منصة ايفولف للخدمات الطلابية

ميد فيزياء للهندسة 1 https://t.me/Evolvesol/7310?single ميد أحياء عامة https://t.me/Evolvesol/7303 ميد تفكير تصميمي: https://t.me/Evolvesol/7300 https://t.me/Evolvesol/7287 ميد علوم البيانات: https://t.me/Evolvesol/7295?single ميد الكهرومغناطيسية https://t.me/Evolvesol/7293?single ميد ايفولف 2 فرع الليث https://t.me/Evolvesol/7288?single

بداية ميد نمذجة البيانات الغترة الأولى الان

ميد نمذجة البيانات الفترة الأولى الان

15. The image is a cat. The model predicts the probability of "cat". Which probability gives the Lowest cross-entropy loss? A. 0.01 B. 0.1 C. 0.5 D. 0.9 16. The following training and validation loss curves show that the model has a ........... (Curve shows both train and val loss decreasing and plateauing together) A. large learning rate, and needs to be decreased B. small learning rate, and needs to be increased C. suitable learning rate, and activation needs to be changed D. suitable learning rate, and nothing needs to be done 17. A model achieves 99% accuracy on the training set but only 70% accuracy on the test set. What is the most suitable action to improve the model? A. Add more layers and neurons to the model B. Train the model for more epochs C. Apply regularization techniques (e.g., dropout) or collect more training data D. Remove the validation set and train on all the data 18. Which of the following describes the Early Stopping technique? A. Training stops as soon as the training loss reaches zero B. The learning rate is reduced after every epoch C. Training stops after the first epoch to save time D. Training stops when the validation error starts to increase 19. To mitigate vanishing/exploding gradient problems, we can use good suitable ................ A. initial model weights and data normalisation B. activation function C. faster optimisers D. All the above 20. When designing a neural network architecture, what is the recommended strategy for dealing with model capacity? A. Start with small model with few neurons and gradually add layers B. Start with bigger model than necessary, then use regularization techniques to control overfitting C. Design model with medium size that suitable to the problem D. Start with medium model without any regularization to reduce the computational time

1. What is the main difference between the McCulloch-Pitts neuron and the Perceptron? A. The McCulloch-Pitts neuron learns its weights, but the Perceptron uses fixed weights B. The Perceptron learns its weights, but the McCulloch-Pitts neuron uses fixed weights C. The Perceptron accepts only inputs of 0 or 1 D. The McCulloch-Pitts neuron does not use a threshold 2. Why do we add a bias term to a Perceptron? A. So the decision boundary does not have to pass through the origin B. To increase the learning rate during training C. To make all the weights positive D. To reduce the number of input features 3. A fully connected network has 2 inputs, two hidden layers with 3 units each, and 1 output unit. How many parameters must be learned? A. 18 B. 7 C. 25 D. 21 4. A neural network has many hidden layers, but it does NOT use any non-linear activation function. Which statement is true? A. It can model any non-linear function B. It is still a linear function, so it cannot model non-linear patterns C. It always overfits the training data D. It does not need a loss function 5. Which activation function should be used in the output layer to classify an image into one of 10 classes? A. ReLU B. Tanh C. Sigmoid D. Softmax 6. What is the correct order of the steps when training a neural network? A. Forward pass -> compute loss -> backward pass -> update weights B. Forward pass -> update weights -> backward pass -> update weights C. Backward pass -> forward pass -> compute loss -> update weights D. Forward pass -> Backward pass -> compute loss -> update weights 7. Why do units with Sigmoid or Tanh activation learn slowly when their input value z is very large or very small? E. Because the derivative is close to zero, so the weights change very little F. Because the derivative becomes very large, so the weights explode G. Because the output of the unit becomes negative H. Because these functions are not differentiable 8. What is the main problem of the ReLU activation function, and how does ELU help? A. ReLU is slow to compute, and ELU is faster B. ReLU saturates for large positive inputs, and ELU does not C. ReLU gives zero gradient for negative inputs so units can "die", and ELU gives a small gradient for negative inputs D. ReLU outputs only negative values, and ELU outputs only positive values 9. What happens if all neurons in a layer start with the same weights and biases? A. The network trains faster B. All neurons compute the same output and learn the same features C. The vanishing gradient problem is solved D. Each neuron learns a different feature 10. The input features of a dataset are on very different scales (for example, age from 0 to 100 and salary from 0 to 100,000). What is the best step before training? A. Standardize the features so the mean is 0 and the standard deviation is 1 B. Increase the learning rate C. Remove the feature with the larger values D. Set all the weights to zero 11. What is an advantage of momentum optimization compared to the vanilla gradient descent? A. It removes the need for a learning rate B. It never oscillates around the minimum C. It does not add any new hyperparameter D. It can roll past local minima 12. During training, the training loss and the validation loss both increase steadily together. What is the most likely cause and fix? A. Overfitting, so we should add dropout B. Underfitting, so we should add more layers C. The learning rate is too high, so we should lower the learning rate D. Training is healthy, so no fix is needed 13. In L2 regularization, Error = Σ(y - ŷ)² + α Σw². What happens when we increase the value of α? A. The weights are pushed closer to 0, and the decision boundary becomes smoother B. The weights become larger, and the decision boundary becomes more complex C. The training error always becomes zero D. The bias terms are penalized more than the weights 14. Each batch normalization layer learns (γ) and (β). If a network has 3 batch normalization layers, how many trainable batch normalization parameters must be learned? A. 3 B. 12 C. 9 D. 6

شرح الاستاذه هدى لنموذج الميد Evolve 2

نماذج الميد الأول ايفولف 3: https://t.me/Evolvesol/5852?single https://t.me/Evolvesol/5995?single نماذج الميد الأول ايفولف 2: https://t.me/Evolvesol/1418 https://t.me/Evolvesol/7185?single https://t.me/Evolvesol/7175?single تجميعات ايفولف 1 الميد الاول: https://t.me/Evolvesol/7177?single

كويز نمذجة البيانات

كويز تكامل ومخازن البيانات

كويز: Applied Statistics for Data Science MTH2902T

كويز البرمجة

كويز تفاضل

كويز الفيزياء

شرح مجاني شامل فترة المد✨ فيزياء : https://t.me/physica1Amqura تفاضل: https://t.me/Math1UmmQura

شرح ايفولف 3 المنهج كامل مجانا