks to outperform traditional machine learning models.
  • Write a short note on Recurrent Networks & LSTM.

Section B – sample questions

  • Explain the mathematical foundation of the Support Vector Machine (SVM) algorithm and describe its kernel trick for non-linear classification problems. How does it differ from logistic regression?
  • Explain the optimization problem in deep learning. Why are deep learning loss surfaces highly non-convex? Discuss with suitable diagrams.
  • Derive the mathematical expression for the output of a single hidden-layer neural network. Explain the role of activation functions.