tag:blogger.com,1999:blog-3963186461850074185.post197322445513629394..comments2022-03-26T19:11:55.299-07:00Comments on Deep Learning: Sky's the Limit?: Part 2: AlphaGo under a Magnifying Glass (The historic match of deep learning AlphaGo vs. Lee Sedol) REVISEDBob van den Hoekhttp://www.blogger.com/profile/16913830394696106545noreply@blogger.comBlogger4125tag:blogger.com,1999:blog-3963186461850074185.post-54682037679095049362017-02-07T07:10:38.875-08:002017-02-07T07:10:38.875-08:00Hey there,
the last graphic is a bit confusing to...Hey there,<br /><br />the last graphic is a bit confusing to me. The game-patterns that are displayed below the convolutional layers range from large patterns near the input to smaller, more detailled ones near the output. Shouldn't this be the other way round?<br /><br />I do agree that convolutional layers farther away from the input usually have smaller feature maps, as depicted in the graphic. However, am I wrong in believing that their neurons' inputs essentially depend on larger parts of the image? Aren't the very local features, like the GO-pattern on the lower right, being detected by the first convolutional layer and the rather large features, like the one on the left, being detected by layers to the right, since they process information about the presence of local features?<br /><br />Maybe I just interpreted the graphic the wrong way or I am making another mistake. Hope you can clear up my confusion.<br /><br />Kind regards,<br />KieronKieronnoreply@blogger.comtag:blogger.com,1999:blog-3963186461850074185.post-15142276535978311052016-05-13T00:37:20.181-07:002016-05-13T00:37:20.181-07:00The dimensions and shape of the L1, L2 L3 in the f...The dimensions and shape of the L1, L2 L3 in the figure shown are just a very simple example of the architecture of a convolutional neural network, analogue to the network architecture as applied to AlphaGo. However, the dimensions and shapes shown are not as have been used for AlphaGo (starting with 48 feature planes of 19x19 images).Bob van den Hoekhttps://www.blogger.com/profile/16913830394696106545noreply@blogger.comtag:blogger.com,1999:blog-3963186461850074185.post-77076836947976033092016-05-12T23:53:08.078-07:002016-05-12T23:53:08.078-07:00Hi Bob,
Is that correct the shape of the weights i...Hi Bob,<br />Is that correct the shape of the weights in L1, L2 and L3 in the last figure(above)?<br />ThnaksUnknownhttps://www.blogger.com/profile/00034606926450649522noreply@blogger.comtag:blogger.com,1999:blog-3963186461850074185.post-44516513431006945642016-04-06T09:35:09.485-07:002016-04-06T09:35:09.485-07:00In this blog, an overview is given of the differen...In this blog, an overview is given of the different AI components of AlphaGo as well as details about the teaching material, learning methods and skills of the program. <br /><br />In the first part (Under a small magnifying glass), the most important neural network components are put to spotlight along with their goal and function as part of AlphaGo. In the second part (Under a large magnifying glass), the characteristics of convolutional neural networks that AlphaGo is based on are covered in more detail.<br />Bob van den Hoekhttps://www.blogger.com/profile/16913830394696106545noreply@blogger.com