Function Pretrain():

i

}|

i

begin

for number of training iterations do

1

m

m

} from the source

1

m

m

i

=1

K

j

i

i

.

end

end

end

Function Finetune():

Data:

i

}|

i

i

}|

i

The number of adaptive layers, l.

The number of steps to apply to the discriminator, k.

i

}|

i

begin

for number of training iterations do

for i ← 1 to k do

m

S

m

T

Update the domain discriminator D by ascending its stochastic gradient:

D

1

m

m

i

i

i

end

m

T

ascending its stochastic gradient:

1

m

m

i

i

end

i

end

end

end

Algorithm 1: Adversarial Adaptive CNN (A2CNN)

9