CS7643l Quizl 2l (Latestl 2025/l 2026l Update)l Questionsl &l Answers|l Gradel A|l 100%l Correctl (Verifiedl Solutions)
Q:l Effectl ofl channelsl onl outputl size
Answer:
Itl doesn'tl havel effectl onl thel outputl size:l wel performl thel dotl productl forl eachl channelsl andl summingl theml up.
Q:l Effectl ofl channelsl onl parameters
Answer:
Eachl channell mightl havel itsl ownl weightsl withl respectl tol thel samel kernel.Ml xl (Chl xl K1l xl K2l +l 1)
Q:l Effectl ofl multiplel kernelsl (featurel extraction)l onl outputl size.
Answer:
Thel kernell sizel shouldl bel equall (K1l xl K2)l forl eachl kernell withinl thel layer.l Thel
outputl size:
(Hl -l K1l +l 1)l xl (Wl -l K2l +l 1)l xl Numberl ofl Kernels
Q:l Effectl ofl multiplel kernelsl (featurel extraction)l onl parameters
- / 2
Answer:
Eachl kernel,l eachl channell hasl itsl ownl setl ofl weights,l butl eachl kernell hasl onlyl 1l biasl term.(K1l xl K2l xl Channelsl +l 1)l xl M wherel Ml isl thel numberl ofl kernels
Q:l Whatl isl thel purposel ofl poolingl layer?
Answer:
Dimensionalityl reduction
Q:l Howl manyl learnedl parametersl doesl al maxl poolingl layerl have?
Answer:
None
Q:l Invariance
Answer:
Ifl thel featurel changes,l movesl orl rotatesl slightlyl onl thel image,l thel outputl valuel remainsl thel same.l (Forl example,l wel classifyl thel imagel ofl al catl regardlessl ofl wherel thel catl isl inl thel image)
Q:l Equivariance
Answer:
- / 2