Institute for Theoretical Information Technology

Seminar Communication Theory (S3/4)

Seminar on Deep Learning - Methodologies and Applications

Dates

Seminar dates

Deadline for the application procedure in the campus system is 02.04.2019.

The introductory meeting takes place on Thursday, 04.04.2019, 10:30 am, in room 333, ICT cubes, Kopernikusstraße 16.

Content

Blind Source Separation

• Requirements: Basic knowledge of mathematics and programming. Fluency in written and spoken English.
• Description: The cocktail party describes the problem of separating a desired signal from the interference and noise caused by other simultaneous active sources.dHere we will be applying important data analytics tools such as Independent Component Analysis (ICA), Principle Component Analysis (PCA) to separate audio signals.dWe also provide the opportunity for real world implementation of the developed solution.

Compressed Sensing using Generative Models

• The paper examines how generative adversarial networks can be used for classical signal processing tasks. In a remarkable result, it is shown how MNIST images can bedreconstructed only from 10 samples. This means that MNIST images can be compressed to around 1 percent of the original dimension and that in in a very efficient way. The method significantly outperforms LASSO, a well known reconstruction method with established theoretical guarantees.

Improving information security via full-duplex jamming receivers

• Requirements: Being familiar with the basics of wireless communication systems, system modeling. Being familiar with (or interested in) the basics of mathematical optimization. Interest in learning the new concepts/technologies in the context of wireless communications.
• Abstract: A Full-duplex (FD) transceiver is characterized by the capability to receive and transmit at the same time and frequency. As a result, an FD information receiver has the capability to transmit a jamming signal while receiving useful information, and thereby deteriorating the information reception by a potential eavesdropper. This enhances the physical layer security of the communication system. The aformetioned system concept and optimization will be the focus of this seminar work.

ISTA-Net: Iterative Shrinkage-Thresholding Algorithm Inspired Deep Network for Image Compr. Sensing

• Iterative thresholding algorithms have been used successfully for signal recovery in context of compressed sensing applied for instance to functional magnetic resonance imaging (fMRI) applications and single pixel camera. Although very fast compared to convex-programming approaches, the method can be stil enhanced if the thresholding is chosen efficiently tailored to different datasets. In this work a deep thresholding network is constructed combining strength of thresholding algorithms and gradient-based neural networks to construct a fast and efficient recovery method.

Optimal sensor deployment under various conditions

• Requirements: Basic knowledge of wireless sensor networks and communication systems. A very good knowledge of mathematics (in particular, various optimization techniques). A very good knowledge of problem modeling in terms of network parameters.
• Abstract: The sensor deployment strategy has a huge impact on the performance of a wireless sensor network (WSN) as well as its effective lifetime. The topic of sensor placement covers various transmitter and receiver deployments and has been widely studied, where different deployment criteria have been considered for specific applications. The majority of the studies have focused on the issues such as coverage and connectivity, target coverage, reliability, cost, delay, and energy and lifetime, in order to define the objective of the optimization problem which yields the desired sensor positions. Recent studies have taken into account the physical parameters of communications, such as the propagation delay and the angle of arrival, in order to find the optimal solutions. This topic can be studied in both two and three dimensional scenarios. The optimal trajectory design for unmanned aerial vehicles (UAVs) is an application example of the generalization of this topic. Due to the high complexity, most of the formulated optimization problems in this context cannot be tackled by conventional techniques.

Prerequisites

A successful participation in at least one course offered by the Chair of Theoretical Information Theory. (Decision on admission in individual cases.)

Aims

Students learn to work autonomously by dealing with a scientific topic of interest for the institute. Normally, a current research paper is thoroughly worked through. Background material is carefully prepared and missing details are provided by the student. Also literature surveys may be composed. Moreover, simulations may be implemented and evaluated. This work will be documented in a student report and presented to other seminarians and colleagues by using modern presentation techniques.

Exam

Transcript after successful participation (written report and presentation).

Introductory Meeting

On this event the prerequisites will be checked, topic distribution will be checked, and a general introduction to the course and the requirements of the seminar will be given.

Consultation hours

Dr. Michael Reyer upon agreement

Back to the list of lectures