<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
</head>
<body text="#000000" bgcolor="#FFFFFF">
Уважаемые сотрудники ИТФ,<br>
<br>
В пятницу 12 апреля 2019 года состоится коллоквиум, на котором будет
заслушан доклад:<br>
<br>
Ян Фёдоров (King's College London)<br>
<b>On reconstructing nonlinearly encrypted signals corrupted by
noise.</b><b><br>
</b><br>
I consider the problem of reconstructing a source vector from its
encrypted image corrupted by an additive Gaussian noise. Assuming
encryption to be given by a random Gaussian mapping, the
reconstruction problem in the framework of the Least Square Scheme
turns out to be equivalent to finding the configuration of minimal
energy in a certain version of spherical spin glass model. As a
measure of the quality of the signal reconstruction one can use the
mean overlap between the original signal and its recovered image.
Thi overlap is analysed in the framework of Parisi scheme of Replica
Symmetry Breaking. If the mapping is quadratic, there exists a
threshold in the noise-to-signal ratio beyond which the
reconstruction is impossible. The behaviour close to the threshold
is controlled by the replica symmetry breaking mechanism and is
characterized by a nontrivial exponent 3/4. <br>
</body>
</html>