Scientific Training Events

Spring School 2016 held in Fraunhofer IDMT, Ilmenau, Germany 4th-8th April 2016.

The SpaRTaN-MacSeNet Spring School on Sparse Representations and Compressed Sensing Spring School was aimed at graduate students, researchers and industry professionals working in this fast moving and exciting area. The five day school was split into two components, during three days, a panel of experts offered lectures and tutorials covering the theory of sparse representations, compressed sensing and related topics; alongside applications of these methods in areas such as image processing, audio signal processing, and signal processing on graphs. The remaining two days were devoted to software carpentry, giving researchers the computing skills they need to get more done in less time and with less pain

Find out more details at the Spring School Webpage.

Research Skills Training Events

The network has held 2 training weeks covering transferable skills for researchers and all our ESRs have attended one of these weeks. The first was in Surrey in September 2015 and the second preceded the Ilmenau Spring School in April 2016. Both training weeks focussed on transferable skills aligned to the Vitae Researcher Development Framework and were provided by SURREY’s Researcher Development Team who as well as providing courses for SURREY’s researchers have had experience running similar events for other ITNs and SEPNet (the South East Physics Network).

We wanted to cover topics which were important to all researchers as well as some of the more specific areas that are likely to come up for our highly mobile ESRs. The theme of the first sessions of both training weeks reflected this by covering unwritten rules, those between different cultures caused by different countries and languages as well as those caused by different stages of career or different education experiences. The other sessions covered the skills researchers need to plan and develop their career, they were formally introduced to the Vitae Researcher Development Framework and were encouraged to look at the skills a researcher needs. In week one this day included an employer panel discussion which allowed the ESRs to talk to senior researchers or managers in SMEs, International Companies, Start-ups and Academia. The panellists were asked to come with an idea of what they looked for when recruiting researchers so that the ESRs could ask questions about what it was like to work in various environments and what skills were specific or common across employers. Following this the ESRs were able to use the Action Planning tools to help them identify areas where they needed further training. During the second training week the employer panel was swapped for networking skills to enable them to make the most of the industry and academic visitors at the summer school that was about to follow. Other sessions looked at the skills researchers need to present their research and engage with different audiences and Dr Sophie Wehrens, the MacSeNet Ethics Advisor, prepared a session on understanding ethics in research with examples for the ESRs to discuss and analyse to give them an insight into how ethics applies to their research.

Project Co-ordinator

Prof. Mark Plumbley, Centre for Vision, Speech and Signal Processing, University of Surrey, UK.

Project Administrator

Dr Helen Cooper, Centre for Vision, Speech and Signal Processing, University of Surrey, UK.

Contact

helen.cooper@surrey.ac.uk

NEWS: Looking for a Research Job? We're still recruiting one more Experienced Researcher, details on our recruitment page

The SpaRTaN Initial Training Network will train a new generation of interdisciplinary researchers in sparse representations and compressed sensing, contributing to Europe’s leading role in scientific innovation.

By bringing together leading academic and industry groups with expertise in sparse representations, compressed sensing, machine learning and optimisation, and with an interest in applications such as hyperspectral imaging, audio signal processing and video analytics, this project will create an interdisciplinary, trans-national and inter-sectorial training network to enhance mobility and training of researchers in this area.

SpaRTaN is funded under the FP7-PEOPLE-2013-ITN call and is part of the Marie Curie Actions — Initial Training Networks (ITN) funding scheme: Project number - 607290

We're Recruiting

SpaRTaN is an EU-funded Marie Curie Initial Training Network, bringing together leading academic and industry groups to train a new generation of interdisciplinary researchers in sparse representations and compressed sensing, with applications in areas such as hyperspectral imaging, audio signal processing and video analytics.

There are eight Marie Curie Early Stage Researcher (ESR) positions, which allow the researcher to work towards a PhD, and two Marie Curie Experienced Researcher (ER) Positions for candidates who already have a PhD or equivalent research experience. The ESRs will be recruited to start during the first half of 2015 for a duration of 36 months and the ERs will be recruited to start towards the end of 2015 for a duration of 24 months.

Each ESR and ER will have secondments linked to their research to other partners in the network. They will also attend ITN progress meetings and Training events throughout Europe and possibly conferences and events internationally.

Marie Curie ESRs and ERs are paid a competitive salary which is adjusted for their host country. Please see the individual positions below to find the annual salary for that host country (figures are given in Euros prior to employer and employee tax being deducted). Since the ESR and ER positions include secondments to other hosts and for the researcher to move countries the EU also provides a Mobility Allowance, this is higher for researchers who have a family (family is defined as persons linked to the researcher by (i) marriage, or (ii) a relationship with equivalent status to a marriage recognised by the national legislation of the country of the beneficiary or of the nationality of the researcher, or (iii) dependent children who are actually being maintained by the researcher).

ESRs should be within four years of the diploma granting them access to doctorate studies at the time of recruitment. ERs need to be within the first five years of years of their research career, i.e from the date of their diploma granting access to doctorate studies. (See the figure to the right)

In addition, to be eligible for a position as a Marie Curie Early Stage Researcher or Experienced Researchers you must not have spent more than 12 months in the host country in the 3 years prior to starting.

In order to be sure that you meet these requirements please fill in the eligibility form and include it with your application.

(Position Filled) ESR1 : Sparse Time-Frequency methods for Audio Source Separation - CVSSP, University of Surrey, United Kingdom

Location: CVSSP, University Of Surrey, United Kingdom

Marie Curie Annual Allowance (Pre-Employer/Employee Tax)*: €51,072

Marie Curie Annual Mobility Allowance (Pre-Employer/Employee Tax)*: €11,289/€16,128

Approximate Country Specific Comparable salary: £36,087.97/£38,836.84 (Gross including Mobility)

Start Date: From April 2015

Duration: 36 Months

Closing Date for Applications: 2015-01-25

Human listeners appear to have an innate and effortless ability to isolate and attend to one sound source while suppressing others. This remarkable phenomenon is called the cocktail party effect. We would like to be able to perform the same task for musical audio signals, and extract a signal corresponding to one instrument from a stereo (or mono, or 5.1 surround sound) mixture of musical audio. Since the mixing and source processes are not known beforehand, this problem is often known as blind source separation (BSS).

The candidate will be responsible for developing source separation algorithms, particularly for the difficult case of more sources than sensor channels. We wish to achieve excellent perceptual quality of reconstructed sources for applications such as: music separation and remixing; audio denoising, restoration and enhancement; music information retrieval, and interactive music performance systems. Methods to be studied include sparse representations, time-frequency masking, Bayesian probabilistic inference, non-negative matrix factorization (NMF), beamforming, and independent component analysis (ICA). This project will also investigate the use of knowledge of the notes from a musical score to identify notes from different instruments to help the separation process.

CVSSP is one of the major research centres of Surrey’s Department of Electronic Engineering (EE), the top ranked UK EE department in both the RAE 2008 and in the national league tables. CVSSP is the largest research centres in the UK focusing on Computer Vision, graphics and signal processing, with 120+ members comprising academic and support staff, research fellows and PhD students.

Informal enquires are welcome and should be made to Dr Wenwu Wang or Prof Mark Plumbley.

Requirements

The successful applicant is expected to have an excellent mathematical and programming background, with an MSc or equivalent related to signal processing and audio engineering. Programming skills in Matlab or C/C++ are required.

Applications

Application via jobs.surrey.ac.uk (job ref 084014).

Please ensure your application includes your completed eligibility form and a letter of motivation including a maximum 1-page statement explaining how your research interests, skills and experience are relevant to the position;

Direct link to this advert (right click and copy link).

(Position Filled) ESR2 : Automatic Music Transcription using Structured Sparse Dictionary Learning - CVSSP, University of Surrey, United Kingdom

Location: CVSSP, University Of Surrey, United Kingdom

Marie Curie Annual Allowance (Pre-Employer/Employee Tax)*: €51,072

Marie Curie Annual Mobility Allowance (Pre-Employer/Employee Tax)*: €11,289/€16,128

Approximate Country Specific Comparable salary: £36,087.97/£38,836.84 (Gross including Mobility)

Start Date: From April 2015

Duration: 36 Months

Closing Date for Applications: 2015-01-25

Human perception of sounds is much more advanced than any technical system so far created. However, transcription of polyphonic music (i.e. converting music with many instruments or notes at the same time into a written "score" notation) is a problem that is hard even for trained human listeners. Previous research into automatic music transcription (AMT) using sparse representations. However, it was discovered that the representations need to be clustered into groups, and that there are strong correlations over time. Recent developments in structured sparse representations, such as group sparsity and tree-structured sparsity, offer the potential to significantly improve these results, as a fruitful tool to encode prior knowledge about the physics of signal processing and machine learning problems.

In this PhD project, the candidate will investigate new methods for automatic music transcription, focussing on dictionary learning methods related for sparse and structured sparse representations. Methods to be investigated include the use group sparsity as part of the decomposition learning process; tree-structured sparsity, with notes modelled as parts of chords with different likelihoods; and structuring over time instead of independent decomposition for each time frame.

CVSSP is one of the major research centres of Surrey’s Department of Electronic Engineering (EE), the top ranked UK EE department in both the RAE 2008 and in the national league tables. CVSSP is the largest research centres in the UK focusing on Computer Vision, graphics and signal processing, with 120+ members comprising academic and support staff, research fellows and PhD students.

Informal enquires are welcome and should be made to Dr Wenwu Wang or Prof Mark Plumbley.

Requirements

The successful applicant is expected to have an excellent mathematical and programming background, with an MSc or equivalent related to signal processing and audio engineering. Programming skills in Matlab or C/C++ are required.

Applications

Application via jobs.surrey.ac.uk (job ref 084314).

Please ensure your application includes your completed eligibility form and a letter of motivation including a maximum 1-page statement explaining how your research interests, skills and experience are relevant to the position;

Direct link to this advert (right click and copy link).

(Position Filled) ESR3 : Sparse Representations and Compressed Sensing - University of Edinburgh, United Kingdom

Location: Institute for Digital Communications, University of Edinburgh, United Kingdom

Marie Curie Annual Allowance (Pre-Employer/Employee Tax)*: €51,072

Marie Curie Annual Mobility Allowance (Pre-Employer/Employee Tax)*: €11,289/€16,128

Approximate Country Specific Comparable salary: £31,924 (Gross) + Mobility Allowance

Start Date: From March 2015

Duration: 36 Months

Closing Date for Applications: 2015-03-02

One PhD position is available jointly at the Institute for Digital Communications and the Brain Research Imaging Centre at the University of Edinburgh, UK. The selected candidate will study the next generation of compressed sensing (CS) techniques for accelerated acquisition in MRI. The project will explore the benefits of CS advanced imaging modalities, such as quantitative imaging, MR spectroscopy and diffusion tensor imaging.

This PhD will investigate the next generation of compressed sensing (CS) techniques for accelerated acquisition in MRI.

Magnetic Resonance Imaging has already shown itself to be an ideal candidate for the application of compressed sensing theory. Excellent image reconstruction has been shown to be possible from undersampled k-space measurements through the application of compressed sensing principles and algorithms. However the real challenge and benefits lie in tackling advanced MR imaging techniques, such as quantitative MRI, MR spectroscopy and diffusion tensor imaging. These techniques require very long acquisition times and can suffer from bad motion artefacts induced during acquisition. These problems go beyond traditional CS solutions, and to tackle them will require the development of new structural signal models and sampling strategies.

The Early Stage Researcher on this project will benefit from the partnership between IDCOM and BRIC in the University of Edinburgh and the other academic institutions and industrial partners within the SpaRTaN project. They will attend initial training events and be exposed to the research activities of all participants at regular six monthly progress meetings. They will engage in training events and secondments to other project partners.

Informal enquires are welcome and should be made to Prof. Mike Davies.

Requirements

Highly motivated, excellent candidates should ideally hold a valid Masters degree with a specialisation in medical imaging and/or signal processing and should be eligible for immediate admission on the PhD programme at the University of Edinburgh.

Applications

Application is via the university job pages, vacancy reference : 032432.

Please ensure your application includes your completed eligibility form and a letter of motivation including a maximum 1-page statement explaining how your research interests, skills and experience are relevant to the position;

Direct link to this advert (right click and copy link).

(Position Filled) ESR4 : Task Based Dictionary Learning for Audio-Visual Tagging - LTS2, EPFL,Switzerland

Location: LTS2, EPFL,Switzerland

Marie Curie Annual Allowance (Pre-Employer/Employee Tax)*: €45,448

Marie Curie Annual Mobility Allowance (Pre-Employer/Employee Tax)*: €10,046/€14,352

Approximate Country Specific Comparable salary: 55938 CHF (Gross including Mobility)

Start Date: February to May 2015

Duration: 36 Months

Closing Date for Applications: 2015-01-15 or 2015-04-15 if not filled

Audio-visual archives are so huge that tagging operations must be particularly computationally efficient and scalable. Machine learning techniques based on semi-supervised learning have proved particularly efficient at reducing expert interactions with large volumes of data, and therefore offer an important technological platform for this problem. In this project we will investigate deep learning algorithms operating on patch based dictionaries of features trained on specific tagging tasks; tackle hierarchies of problems, from classifying musicians to classifying audio-visual events; and contribute new technologies to organise, interact with it and offer new human-computer interfaces to explore the Montreux Jazz Festival audio-visual archives.

This is one of two doctoral student positions open at the Laboratory of Signal Processing 2, LTS2, at EPFL, Lausanne, Switzerland, focusing on the application of compressive sensing and sparsity based methods to machine learning with applications to audio-visual data.

The LTS2 is specialized in signal/image/data processing, graph and networks analysis as well as machine learning.

The lab offers a stimulating research environment with an open minded and collaborative team, state of the art IT facilities (multicore servers and GPU units), as well as acoustics facilities (anechoic and reverberant rooms).

Informal enquires are welcome and should be made to Prof. Pierre Vandergheynst or Dr Benjamin Ricaud.

Conditions of employment

This is a full time position as PhD researcher within the Laboratory of Signal Processing 2. The candidate will dedicate its time to research as well as be interacting with Master students and participate in teaching.

Requirements

A candidate with an MSc degree or equivalent either

  • in electrical engineering, computer science or applied mathematics with a strong interest in acoustics.
  • in acoustics with an interest for mathematics and a good experience in programming and scientific numerical computations.

Experience with programming in one or more languages is mandatory, a good knowledge of Matlab or Python is a plus. We expect a good level of English. Ability to travel within the network is essential, as currently one position includes secondments in Edinburgh (three months) and Finland (three months) and the other in London (six months).

Applications

To apply the candidate must register and get admitted to the EPFL doctoral school « EDEE » before 15th of January 2015 or if the post is not filled after this date « EDIC » before 15th of April 2015.

EDEE Application EDIC Application

Please ensure your application includes your completed eligibility form and a letter of motivation including a maximum 1-page statement explaining how your research interests, skills and experience are relevant to the position;

Direct link to this advert (right click and copy link).

(Position Filled) ESR5 : 1-bit Compressive Imaging - LTS2, EPFL,Switzerland

Location: LTS2, EPFL,Switzerland

Marie Curie Annual Allowance (Pre-Employer/Employee Tax)*: €45,448

Marie Curie Annual Mobility Allowance (Pre-Employer/Employee Tax)*: €10,046/€14,352

Approximate Country Specific Comparable salary: 55938 CHF (Gross including Mobility)

Start Date: February to May 2015

Duration: 36 Months

Closing Date for Applications: 2015-01-15 or 2015-04-15 if not filled

One of the main promised impacts of CS is the development of new, more efficient sensors. Recently, it has been shown that CS measurements can be radically quantised to only a single bit (0/1), and yet the original signal can still be efficiently and robustly recovered. This striking result could open the way to dramatically simplified sensors, such as cameras with ultra-simple analog and digital electronics. In parallel, many current computer vision and augmented reality applications describe image content by means of localised binary descriptors that drive classifiers. These binary descriptors can be implemented on mobile platforms very efficiently. There is therefore a convergence of technologies based on 1-bit measurements that has not yet been explored. The outcome of this research could result in brand new imaging chips for mobile applications that would be ultra efficient in terms of power consumption but would also allow the direct extraction of low level features for computer vision applications. In this project, we will interact with a hardware group and help design novel hardware based on 1-bit sensing; demonstrate a standalone visual sensor that directly collects 1-bit measurements; decode measurements to form images; and apply with no decoding for high-level computer vision and augmented reality applications.

This is one of two doctoral student positions open at the Laboratory of Signal Processing 2, LTS2, at EPFL, Lausanne, Switzerland, focusing on the application of compressive sensing and sparsity based methods to machine learning with applications to audio-visual data.

The LTS2 is specialized in signal/image/data processing, graph and networks analysis as well as machine learning.

The lab offers a stimulating research environment with an open minded and collaborative team, state of the art IT facilities (multicore servers and GPU units), as well as acoustics facilities (anechoic and reverberant rooms).

Informal enquires are welcome and should be made to Prof. Pierre Vandergheynst or Dr Benjamin Ricaud.

Conditions of employment

This is a full time position as PhD researcher within the Laboratory of Signal Processing 2. The candidate will dedicate its time to research as well as be interacting with Master students and participate in teaching.

Requirements

A candidate with an MSc degree or equivalent either

  • in electrical engineering, computer science or applied mathematics with a strong interest in acoustics.
  • in acoustics with an interest for mathematics and a good experience in programming and scientific numerical computations.

Experience with programming in one or more languages is mandatory, a good knowledge of Matlab or Python is a plus. We expect a good level of English. Ability to travel within the network is essential, as currently one position includes secondments in Edinburgh (three months) and Finland (three months) and the other in London (six months).

Applications

To apply the candidate must register and get admitted to the EPFL doctoral school « EDEE » before 15th of January 2015 or if the post is not filled after this date « EDIC » before 15th of April 2015.

EDEE Application EDIC Application

Please ensure your application includes your completed eligibility form and a letter of motivation including a maximum 1-page statement explaining how your research interests, skills and experience are relevant to the position;

Direct link to this advert (right click and copy link).

(Position Filled) ESR6 : Analysis Dictionary Learning Beyond Gaussian Denoising - Instituto de Telecomunicações, Portugal

Location: Instituto de Telecomunicações, Portugal

Marie Curie Annual Allowance (Pre-Employer/Employee Tax)*: €32,300

Marie Curie Annual Mobility Allowance (Pre-Employer/Employee Tax)*: €7,140/€10,200

Approximate Country Specific Comparable salary:

Start Date: April to September 2015

Duration: 36 Months

Closing Date for Applications: 2015-2-15 or 2015-04-15 (if not filled)

The dominant paradigm in sparsity-based regularization of inverse problems is that of synthesis sparsity, where the unknown signal is modeled as having a sparse representation on a dictionary of functions. Originally these dictionaries were fixed (e.g. wavelets bases or frames) but the current state-of-the-art methods use dictionaries that are learned from data. Dictionary learning methods were initially limited to simple Gaussian denoising problems, and have only recently been extended to harder problems, such as image deconvolution and reconstruction, or to problems involving non-Gaussian noise.

Analysis formulations are dual of the synthesis ones, in that the unknown signal is modeled as yielding a sparse result, when a so-called analysis operator is applied to it. Although analysis formulations have been widely used in denoising, deconvolution, and reconstruction, with Gaussian and other types of observation noise, only fixed analysis operators have typically been used. Very recently, the problem of learning operators for analysis sparsity formulations has started to be addressed, but so far only for the classical Gaussian denoising problems. The objective of this work is to extend analysis dictionary learning to scenarios harder than Gaussian denoising, namely, deconvolution, reconstruction, and non-Gaussian observations.

Informal enquires are welcome and should be made to Prof. Mário Figueiredo.

Requirements

Candidate are expected to hold an MSc degree (or equivalent), in electrical engineering, computer science, applied mathematics, or related areas, with a solid background on mathematics and, preferably, signal processing. Candidates are also expected to have a good scientific programming experience, preferably in Matlab and/or Python, a very good level of English, and availability to travel within the network.

Applications

To apply for the position, please provide:

(i) a letter of motivation including a maximum 1-page statement explaining how your research interests, skills and experience are relevant to the position;

(ii) a CV including publication list;

(iii) names and contact details of three referees willing to write confidential letters of recommendation.

(iv) Your completed eligibility form.

All materials should be emailed as a single PDF file (<5 Mb) to: mtf@lx.it.pt with ‘PhD application SpaRTaN ESR6′ in the subject line.

Direct link to this advert (right click and copy link).

