Post by account_disabled on Feb 18, 2024 18:10:05 GMT 14
In images, the details analyzed by the network are located close to each other. And in a text, the connections between elements are long, because the beginning and end of a sentence can be separated by many words. Such neural networks are not suitable for tabular data. This is due to the fact that, unlike pixels, tabular data is heterogeneous (text, date, various lowercase and mathematical symbols). to analyze multi-sized images.
For example, when there are several documents and they are all of different sizes (A3, A4 and A5), but contain the same text. And if you don’t adjust all documents to the same size, then the neural network is unlikely to Phone Number List understand that the content of the text is the same for all of them. Frequently asked questions about convolutional neural networks When was the foundation of convolutional neural networks laid? The idea itself dates back to the 1950s, but interest in them was revived in 2012 after University of Toronto mathematicians Alec Krizhevsky and Geoff Hinton won the ImageNet image recognition competition.
Their neural network made it possible to reduce image recognition errors from 26% to 15%, which was a record figure. Currently, the accuracy of the convolutional neural network algorithm in recognizing faces in a crowd is higher than that of a person and is 99.8%. We are giving a discount of 60% on training “Artificial Intelligence from Scratch” until February 18 In just 9 months you can get a job with an income of 150,000 rubles Book a discount What are the features of convolutional neural networks compared to traditional ones? They can work effectively with images and videos due to their ability to capture local spatial dependencies in the data.