AutoencoderCIFAR-10

Here the results of the first 8 topologies: two of them are wrong becouse of their latent space (#6, #7).

Autoencoder id	#1	#2	#3	#4	#5	#6	#7	#8
Loss (e^-3)	2.30	1.50	1.50	0.89	0.99	0.92	0.44	2.40
Latent Space	2048	2048	2048	2048	2048	8192	8192	2048

Autoencoder topology comparison w.r.t. different latent spaces

Autoencoder id	#1	#2	#3	#4	#5
Loss (e^-3) 2048 LS	2.30 5️⃣	1.50 4️⃣	1.50 3️⃣	0.89 🥇	0.99 2️⃣
Loss (e^-3) 1024 LS	5.20 5️⃣	5.00 4️⃣	4.70 2️⃣	3.00 🥇	4.70 3️⃣
Loss (e^-3) 512 LS	5.90 5️⃣	5.20 4️⃣	5.20 2️⃣	4.00 🥇	5.60 4️⃣
Loss (e^-3) 256 LS	----	6.10 3️⃣	5.70 2️⃣	4.80 🥇	15.80 4️⃣
Loss (e^-3) 128 LS	8.20 5️⃣	7.10 3️⃣	6.80 2️⃣	6.50 🥇	8.20 4️⃣
Loss (e^-3) 64 LS	10.20 4️⃣	9.50 3️⃣	8.80 🥇	9.30 2️⃣	18.90 5️⃣
Loss (e^-3) 32 LS	----	14.10 3️⃣	12.90 🥇	13.10 2️⃣	14.20 4️⃣
Loss (e^-3) 16 LS	----	18.80 3️⃣	17.30 🥇	18.00 2️⃣	19.50 4️⃣

Batch Normalized Topologies (50 epochs)

Obtained with learning rate 1e-04 and batch size 128

Autoencoder id	#1	#2	#3	#4	#5
Loss (e^-3) 2048 LS	2.9	2.5	24	1.2 🏆	1.6
Loss (e^-3) 1024 LS	2.8	2.6	2.4	1.3 🏆	2.3
Loss (e^-3) 512 LS	3.0	2.8	2.7	1.6 🏆	2.9
Loss (e^-3) 256 LS	3.4	3.2	3.5	2.9 🏆	4.9
Loss (e^-3) 128 LS	4.9	4.9	5.0 🏆	5.1	7.9
Loss (e^-3) 64 LS	7.8	7.7 🏆	7.8	8.1	10.1

Batch Normalized Topologies (75 epochs)

Obtained with learning rate 1e-04 and batch size 128

Autoencoder id	#1	#2	#3	#4	#5	loss gain
Loss (e^-3) 2048 LS	2.6	2.5	11	0.9 🏆	2.8	-0.3:heavy_check_mark:
Loss (e^-3) 1024 LS	2.5	2.5	2.2	1.2 🏆	2.0	-0.1:heavy_check_mark:
Loss (e^-3) 512 LS	2.7	2.5	2.5	1.5 🏆	2.7	-0.1:heavy_check_mark:
Loss (e^-3) 256 LS	3.2	3.2	3.2	2.8 🏆	4.5	-0.1:heavy_check_mark:
Loss (e^-3) 128 LS	5.0	4.9	4.9 🏆	5.0	4.8	-0.1:heavy_check_mark:
Loss (e^-3) 64 LS	7.7	7.7 🏆	7.9	8.1	10.1	+0.0:heavy_minus_sign: