AI & machine learning quiz

Deep Learning Quiz: Neural Networks, CNNs, RNNs and Transformers MCQs

Test your deep learning understanding: neurons and activation functions, loss functions, backpropagation and optimisers, regularisation, CNN shape calculations, RNNs and LSTMs, attention and transformers and transfer learning.

✅ Free · no login60 questions⏱ Timer📝 Explanations🖨 Mark sheet & PDFIntermediate to advanced
Deep Learning Quiz: free online quiz with timer and mark sheet by Codanics
Loading…

Deep Learning Quiz

60 questions · 7 topics · timer · instant explanations · printable mark sheet.

  • Questions60 in the bank, 10–30 per attempt
  • Levels20 easy · 24 medium · 16 hard
  • TimeAbout 5–20 minutes
  • PriceFree, no login needed
  • Made byDr. Aammar Tufail, Codanics
  • Updated18 November 2026

In short: the Deep Learning Quiz is a free online multiple-choice quiz by Codanics with 60 questions on neural network basics; training and optimisation; regularisation techniques; convolutional neural networks; RNNs and LSTMs; transformers and attention; and transfer learning and frameworks. You choose the number of questions and the difficulty, answer against a timer, see an explanation after every answer and finish with a printable mark sheet.

🚀 Follow Codanics for more free quizzes and lessonsNew quizzes, free lessons, scholarships and PhD positions, straight to your phone.

What the Deep Learning Quiz covers

Deep learning powers image recognition, speech, translation and today's large language models. This quiz checks whether you understand what happens inside a network while it trains, from activations and gradients to attention.

Hard questions include output-shape and parameter-count calculations for convolutional layers, a favourite in interviews and exams.

  • Neural network basics 9 questions
  • Training and optimisation 11 questions
  • Regularisation techniques 8 questions
  • Convolutional neural networks 10 questions
  • RNNs and LSTMs 7 questions
  • Transformers and attention 8 questions
  • Transfer learning and frameworks 7 questions

Who it is for: AI and data science students, engineers moving into deep learning and anyone preparing for a deep learning or computer vision interview.

How the quiz works

  1. Type your name if you want it printed on the mark sheet (optional; it never leaves your browser).
  2. Choose 10, 20 or 30 questions, a difficulty (Easy, Mixed or Hard) and the timer: Standard, Relaxed or none.
  3. Answer each question before the timer runs out. Press A–D or 1–4 on a keyboard, or tap an option on your phone.
  4. Read the explanation after every answer, then press Next.
  5. At the end, get your score, grade, topic-wise breakdown and a full answer review. Print it or save it as a PDF.
  6. Retake the quiz for a fresh random set, or practise only the topics you found hard.

Free video lessons from Codanics

Score lower than you hoped? These free lessons from the Codanics YouTube channel cover the same topics, explained step by step.

▶ Full playlist: Deep learning-101
Free on the Codanics YouTube channel, taught in Urdu/Hindi.

Sample questions with answers

Here are 9 questions from the bank, with answers and explanations. The quiz above draws from all 60.

1. What does a single artificial neuron compute before applying its activation function?
  1. The maximum of its inputs
  2. A weighted sum of inputs plus a bias
  3. The average of its weights
  4. A random number between 0 and 1

Answer: B. A weighted sum of inputs plus a bias
A neuron computes z = w1*x1 + w2*x2 + … + b, then passes z through an activation function such as ReLU or sigmoid.

2. Applying softmax to the logits [1, 2, 3] gives approximately which result?
  1. [0.167, 0.333, 0.5]
  2. [0.333, 0.333, 0.333]
  3. [0.0, 0.0, 1.0]
  4. [0.09, 0.245, 0.665]

Answer: D. [0.09, 0.245, 0.665]
Softmax uses exponentials: e^1, e^2, e^3 divided by their sum. That is not the same as dividing by the plain sum (which gives 1/6, 2/6, 3/6).

3. What does one epoch mean in neural network training?
  1. One update of the weights
  2. One full pass through the entire training set
  3. One mini-batch of 32 samples
  4. One evaluation on the test set

Answer: B. One full pass through the entire training set
An epoch is complete when every training example has been used once. With mini-batches there are many weight updates per epoch.

