2019 lightning-talk12
- 1. ConvNet Max pooling 螻 Full Network Pooling Layer (Sampling) 伎伎 觸 襷 螻 蟆 襷 Max Pooling: 2x2 螳 螳 觸 れ伎襷 一危 蠍磯 filter 蠍一 stride 蠍一 磯 覲蟆暑 Fully Connected Layer (FC Layer) Contains neurons that connect to the entire input volume, as in ordinary NN
- 3. ConvNet 覈 LeNet-5 (1998) Noise 麹 螳
- 4. ConvNet 覈 AlexNet (2012): Error rate 18.2% Ensemble 15.4% GoogLeNet (2015): Error rate 6.7% ResNet (2015): Error rate 3.6% 譴螳 蟇企磯 Fast Forward るジ れ 2 10-20% 蟆谿 AlphaGo (2016)