Epoch when the rule-based paradigm lost to the weighted sum paradigm

from Think of society as a human computation Epoch in which the rule-based paradigm lost to the weighted paradigm The epoch in which the logical block (hand-designed feature/rule-based) paradigm was decisively defeated by the weight-learning deep learning (CNN) paradigm was AlexNet's victory in the ILSVRC-2012 (ImageNet Large Scale Visual Recognition Challenge) on September 30, 2012. Challenge) on September 30, 2012. AlexNet won by more than 10 percentage points, recording 15.3% of the top 5 errors compared to 25% in 2011, the previous year when hand-designed features dominated. Since that moment, the "weighted paradigm" has continued to dominate SOTA in image recognition, reaching 3.57% of ResNet in 2015 and nearly 91% Top-1 accuracy for ViT-G/14 and CoCa since 2023.

• 2013-2015: Rapid CNN expansion: ZFNet, GoogLeNet, etc., with errors going from 10 % to 6-7 % range; ResNet-152 (2015) achieved 3.57 % in Skip Connection, below humans (5-6 %)
• The essence of what brought about the shift was the value shift of "not designing features, but letting them learn through data and computational power.

This page is auto-translated from /nishio/ルールベースパラダイムが重み付き和パラダイムに負けたエポック using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.