4. What does the Adam optimiser combine?
  1. Dropout with batch normalisation layers
  2. L1 and L2 penalties on the weights
  3. Momentum and per-parameter adaptive learning rates
  4. Random search with early stopping rules

Answer: C. Momentum and per-parameter adaptive learning rates
Adam keeps a moving average of gradients (momentum) and of squared gradients, which scales the step size for each parameter separately.

5. What does dropout do during training?
  1. Removes the worst training samples
  2. Deletes layers that are not learning
  3. Randomly sets a fraction of neuron outputs to zero
  4. Lowers the learning rate every epoch

Answer: C. Randomly sets a fraction of neuron outputs to zero
Dropout randomly switches off units on each training step, so the network cannot rely on any single neuron. It is turned off at inference time.

6. What does a 2×2 max pooling layer with stride 2 do to a 28×28 feature map?
  1. Increases it to 56×56
  2. Reduces it to 27×27
  3. Leaves it at 28×28
  4. Reduces it to 14×14

Answer: D. Reduces it to 14×14
Each 2×2 block is replaced by its maximum value, halving both height and width.

7. A 32×32 image passes through a 5×5 convolution with stride 1 and no padding. What is the output height and width?
  1. 32×32
  2. 27×27
  3. 30×30
  4. 28×28

Answer: D. 28×28
Output size = (n + 2p – k) / s + 1 = (32 + 0 – 5) / 1 + 1 = 28.

8. What does an attention mechanism let a model do?
  1. Skip the training step entirely
  2. Reduce images to black and white
  3. Increase the batch size automatically
  4. Focus on the most relevant parts of the input

Answer: D. Focus on the most relevant parts of the input
Attention computes weights that show how much each input element should contribute when producing a given output.

9. What is transfer learning?
  1. Moving a model from CPU to GPU
  2. Copying training data between folders
  3. Training two models at the same time
  4. Reusing a model trained on one task for a related task

Answer: D. Reusing a model trained on one task for a related task
A model pre-trained on a large dataset already knows useful features, so you can adapt it to a new task with far less data and compute.

Free guides and blogs to study

Free tools for your studies and research

Codanics also has 160+ free online tools for students and researchers: statistics calculators, meta-analysis, PDF and writing tools. They run in your browser, with no login.

Browse all free tools →

Frequently asked questions

Is the Deep Learning Quiz free?

Yes. The Deep Learning Quiz is completely free, with no sign-up, no login and no limit on attempts.

How many questions are in the Deep Learning Quiz?

The question bank has 60 multiple-choice questions (20 easy, 24 medium and 16 hard) across 7 topics. Each attempt draws 10, 20 or 30 of them at random, so every attempt is different.

How does the timer work?

Each question has its own countdown: about 25 seconds for easy, 40 for medium and 55 for hard questions, plus 15 seconds when there is code to read. Choose Relaxed to double the time or switch the timer off to study at your own pace. Unanswered questions count as skipped.

Can I get a certificate or mark sheet?

Yes. When you finish, you get a mark sheet with your name, score, grade, time, topic-wise performance and every answer explained. Use Print or save as PDF to keep it. It is a self-assessment result, not an official certificate.

Are the answers and explanations correct?

Every question was written and reviewed for one clearly correct answer, and code questions were checked by running the code. If you think an answer is wrong, tell us on the Codanics WhatsApp channel or in the YouTube comments and we will fix it.

Is the deep learning quiz framework-specific?

No. Concepts apply to PyTorch, TensorFlow and Keras. A few questions mention framework ideas such as fine-tuning a pre-trained model, without version-specific details.

Should I take the machine learning quiz first?

If you are new to the field, yes. Deep learning builds on ML ideas such as loss, overfitting and validation.

Is my data stored anywhere?

No. The quiz runs entirely in your browser. Your name, answers and score are not uploaded. Your best score and settings are remembered only in this browser so you can track your progress.

More free quizzes

See all Codanics quizzes → · Courses · Free tools · Blog

Follow Codanics

Created by Dr. Muhammad Aammar Tufail, Codanics. Published 18 November 2026, last updated 18 November 2026. Found a mistake or want a quiz on another topic? Tell us on the WhatsApp channel or contact us.