(Position Filled) ESR7 : Compressed Sensing for Hyperspectral Imaging - Instituto de Telecomunicações, Portugal

Location: Instituto de Telecomunicações, Portugal

Marie Curie Annual Allowance (Pre-Employer/Employee Tax)*: €32,300

Marie Curie Annual Mobility Allowance (Pre-Employer/Employee Tax)*: €7,140/€10,200

Approximate Country Specific Comparable salary:

Start Date: April to September 2015

Duration: 36 Months

Closing Date for Applications: 2015-2-15 or 2015-04-15 (if not filled)

In Hyperspectral imaging (HSI), the sensors measure the electromagnetic energy scattered in their instantaneous field view in hundreds (even thousands) of spectral channels. The very high spectral resolution of HSI enables a precise identification of the sensed materials via spectroscopic analysis, facilitating countless applications; e.g., earth observation and remote sensing, food safety, pharmaceutical process monitoring and quality control, as well as biomedical, industrial, and forensic applications.

This projects focus on compressive hyperspectral imaging (CHSI), whereby data compression is implemented simultaneously with the acquisition, by computing a number of projections, termed measurements, of the original data onto a set of predesigned vectors. Assuming that the original data admits a sparse representation on a given basis or frame, it can be recovered from the projections by solving a suitable optimization problem. CHI is of paramount importance in spaceborne systems, due to the extremely large volumes of data collected by the imaging spectrometers onboard satellites, and the low bandwidth of the connections between them and the ground stations.

The success of CHSI stems from the very high spectral and spatial correlation of this type of data, meaning that it is compressible, i.e., it admits a representation on a given frame in which most of the coefficients are small and, thus, it is well approximated by sparse representations. Due to the huge number of optimization variables involved in a typical CHI problem (values between 10^8 and 10^10 are usual), the solutions of these optimization problems are very demanding from the computational point of view. To make the problem even more challenging, the sparse representations are not known beforehand and have to be learned from the measurements.

Informal enquires are welcome and should be made to Prof. Jose Bioucas Dias.

Objectives

The main goals of this project are a) to develop effective adaptive sparse representations for hyperspectral data, namely by exploiting that fact that the spectral vectors usually leave in a data-dependent low dimensional subspace, b) to develop optimal measurement matrices in the sense of minimal number of measurements, and c) to develop optimization strategies to infer simultaneously the sparse representation and the original hyperspectral image.

Requirements

Candidates should hold an MSc degree in Electrical and/or Computer Engineering, Computer Science, or Applied Mathematics. The required skills include, with a flexible balance, advanced programming skills (C/C++, Matlab),strong background in statistics, signal processing and/or optimization, a good knowledge of English, and a strong motivation to work as part of a team.

Applications

To apply for the position, please provide:

(i) a letter of motivation including a maximum 1-page statement explaining how your research interests, skills and experience are relevant to the position;

(ii) a CV including publication list;

(iii) names and contact details of three referees willing to write confidential letters of recommendation.

(iv) Your completed eligibility form.

All materials should be emailed as a single PDF file (<5 Mb) to: bioucas@lx.it.pt with ‘PhD application SpaRTaN ESR7′ in the subject line.

Direct link to this advert (right click and copy link).

(Position Filled) ESR8 : Large-scale signal processing - INRIA, France

Location: INRIA/CNRS/ENS Paris, France

Marie Curie Annual Allowance (Pre-Employer/Employee Tax)*: €44,118

Marie Curie Annual Mobility Allowance (Pre-Employer/Employee Tax)*: €9,752/€13,932

Approximate Country Specific Comparable salary: €40,800 (Gross)

Start Date: April to September 2015

Duration: 36 Months

Closing Date for Applications:2015-02-13

Machine learning methods are becoming widespread for signal processing, in most areas of science, engineering and the economy. These data-driven approaches require an explicit management of uncertainty. The PhD student is expected to develop methodologies to compute a degree of confidence in predictions, that allows subsequent users to appropriately take decisions. Within this project, we will particularly focus on high-dimensional problems with potentially a large number of observations and predictions based on convex optimization.

The 36 month project will be undertaken within the Computer Science Department of Ecole Normale Superieure, located in downtown Paris, within the INRIA/CNRS/ENS project-team SIERRA.

Informal enquires are welcome and should be made to Dr Francis Bach

Requirements

The applicant must have a Masters degree in Computer Science or Applied Mathematics (or any equivalent diploma), with knowledge of convex optimization, machine learning, and numerical linear algebra.

Applications

To apply for the position, please provide:

(i) a letter of motivation including a maximum 1-page statement explaining how your research interests, skills and experience are relevant to the position;

(ii) a CV including publication list;

(iii) names and contact details of three referees willing to write confidential letters of recommendation.

(iv) Your completed eligibility form.

All materials should be emailed as a single PDF file (<5 Mb) to: francis.bach@inria.fr with ‘PhD application SpaRTaN ESR8′ in the subject line.

Direct link to this advert (right click and copy link).

ER1 : Video Analytics for Large Camera Networks - VisioSafe, Switzerland

Location: VisioSafe, Switzerland

Marie Curie Annual Allowance (Pre-Employer/Employee Tax)*: €69,966

Marie Curie Annual Mobility Allowance (Pre-Employer/Employee Tax)*: €10,046/€14,352

Approximate Country Specific Comparable salary:

Start Date: Negotiable From Feb 2016

Duration: 24 Months

Closing Date for Applications: Open until filled

One Experienced Researcher position is available.

Over the past years, VisioSafe have installed hundreds of cameras to get in-depth insights on human mobility in large public scenes such as train terminals. We have collected more than 75 million trajectories aggregated as temporal Origin-Destination (OD) matrices from two train terminals during one year.

