Скачать или смотреть Resnets in 60 seconds

This video is dedicated to Sun Jian (1976 - 2022), one of the authors of the ResNet, Faster R-CNN, ShuffleNet and other notable publications. His contributions to the AI community will always be rememberd and celebrated.
----------------------
As Neural Networks become deeper and deeper, they are subject to accuracy degradation. This decline is represented by a vanishing or exploding gradient. It occurs when the training and test error rate increases as the number of layers is increased.

Resnet was designed to solve this problem. Instead of having each layer connected to the next layer, activations of a layer are connected to further layers by skipping some layers in-between, forming residual blocks. Resnets are then created by stacking these residual blocks together.

ResNet is considered one of the fastest image classification algorithms, taking first place in the imagenet and coco detection tasks. The popular versions of resnets include resnet-50, resnet-101 and resnet-152. There's actually a resnet-1202 which means 1,202 layers but the test result was worse than a 110-layer network, although they had fairly similar training error.
#computervision #artificialintelligence #shorts
----------------------

I'll be adding more shorts to this new playlist along with long, say 8 minute how-to videos in the coming days. If real artificial intelligence also excites you, be sure to comment/email me what topic to simplify next.

Cheers, and happy coding!

P.S.: I'll be super surprised if the YouTube algorithm drives even the slightest traffic here in the first few weeks. Prepped for long rainy days so please share the channel link with your friends in tech :)