The Story of Yann LeCun rejection by ComVis orthadoxy (NCR days research)

For several years, YEM pushed convolutional neural networks in computer vision, but the vision community dismissed them as only suitable for small tasks like handwritten digits. Despite submitting a paper that outperformed other systems on a benchmark, it was rejected because the system learned everything without explicitly stating the knowledge or heuristics used. The computer vision paradigm dictated that knowledge should be explicit and mathematically implemented. By contrast, YEM's approach relied on learning everything without explicitly defining the knowledge. The machine learning community respected YEM but thought he was on the wrong path. It was not until Fei-Fei Li and her collaborators introduced the ImageNet competition that neural networks were proven to work well. Yann LeCun tried to get students to attempt the ImageNet competition with convolution nets but couldn't find any interested. Meanwhile, Ilio and Alex Fyshevsky became interested and successfully demonstrated the effectiveness of convolution nets. This turn of events was unfortunate for Yann LeCun as it was not his group that convinced the computer vision community about the superiority of neural networks over traditional methods.