The selected candidate will analyze such Big Data to identify behavioural patterns in transportation settings. More specifically, we aim to address how predictable is human behaviour. Can we learn structures in the OD matrices to infer recurrent and abnormal patterns?

The outcome of the project will help to understand the space usage, to forecast human flows, and to simulate the impact of infrastructure changes.

Planned Secondments

Scientific: EPFL (3 months): for training in sparse representations on graphs

Requirements

Candidates should hold a masters degree in image processing, computer science and/or engineering, or related areas - the degree must have been completed 3-5 years before the date they start their position in order to be eligible for this funding. Experience in image-processing algorithms is essential. Candidates are also expected to have good skills in programming (C and C++), proficiency in English, the right to work in Switzerland and availability to travel within the SpaRTaN network.

Applications

To apply for the position, please provide:

i a letter of motivation explaining how your research interests, skills and experience are relevant to the position.

ii a detailed CV.

iii your completed eligibility form.

All materials should be emailed to info@visiosafe.com with ‘Application SPARTAN ER1′ in the subject line.

Direct link to this advert (right click and copy link).

(Position Filled) ER2 : Image and Video Restoration with Adaptive Transforms - Noiseless Imaging, Finland

Location: Noiseless Imaging, Finland

Marie Curie Annual Allowance (Pre-Employer/Employee Tax)*: €69,849

Marie Curie Annual Mobility Allowance (Pre-Employer/Employee Tax)*: €10,029/€14,328

Approximate Country Specific Comparable salary:

Start Date: Negotiable From July 2015

Duration: 24 Months

Closing Date for Applications: 2015-07-03

One Experienced Researcher position is available. The selected candidate will study and design computationally efficient algorithms for image, video, and multidimensional data restoration, including denoising, deblurring, blind deblurring and super-resolution. In particular, the considered algorithms will be based on a sparse representation with respect to data-driven adaptive transforms, where the adaptivity follows from a nonlocal spatial or spatio-temporal analysis of the data. The results shall be applicable to challenging problems of industrial significance.

Planned Secondments

Scientific: Instituto de Telecomunicações, Portugal, 3 months.

Requirements

Candidates should hold a master or doctoral degree in image processing, computer science and/or engineering, applied mathematics, or related areas, with a strong background in linear algebra. Experience in image-processing algorithms is essential. Candidates are also expected to have good skills in scientific programming (preferably Matlab and/or C), proficiency in English, and availability to travel within the SpaRTaN network.

Applications

To apply for the position, please provide:

i. A letter of motivation (maximum one page) explaining how your research interests, skills and experience are relevant to the position.

ii. A detailed CV, including list of publications. If any, copies of the publications most relevant to the position.

iii. Names and contact details of three referees willing to write confidential letters of recommendation.

iv. Your completed eligibility form.

All materials should be emailed as a single PDF file to ni@noiselessimaging.com with 'Application SPARTAN ER2' in the subject line.

Direct link to this advert (right click and copy link).

* Salary adjusted for host country prior to employer and employee tax being deducted, as such the gross and net amount received by the ESR/ER will be different from that listed. Where possible the mobility allowance will be untaxed but this will depend on country tax regulations. Salaries converted into local currency are approximate at the time of writing and may change, each ESR/ER will receive their full Euro entitlement after taxes regardless of the host currency.

Our Researchers

ESR1 Alfredo Zermini: Sparse Time-Frequency methods for Audio Source Separation - CVSSP, University of Surrey, United Kingdom

Alfredo was born in Senigallia (Italy), he has lived in several Italian towns during his life. He studied physics at the University of Bari (Italy) and then particle physics during his Masters at the University of Pisa.

During his Masters thesis Alfredo built and studied a series of particle detectors prototypes: his results have been used to design the new drift chamber prototype for the MEG II experiment in Zurich (Switzerland).

Alfredo decided to move to engineering for a variety of reasons; the large collaborative groups in particle physics make it difficult to feel ownership of any new discovery and the idea of working on audio has always been something that had appealed, even before starting University. This project offered the right opportunity to merge his interests both in audio and the pure science fields. He is currently studying for his PhD on audio source separation at the University of Surrey (UK).

This individual project focusses on using Deep Neural Networks (DNNs) to predict the Direction Of Arrival (DOA) of audio sources in an audio mixture with respect to the listener. The DOA is then used to generate soft/binary time-frequency masks for the recovery/estimation of the individual audio sources.

ESR2 Cian O'Brien: Automatic Music Transcription using Structured Sparse Dictionary Learning - CVSSP, University of Surrey, United Kingdom

Cian is from Tipperary, Ireland. He attended Trinity College Dublin where he obtained an Honours degree in Mathematics, during which time he also studied classical and jazz music at the Royal Irish Academy of Music. For his Masters degree, he spent two years at Georgia Institute of Technology and graduated with an MSc in Music Technology. His work focussed on sparse dictionary learning applications to Music Information Retrieval problems such as genre recognition and music mood estimation.

This individual project is Automatic Music Transcription (AMT). Given an audio file of a piece of music, the goal of AMT is to produce a "pitch-time" representation which gives the musical pitches present in the signal at each time-frame.

