Maswali yanayounganishwa na mwongozo · Ngumu Kiwango

Maswali ya Superposition na Polysemancity

Hundi huangazia nafasi ya juu zaidi, uchache, niuroni za polisemantiki, maelekezo yanayopishana na jiometri ya uwakilishi uliobanwa.

Njia za mwongozo zinazohusianaSuperposition na Polysemancity
Swali 1 ya 8

'Superposition' inarejelea nini katika mitandao ya neural?

Hakiki kila swali8 maswali

Majibu na maelezo huonekana unapojibu maswali.

  1. 'Superposition' inarejelea nini katika mitandao ya neural?
    • Inawakilisha vipengele vingi kuliko neurons kwa kuingiliana maelekezo yao
    • Kuweka miundo mingi juu ya nyingine
    • Kuendesha mfano mara mbili na wastani
    • Kuongeza kelele kwa pembejeo wakati wa mafunzo
  2. Ni hali gani hufanya superposition kufanya kazi vizuri licha ya kuingiliwa kwa kipengele?
    • Vipengele vinavyotumika sana wakati wote
    • Vipengele vikiwa vichache, kwa hivyo vichache vinatumika kwa wakati mmoja
    • Kwa kutumia uanzishaji chanya pekee
    • Kuwa na nyuroni nyingi kuliko vipengele
  3. Niuroni ya 'polysemantiki' ni nini?
    • Neuroni ambayo haiwashi kamwe
    • Neuroni inayotumika tu kwenye safu ya pato
    • Neuroni inayojibu dhana nyingi zisizohusiana
    • Neuroni isiyo na miunganisho
  4. Ni karatasi gani ya Anthropic iliyoanzisha utafiti unaodhibitiwa wa hali hii mnamo 2022?
    • Umakini Ndio Wote Unaohitaji
    • Uainishaji wa ImageNet na CNN za kina
    • Inacheza Atari na Deep RL
    • Mifano ya Toy ya Superposition
  5. Ikiwa safu lazima ihifadhi vipengele vingi kuliko ilivyo na vipimo, ni nini kinachoweza kuepukika kijiometri?
    • Baadhi ya vipengele lazima vishiriki maelekezo yanayopishana, yasiyo ya orthogonal
    • Vipengele vinaweza kukaa kikamilifu katika orthogonal
    • Mtandao lazima uvunjike
    • Vipengele vyote vinafanana
  6. Kwa nini utangulizi hufanya ukalimani kuwa mgumu zaidi?
    • Inafanya mfano kuwa polepole kukimbia
    • Neuroni moja haipati tena ramani kwa ukamilifu kwa dhana moja
    • Inaondoa vipengele vyote kutoka kwa mfano
    • Inathiri tu safu ya pembejeo
  7. Why does sparse feature activity reduce interference when more features than dimensions are represented?
    • Sparse activity makes all feature directions orthogonal
    • It increases the number of neurons automatically
    • Fewer features are active at the same time, so overlapping directions collide less often
    • It removes the need for nonlinear computation
  8. What does an antipodal pair of feature directions look like geometrically?
    • Two vectors at a right angle in separate dimensions
    • Two vectors with identical directions and magnitudes
    • A vector with no coordinates
    • Two directions that point opposite ways along the same line