Home › AKTU PYQ › Deep Learning › Exam Strategy

Deep Learning (BAI701) AKTU exam strategy

Focus on Unit 2 (Deep Networks) and Unit 4 (Optimization & Generalization). Together they account for about 72 marks among all questions printed per paper. Unit 5 (Case Studies & Applications) is safest to skim if you're short on time.

Unit priority

  • Unit 2 (Deep Networks) – Must Prepare, 38 marks avg. Unit 2 (Deep Networks) averages 38 marks per paper — one of the heaviest units in this subject. Convolutional Networks and Generative Adversarial Networks (GAN) have appeared in every paper analysed — these are must-dos. Must-do: Convolutional Networks, History & Theory of Deep Learning, Regularization & Batch Normalization, Generative Adversarial Networks (GAN).
  • Unit 4 (Optimization & Generalization) – High Priority, 34 marks avg. Unit 4 (Optimization & Generalization) averages 34 marks per paper — one of the heaviest units in this subject. Optimization in Deep Learning and Recurrent Networks & LSTM have appeared in every paper analysed — these are must-dos. Must-do: Optimization in Deep Learning, Recurrent Networks & LSTM, Generalization in Neural Networks, Deep Reinforcement Learning.
  • Unit 1 (Introduction to Neural Networks) – High Priority, 30 marks avg. Unit 1 (Introduction to Neural Networks) averages 30 marks per paper — one of the heaviest units in this subject. Linear Models (SVM, Perceptron, Logistic Regression) and Neural Networks & Shallow Networks have appeared in every paper analysed — these are must-dos. Must-do: Training a Network (Loss, Backprop, SGD), Linear Models (SVM, Perceptron, Logistic Regression), Neural Networks & Shallow Networks, Introduction to Machine Learning.
  • Unit 3 (Dimensionality Reduction) – Medium Priority, 26 marks avg. Unit 3 (Dimensionality Reduction) averages 26 marks per paper — one of the heaviest units in this subject. Linear Dimensionality Reduction (PCA, LDA) and ConvNet Architectures (AlexNet, VGG, ResNet) have appeared in every paper analysed — these are must-dos. Must-do: Linear Dimensionality Reduction (PCA, LDA), ConvNet Architectures (AlexNet, VGG, ResNet), Training a ConvNet (Weight Init, Hyperparameters), Autoencoders.
  • Unit 5 (Case Studies & Applications) – Safe to Skim, 25 marks avg. Unit 5 (Case Studies & Applications) averages 25 marks per paper — one of the heaviest units in this subject. Must-do: Audio WaveNet, Image Classification & Detection, NLP & Word2Vec, Face Recognition.

Topics AKTU keeps repeating

  • Linear Models (SVM, Perceptron, Logistic Regression): Asked in 3 out of 5 papers. If you prepare only one topic from its unit, make it this one.
  • Neural Networks & Shallow Networks: Asked in 3 out of 5 papers. If you prepare only one topic from its unit, make it this one.

Short on time?

  • Among all units, Unit 5 (Case Studies & Applications) is the lowest priority at 25M avg. If you're running low on time, prioritize the top 3 units and give this one a quick read. But even here — Image Classification & Detection, NLP & Word2Vec, Audio WaveNet are high-repeat topics. Cover these at minimum to stay safe.

How the paper is marked

  • Section A: 7 questions × 2 marks = 14 marks. Short definition/concept questions from all 5 units. Prepare 2–3 key definitions per unit — this section is the easiest 14 marks you can secure.
  • Section B: Attempt any 3 of 5 questions × 7 marks = 21 marks. Choose the 3 questions from units you've prepared most. Don't attempt all 5 — pick your strongest 3 and write them well.
  • Section C: 5 OR-pairs × 7 marks = 35 marks. One pair per unit — you must attempt all 5. This is where the exam is won or lost. Focus your deep preparation on the top 3 priority units.

Exam-day tips

  • Start with Section A to build momentum — budget 15–20 minutes max for all 7 questions.
  • In Section B, Unit 2 (Deep Networks) questions have the highest repeat rate — always attempt that one first.
  • For Section C OR-pairs, read both options before choosing — sometimes the 'b' option is simpler despite appearing longer.
  • If you've prepared the VERY LIKELY topics, you can comfortably score 50+ marks without touching any new topic on exam day.
  • This strategy is built on 5 past papers — the patterns here are highly reliable.