GoConqr has a new look! Let us know what you think.

Sign up free

Flashcards

NN

Martin Honěk English 55 cards

NN

Cards in this deck (55)

  1. 1Define activity - semantics of a networkDefine activity - semantics of a network
  2. 2basic definition of neural networkbasic definition of neural network
  3. 3define architecture of neural networkdefine architecture of neural network
  4. 4Describe basic types of NN architecturesDescribe basic types of NN architectures
  5. 5Describe activity propertiesDescribe activity properties
  6. 6Define activity - semantics of a networkDefine activity - semantics of a network
  7. 7Write a inner potential formulasWrite a inner potential formulas
  8. 8Write typical activation functions and draw graphsWrite typical activation functions and draw graphs
  9. 9Draw XOR neural networkDraw XOR neural network
  10. 10Write a line equation for weights 2, 2 and bias -1Write a line equation for weights 2, 2 and bias -1
  11. 11Define properties of LearningDefine properties of Learning
  12. 12What is the goal of the learning?What is the goal of the learning?
  13. 13Name the adventages of using neural networkName the adventages of using neural network
  14. 14Draw NN for boolean functions + !x1 ^ (!x2 U x1)Draw NN for boolean functions + !x1 ^ (!x2 U x1)
  15. 15Draw example of nonlinear separation for boolean functionsDraw example of nonlinear separation for boolean functions
  16. 16Proof that two layers MLP can compute arbitrary boolean functionsProof that two layers MLP can compute arbitrary boolean functions
  17. 17Proof (by image) that arbitrary input space R^k can be approximated by three layer networkProof (by image) that arbitrary input space R^k can be approximated by three layer network
  18. 18Describe cybenko theorem for non-linear separationDescribe cybenko theorem for non-linear separation
  19. 19ALVINN - describe. how many neurons etc.Viz. ipad
  20. 20Proof that arbitrary function can be approximated by three layer networkProof that arbitrary function can be approximated by three layer network
  21. 21Describe cybenko approximation theoremDescribe cybenko approximation theorem
  22. 22Describe computability for recurrent networkDescribe computability for recurrent network
  23. 23Summarize neural network - how strong they are boolean, function approximationSummarize neural network - how strong they are boolean, function approximation
  24. 24describe neural network vs classical computers (table)describe neural network vs classical computers (table)
  25. 25Why online algorithm work? describe theoremWhy online algorithm work? describe theorem
  26. 26Proc pouzivame sogmoid pro modelování pravdědobnosti?Pravdědopobnost nedokážeme modelovat pomocí linárního modelu. Transformujeme tedy pravdědobnost na odds.tu zlogaritmujeme dostaneme zkoro lineární funkci.
    DUL.
    Jestliže namodelujeme log odds pomocí lineární funkci, pravděpodobnost získáme aplikací logistic sigmoid na výsledek lineární funkce.
  27. 27Jake jsou problemy malé gradient descentJake jsou problemy malé gradient descent
  28. 28Popis problem lokalnich minim u malych siti a identifiability problem?Popis problem lokalnich minim u malych siti a identifiability problem?
  29. 29Jake jsou horsi problemy gradient descent a vysvětlitJake jsou horsi problemy gradient descent a vysvětlit
  30. 30Teorém týkající se gradient descent?Teorém týkající se gradient descent?
  31. 31Popis vanishing a exploding? Variance estimationg + vzorec?Popis vanishing a exploding? Variance estimationg + vzorec?
  32. 32Popis minibatch size vlastnosti?Popis minibatch size vlastnosti?
  33. 33Popis memomentumPopis memomentum
  34. 34Jak najit optimalni learning rate?Jak najit optimalni learning rate?
  35. 35Popis druhy adaptivniho learning rate?Popis druhy adaptivniho learning rate?
  36. 36Na jake znalosti funguje adagrad ohledne learning ratu?Na jake znalosti funguje adagrad ohledne learning ratu?
  37. 37Napis a popis vzorec pro adaGrad?Napis a popis vzorec pro adaGrad?
  38. 38Na jake znalosti funguje rmtrop ohledne learning ratu?Na jake znalosti funguje rmtrop ohledne learning ratu?
  39. 39Napis a vysvetli rmspropNapis a vysvetli rmsprop
  40. 40Popis other optimalizacePopis other optimalizace
  41. 41Jake vlastnosti by mela mit dobra aktivacni funkceJake vlastnosti by mela mit dobra aktivacni funkce
  42. 42hyperbolic tangens, proc 1.729 atd.Viz poznamky.
  43. 43Proc chceme input preprocessing? Jakym zpusobem?Proc chceme input preprocessing? Jakym zpusobem?
  44. 44Inicializace vah pro hyperbolic tangens?Inicializace vah pro hyperbolic tangens?
  45. 45Obecne popis inicializaci vah, vypocetObecne popis inicializaci vah, vypocet
  46. 46popis glorot a bengo inicializacipopis glorot a bengo inicializaci
  47. 47Smooth relu vzorecSmooth relu vzorec
  48. 48Jaké aktivacni funkce pro output neurony?Jaké aktivacni funkce pro output neurony?
  49. 49Popis sigmoidal functions with cross enthropyPopis sigmoidal functions with cross enthropy
  50. 50Popis generalizationPopis generalization
  51. 51Early stoping?Early stoping?
  52. 52Size of the network?Size of the network?
  53. 53Ensemble methods?Ensemble methods?
  54. 54dropout?dropout?
  55. 55Weight decay a L2 regularizationWeight decay a L2 regularization

Information

Description
Flashcards on NN, created by Martin Honěk on 22/01/2021.
Created by
Martin Honěk
Views
0
Rating

Feedback

What is it about?

What do you think of the new look?

How do you like it?