Tutorials on Recurrent Neural Networks and Signature Verification

Részletek / Details: részletes szakmai program, részletek/details

Recurrent Neural Networks

In this tutorial several Recurrent Neural Networks (RNNs) and their application to Pattern Recognition will be described. First, a brief history of RNNs is presented. Next, several problems of simple RNNs are described and the Long Short-Term Memory (LSTM) is presented as a solution for those problems. For a better understanding of the network, its behaviour on several toy problems and real-world PR-applications is investigated. Finally, extended architectures, such as the bi- and multi-directional LSTM will be proposed and their application to speech, handwriting and other PR-domains will be given. Existing Open-Source Toolkits implementing the LSTM and some extensions will be presented and an introduction of how to use these tools will be given.
Recently, LSTMs became quite popular. They work reliably on speech data and won the first place on many international handwriting competitions. Surprisingly, they do not require any preprocessing, nor feature extraction, they work on raw pixel data.

Signature Verification

In this tutorial several issues related to automatic signature verification and its application in real world forensic scenarios will be discussed. First, signature verification will be broadly defined from the Pattern Recognition (PR) perspective and working of such systems will be briefly described. Later the Forensic Document Examiners' (FDEs) perspective of signature verification will be highlighted and an introduction to various genres of handwriting FDEs deal will be provided. In a hands-on session, participants will be involved in solving some example cases in order to make them understand various implications associated with forensic signature verification. Then, a detailed man vs. machine comparison, w.r.t. signature verification, will be provided where machines show a potential to assist humans in real world forensic scenarios. Finally, various issues raised by FDEs concerning the output produced by today's automatic signature verification systems will be discussed and some potential solutions will be delivered.