ESR3 Wajiha Bano: Sparse Representations and Compressed Sensing - University of Edinburgh, United Kingdom

Wajiha is from Pakistan. She did her Masters at Abo Akademi University, Finland. During her masters, she received a scholarship from Zeno Karl Schindler foundation, Geneva for her masters thesis in cardiovascular magnetic resonance (CVMR) research group in University Hospital Lausanne. The aim of the thesis was to characterize the accuracy and precision of cardiac T2 mapping as a function of different influences such as signal-to-noise ratio, cardiac phases, off-resonance frequencies using both numerical simulations and in-vivo imaging. The project was presented in International Society of Magnetic Resonance in Medicine meeting and received Magna Cum Laude award and subsequently was published as in magnetic resonance in medicine (MRM).

Also during her masters, Wajiha worked on another research project in the University of Oulu, Finland that involved studying imaging characteristics of articular cartilage degeneration with clinical ultrasound device which was presented in the Osteoarthritis Research Society meeting. Wajiha's research interests are accelerated image acquisition and reconstruction in Magnetic Resonance Imaging and its clinical applications.

The general aim of the project is to utilize the compressed sensing theory and algorithm to accelerate Magnetic Resonance Imaging (MRI). MRI is a useful clinical tool to diagnose diseases but is limited due to the lengthy acquisition time. This is important as less amount of information is acquired due to slow acquisition that subsequently affects the image quality and also the patient have to spend substantial amount of time in the scanner. The qualitative nature of the MR images can be addressed by quantifying the parameters (T1 and T2 relaxation) used to produce contrast in the image. This is known as magnetic resonance parametric mapping. It is considered a valuable tool for the quantitative assessment of brain structure and function. It has been proved to be useful in the study of brain ageing, Parkinson’s disease, epilepsy, multiple sclerosis etc. However, the clinical utility of conventional parameter mapping is limited due to the lengthy acquisition times which subsequently results in low spatial resolution and employing low number of time points for parameter fitting. The acquisition of undersampled data is a potential solution for faster parameter mapping which can be achieved by either parallel imaging\cite{MRM:MRM10044} or by compressed sensing (CS)\cite{lustig2007sparse}. Compressed sensing has three major components; signal sparsity, incoherent sampling and non-linear sparsity-promoting reconstruction. All three conditions are met naturally by MRI and thus making it an ideal application of CS.

ESR4 Rodrigo Pena: Task Based Dictionary Learning for Audio-Visual Tagging - LTS2, EPFL,Switzerland

Rodrigo was born in Brasilia, Brazil, in 1991. He did most of his undergraduate studies at the University of Brasilia (UnB), earning a degree in Electrical Engineering in 2014. In the meantime, he also studied Electronic Engineering for a year at ENSEIRB-MATMECA, in Bordeaux, France, in the context of a Brazil-France exchange program put in place by their governments. He started his graduate studies at UnB, focusing on image and video processing, more specifically on the incorporation of models for prediction of salient regions to automatic image and video quality estimation. Now a PhD candidate at EPFL, Rodrigo is working on the Signal Processing Laboratory 2 (LTS2), supervised by Prof. Pierre Vandergheynst.

ESR5 Konstantinos Pitas: 1-bit Compressive Imaging - LTS2, EPFL,Switzerland

Konstantinos is a PhD student at the Signal Processing Laboratory LTS2 at Ecole Polytechnique Federale de Lausanne. His hometown is Thessaloniki located in the northern part of Greece. It is there that he completed his Bachelor and Master's degrees in Electrical and Computer Engineering. Konstantinos completed his masters project as an exchange student at EPFL, on the subject of dictionary learning for graph signals. He is interested in supervised dictionary learning, graph signal processing and neural networks.

ESR6 Milad Niknejad: Analysis Dictionary Learning Beyond Gaussian Denoising - Instituto de Telecomunicações, Portugal

Milad is a PhD student in Instituto Superior Tecnico (IST), under supervision of Professor Mario Figueiredo. He got his Bachelor and Masters degrees both from Iran. His current research interests are statistical signal processing and image processing.

The topic of this research project is in patch-based image restoration, and is related to the work previously covered in Milad's MSc thesis, which has led to journal and conference publications. The general goal of the current research is to improve and extend that previous in several directions.

ESR7 Lina Zhuang: Compressed Sensing for Hyperspectral Imaging - Instituto de Telecomunicações, Portugal

Lina is from China and received a Bachelor of Science in geographic information system and a Bachelor of Economics from South China Normal University, Guangzhou, China, in 2012. After that, she got a M.S. degree at the Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing, in 2015. Currently, she is pursuing a Ph.D. degree in Instituto de Telecomunicacoes, Instituto Superior Tecnico, Lisbon, Portugal. Lina is working on inverse problems in hyperspectral imaging (namely, in unmixing, denoising, and superresolution), supervised by Prof. Jose M. Bioucas-Dias and Prof. Mario Figueiredo.

In hyperspectral imaging (HSI), the sensors measure the electromagnetic energy scattered in their instantaneous field view in hundreds (even thousands) of spectral channels. The very high spectral resolution of HSI enables a precise identification of the sensed materials via spectroscopic analysis, facilitating countless applications; e.g., earth observation and remote sensing, food safety, pharmaceutical process monitoring and quality control, as well as biomedical, industrial, and forensic applications.

My research focuses on inverse problems in hyperspectral imaging, namely unmixing, denoising, and superresolution. Due to low spatial resolution of hyperspectral cameras, microscopic material mixing, and multiple scattering, the measurements acquired by those cameras are mixtures (linear and nonlinear) of spectra of materials in the scene under study. Thus, accurate estimation from acquired HSIs requires unmixing, which is a blind source separation (BSS) problem. Hyperspectral unmixing is a challenging ill-posed inverse problem, not only because it is a BSS, but also due to model inaccuracies, observation noise, environmental conditions, endmember variability, and data set size. Among these degradation mechanisms, endmember variability and nonlinear unmixing are studied in my research under the Bayesian framework.

