Sie befinden sich hier: Termine » Prüfungsfragen und Altklausuren » Hauptstudiumsprüfungen » Lehrstuhl 5 » dl-august-19   (Übersicht)

Unterschiede

Hier werden die Unterschiede zwischen zwei Versionen gezeigt.

Link zu dieser Vergleichsansicht

pruefungen:hauptstudium:ls5:dl-august-19 [29.08.2019 13:35] (aktuell)
Muetzi angelegt
Zeile 1: Zeile 1:
 +First he wanted me to give an overview over the topics of the lecture by drawing a mindmap
  
 +Then he wanted me to explain the Perceptron
 + => Schematic
 + => XOR → MLP
 + => Universal Function Approx
 + => Multiple Layer: Deep Learning
 +
 +Name a loss functions that we discussed?
 + => L2 for regression, CE for classification,​ both optimal wrt. MLE
 + => Derivation for L2 by assuming Gauss
 +
 +How does the network learn now?
 + => Backpropagation + gradient descend
 + => Chain rule: Multiplication of gradients + weight update
 + => Exploding/​vanishing gradients
 +
 +What other method did we use to encode the Information?​ (Not quite sure about the wording here)
 + => Activation Functions: Sigmoid/​Tanh → ReLU 
 + => prevent vanishing gradients
 +What about Dying ReLU?
 + => Leaky ReLU
 +
 +What is an regularization alternative fighting the internal covariate shift?
 + => Batchnormalization
 +
 +Can you draw and explain the LSTM structure?
 +
 +What are GANs?
 + => Generator vs Discriminator trained by MiniMax
 +There was a problem called Mode Collapse, please explain it.
 +
 + => TODO
 + => D focuses only on one feature → G also
 +
 +
 +Can you explain Cycle Consistent GANs?
 + => Principle(trainiable inverse mapping) + combined loss function explained
 +
 +What is an Autoencoder?​
 + => Encoder-Decoder
 + => Undertetermined AE + Sparse AE
 +
 +
 +What is common for U-Net and Autoencoder?​ Whats the difference?
 + => Same: Encoder-Decoder structure
 + => Different: (Conv Layer), but mainly skip connections
 +
 +How good would U-Net be in comparison to the AE?
 + => thanks to the skip connections it could basically copy the input