Jul 13
  • CiteSeerX Search

    Free-access scientific literature digital library and search engine.

  • Journal Impact Factors

    Browse and download the latest free-access impact-factor listing of more than 6200 scientific journals from the Information Centre for Aerospace Science & Technology (ICAST), India.

  • Open Access Journals

    Browse through a huge listing of open-access scientific journals.

  • MammoImage.org

    Mammography-related papers, publications, databases and many more from the University of Zagreb, Croatia.

  • EKT.gr

    EKT - National Documentation Centre of Greece, access to the digital library and a vast volume of on-line material.

ASSURANCE: Material and Results PDF Print E-mail
Written by Harris Georgiou   
Saturday, 22 June 2013 00:00


ASSURANCE: "Distributed Sparse Models for Machine Learning in Telecommunications"


The purpose of this research proposal is to develop models as well as training strategies and algorithms for ML, that apply both to classification as well as to regression in a unified way. Thus, although our emphasis is on Signal Processing, the algorithms to be developed can be used in any related ML task. Our approach will incorporate, in the stage of training, all the previously mentioned characteristics that a modern ML system is required to possess, by more carefully "listening" on how human intelligence operates. More specifically, our methodology and algorithmic framework will be developed so that:
a) to be online and time adaptive, in order to be able to track variations of the statistics of the environment,
b) to incorporate a-priori knowledge in the form of constraints,
c) to operate in a distributed learning environment, where a number of nodes exchange information,
d) to be sparsity-aware with respect to the structure of the target system's model and be able to incorporate this information in an efficient way,
e) to be able to accommodate non-linear models in a computationally efficient manner.
Non-linearity is also another feature of the involved processing in the human brain. Controlled non-linearity can drastically improve performance in many practical systems, at the expense of higher computational complexity.
Keywords: Machine learning, sparse models, distributed systems, telecommunications

Material and Results:

Presentations and work conducted by AUA (Harris Georgiou):

  1. Brain Signals & Sparsity: Update on current state-of-the-art of EEG& fMRI processing with sparsity-aware methods (Athens, GREECE, Feb.2013).
    http://dx.doi.org/10.5281/zenodo.59608 [download: pdf]
Technical reports co-authored by CTI/UoA (Harris Georgiou):

  • ...

Software and Datasets:

There are several currently running code & data development side-projects linked with the main ASSURANCE project. These are some of the resources I have personally developed and can publish here as-is (read the licensing info included). Please note that these resources are under continuous development and updates (beta versions), so some features may not yet be fully implemented or stable.

  • fMRI-Sparse toolbox (Matlab): Currently under development, a minimalistic collection of low-level data handling (matrix) functions for fMRI processing, block-based & event-based test pattern series, as well as "realistic" simulated fMRI data series for algorithm benchmarking, template scripts for various fMRI decomposition methods (GLM, PCA, ICA, BP, ...), analysis of components & activation maps, etc. Some features are not yet fully implemented. For detailed info and downloads see main page.


Creative Commons License EU Public License
All the documents and related material by Harris Georgiou are licensed, in parts and as a whole, under a Creative Commons Attribution-Non-Commercial-Share Alike 3.0 Unported License. All the code sources and related material by Harris Georgiou are licensed, in parts and as a whole, under a EU Public License.

Last Updated on Thursday, 04 August 2016 16:37