HSIs acquired by imaging instruments are often noisy owing to a number of degradation mechanisms such as electronic noise, Poissonian noise, quantization noise, and atmospheric effects. My research mainly considers the principal sources of noise, including Gaussian noise, Poissonian noise, and mixture of both noise. Owing to a very high spatial-spectral correlation and spatial self-similarity, this class of images has low rank structure and admit sparse representations on suitable frames. These characteristics of HSIs will be explored to devise fast and effective HIS denoising algorithms.

My research also involves the joint design of fast hyperspectral sensing strategies onboard and superresolution and fusion algoritms on ground stations. A computational imaging perspective is adopted in the joint design of “smart” sensing and computational algorithms to recover the original data. According to this perspective, the image is computed by algorithms designed cooperatively with the sensing strategy. The design of fast hyperspectral sensing strategies is of paramount importance in spaceborne hyprespectral systems due to the extremely large volumes of data collected by the imaging systems onboard satellites and the low bandwidth of the connections between them and the ground stations The fast sensing strategies will be conceived to yield easy, from a conditioning point of view, superresolution and fusion inverse problems. I’ll research acquisition strategies which adaptively control the blurriness and be able to acquire data with different resolutions for different spectral channels, having into consideration that hyperspectral images have low rank and a strong degree of selfsimilarity.

ESR8 Damien Scieur: Large-scale signal processing - INRIA, France

Damien started his studies in 2010 in Faculte Polytechnique de Mons (Belgium) in engineering, then he moved 2012 in École Polytechnique de Louvain (Belgium). His masters was in applied mathematics with a particular focus on optimization.

He did his master thesis with Yurii Nesterov on "Global complexity analysis for the second-order methods". The goal was to design efficient methods for some complex functional classes using both first and second order information.

Damien also worked inside the university with Raphael Jungers and Julien Hendrickx on switched systems. The goal was to implement efficient algorithms to compute certificates on the stability of such systems in the context of a Matlab toolbox (\textit{JSR toolbox}, available online on the Mathworks website).

He also worked with Anthony Papavasiliou and Leopold Cambier on a Matlab toolbox which solve structured stochastic optimization problems (available on GitHub and on the website http://www.baemerick.be/fast/). For example, this toolbox can be used to manage electricity production and consumption when some uncertain amount of energy is available (wind energy, solar panel, water dam, ...).

This project is working on acceleration algorithms. Like the name suggests, the goal of such an algorithm is to make other methods faster. The idea is to design a generic method which is able to accelerate the rate of convergence of some other methods without any knowledge of the methods.

The goal of this algorithm is non-trivial. If we can achieve this goal, this will lead to more efficient methods, allowing us to solve bigger problems in less time, or more accurately.

ER1 : - VisioSafe, Switzerland

Position Still Vacant

ER2 Zhongwei Xu: Image and Video Restoration with Adaptive Transforms - Noiseless Imaging, Finland

Zhongwei Xu received the B.S. in Electrical Engineering, and M.S. in computer science, from the Xidian University, China, in 2008, and 2011, respectively. He received the Ph.D. in computer science from the Universitat Autonoma de Barcelona, Spain, in 2015. His research interests are the restoration and enhancement of digital images, and image and video coding.

The objective of this post-doc project is: multi-spectral 3D scene reconstruction using images from different bands. It involves two main parts:
1) non-local 3D point cloud filtering, based on enhanced sparse representation in transform domain;
2) automatic co-registration for images from different spectral bands.

Full Partners


Centre For Vision Speech and Signal Processing
University of Surrey
www.surrey.ac.uk/CVSSP
Mark Plumbley
Wenwu Wang
Institute for Digital Communications
The University of Edinburgh
www.eng.ed.ac.uk/research/institutes/idcom
Mike Davies
LTS2
Ecole Polytechnique Fédérale de Lausanne
lts2www.epfl.ch
Pierre Vandergheynst
Benjamin Ricaud
Instituto de Telecomunicações (IT)
www.it.pt
Jose Bioucas Dias
Mario Figueiredo
Institut National de Recherche en Informatique et en Automatique
www.inria.fr / www.di.ens.fr
Francis Bach
Visiosafe S.A.
www.visiosafe.com
Noiseless Imaging
www.noiselessimaging.com

Associate Partners

Tampere University of Technology
www.tut.fi
Fraunhofer-Gesellschaft zur Foerderung der Angewandten Forschung E.V.
www.idmt.fraunhofer.de
Cedar Audio Ltd.
www.cedar-audio.com

Peer Reviewed Publications

2016
[2] Lina Zhuang, José M Bioucas-Dias, "Fast Hyperspectral Image Denoising Based on Low Rank and Sparse Representations", In Geoscience and Remote Sensing Symposium (IGARSS), 2016 IEEE International, 2016. [bib]
[1] Rodrigo Pena, Xavier Bresson, Pierre Vandergheynst, "Source Localization on Graphs via l1 Recovery and Spectral Graph Theory", In ArXiv e-prints, 2016. [bib] [pdf]

Powered by bibtexbrowser