BrainOn Quiz

What is the main motivations behind physical neural networks :

• The existed models work well only with the input values between theminimum and the maximum values

• The existed models are less performant

• We can’t model all the problems linearly

• To create new type of artificial intelligence

In which case gradient descent performs well, in case our loss function is :

• Sigmoid

• Cubic

• Convex

• Affine

• Sinusoïdal

How do we call the gap between human-level performance and training error :

• bias

• avoidable bias

• variance

• human-level error

Which ones are object detection algorithms :

• Pose estimation

• Image classification

• YOLO

• Faster R-CNN

Which is the type of neural networks used in Face recognition

• Dense neural networks

• Siamese neural networks

• Generative Adversarial Networks

• Hopfield Networks

Convolutions are used in CNNs (Convolutional Neural Networks), they generally taketime to be run on the computer (high complexity), which technique can optimize it :

• Second-order optimization methods

• Grid Search

• Gradient Descent

• Fast Fourier Transform

What problems do LSTMs resolve :

• Gradient Vanishing

• Overfitting

• Out-Of-Vocabulary (OOV) words

• Limited generalization

What is the appropriate mathematical tool to model Graph Neural Networks:

• Graphs

• Trees

• Hypergraphs

• Forests

Let’s say we have large dataset of 1, 000, 000 rows, what is the best way to split ourdata :

• 70% training, 15% validation, 15% test

• 60% training, 20% validation, 20% test

• 30% training, 30% validation, 30% test

• 98% training, 1% validation, 1% test

What activation function do we generally use in Recurrent Neural Networks (RNNs) :

• Tanh

• Mish

• Relu

• SoftPlus