General Information Regarding All PhD Theses :

1. Salary :
PhD students are salaried employees of MINES ParisTech, net salary being around 1500 euros per month.

2. Location :

Centre de recherche en informatique - CRI
MINES ParisTech
35 rue Saint-Honoré
77305 Fontainebleau Cedex

A free, private bus is running daily from Paris, except during school holidays, when public transportation via RATP has to be used.


Offres de thèse :

  1. Comparison between abstract numerical domains
  2. Compilation for embedded real-time multimedia systems
  3. Structured Code and Non-Deterministic Automaton
  4. Generalized expression evaluation optimization
  5. Energy-Centric Optimization of Computer Applications
  6. Energy-Aware Compilation for Communicating Software Systems

  7. Formal Analysis and Implementation of the Faust Programming Language
  8. Design and Implementation of Innovative Musical Applications for New Multimedia Technologies
  9. Movement-Based User Interface Language Descriptions: Application to Music Therapy for Physically-Challenged Users

Description :


PhD Thesis 1 « Comparison between abstract numerical domains » :

Compilers and static program analyzers rely on some kind of abstract interpretation to gather information about the values of variables and then to optimize the generated code or to prove some code property. A classical optimization is constant propagation. A useful property is array access legality: a reference to an array should be within its bounds or a buffer overflow may happen.

How accurate should the underlying abstract domain be? The more accurate it is, the slower the abstract interpretation is. What is or are the best trade-offs?

To address this question, a generic abstract domain interface was developped and implemented for a few abstract domains such as polyhedra and octagons and intervals within the APRON project, but more implementations and more abstract domains such as Presburger formulae should be taken into consideration. The numerical robustness of the implementations, already addressed by Duong Nguyen in his PhD, should be checked and improved as exceptions not caught or not handled properly lead to wrong or imprecise results. Furthermore, speed and accuracy improvements for polyhderon operators would be welcome.

Our static program analyzer PIPS was designed before the APRON interface and no experiments can be carried out before it is updated to use it. This done, the PhD candidate is expected to define and carry out an experiment plan aiming at exploring different trade-offs for different needs. He/she is also expected to suggest improvements in exception handling and in execution time, based on experimental results.

Contact : François irigoin (email : francois.irigoin--at--mines-paristech.fr)

PhD Thesis 2 « Compilation for embedded real-time multimedia systems » :

Embedded real-time multimedia systems require up to a teraoperation per second, with a limited power budget. Low frequency multiprocessor systems-on-a-chip (MSoC) are used to provide exactly the required processing power at the lowest possible electrical power and energy.

Such chips contains several processing units and they are extremely difficult to program. At the lowest level, assembly languages are used because no compiler is developped and/or because no compiler could generate efficient code for complex SIMD units. At the middle level, communications and buffer allocations must be dealt with. At the highest level, the chip and its resources must be controlled globally.

To reduce the development time and cost, development environments are used to allocate resources to computational tasks, to allocate buffers and to generate communications between the tasks. However, this is only possible for limited parts of the application. The purpose of this work is to pre-process the C code of an application and to extract all parts and information that can be dealt with by such an environment.

Within the framework of the ASTAR project, the candidate will study a few industrial applications provided by French companies and extend as needed the analyses implemented in PIPS, our program analysis framework. Information about access patterns to data and about computational intensity/loads is of primary importance to characterize with code parts can be successfully processed by hardware accelerators.

This work requires cooperation with several industrial and academic partners.

Contact : François irigoin (email : francois.irigoin--at--mines-paristech.fr )

PhD Thesis 3 « Structured Code and Non-Deterministic Automaton » :

To compile and optimize imperative code, information about the values of variables is key to many program transformations such as automatic parallelization and dead code elimination. To analyze the safety property of non-deterministic automata used to control systems, the relationship between components of the state is key.

Within the APRON project, we looked at the relationship between code analysis and automaton analysis. Code analysis is mostly based on structured code analysis and unstructured code is restructured for analysis. To analyze large pieces of code, function calls are abstracted as transformers. These transformers are used to propagate preconditions, a.k.a. abstract state. Automaton analysis is based on unstructured specifications and abstract state propagation. Abstract commands, i.e. transformers, are not used, but a heuristic technique known as widening is used to force the convergence of the state propagation over control paths.

The code analysis technique can be applied to non-deterministic automata when they are expressed as sequential code with non-deterministic choices. It can provide results with are better or worse than automaton analysis, and it always provide results directly, without a convergence phase.

These encouraging results could be due to the fact that some difficulties are solved implicitly by the manual compilation of automata into sequential code. To answer this question an automatic compilation technique should be developped.

The second problem is the scalability of the approach. What happens with large automata? Is it still possible to reduce them to structured codes? How are theses code sizes growing with the number of states? Are they still efficiently analyzable? Experiments with a large number of academic benchmarks will be carried out to understand the complexity of such analyses.

The frameworks used are currently based on polyhderal abstract domains and abstract interpretation. An extension to Presburger formulae could be studied too. This work is mostly academic, but industrial applications are waiting for such new techniques. Cooperation with other teams in the field will be useful.

Contact : François irigoin (email : francois.irigoin--at--mines-paristech.fr)

PhD Thesis 4 « Generalized expression evaluation optimization » :

A compiler can evaluate expressions in many different ways, depending on the number of registers and of arithmetic and logical units available. More and more registers and units are integrated on a chip. More complex unit can also perform several operations as a unique compound expression, e.g. multiply-add, complex multiply, etc.
To increase potential gains, expressions are transformed by using forward substitution, invariant expression detection, etc.

The goal of this work is to extend previous techniques applied to arithmetic and logical expressions to general expressions, dealing with any kind of objects, starting with images. It is assumed that an image processing accelerator is available together with its API.
The accelerator can be seen as a set of very long pipelines. Each pipeline stage can be programmed to perform one elementary image processing step, but each pipeline can only read one pixel per cycle and write one pixel per cycle.

Another hardware target is a SIMD accelerator which processes intermediate small images, and can combine several processings stages together.

This work will be performed in close cooperation with CMM (Centre de Morphologie Mathematique) who is in charge of the image processing accelerator and of its future version(s). The compiler optimizations will be implemented within the PIPS framework of CRI. Hopefully, this work will be part of a research project with several industrial partners (FREIA).

Contact : Fabien Coelho (email : fabien.coelho--at--ines-paristech.fr)

PhD Thesis 5 « Energy-Centric Optimization of Computer Applications » :

PhD Thesis Subject (CIFRE scholarship from SAGEM)

The domain of embedded computing (portable computers, smartphones, electric cars, other devices...) is significantly constrained by energy usage, in particular when the energy source is limited. Moreover, some applications have real-time constraints that require minimal performance results for such applications to be deemed useful. 

Many factors impact such energy consumption in a given system, such as its hardware technology or whether its run-time frequency can be tuned or some unused components switched off. The very way the application operates, in particular its memory usage pattern or specific assembly-level instructions, has an impact on energy consumption. 

The goal here is to design automatic analysis or run-time instrumentation techniques to assess the energy usage of computer applications running on specific devices (multicore processors, GPGPU, specialized accelerators, SoC, embedded processors such as the ARM Cortex-A9) and then to suggest program transformations and, possibly, dedicated run-time environments that lead to an energy usage decrease for specific classes of applications and hardware platforms. 

Contact :Please send a CV and motivation letter to:
- Fabien Coelho (email : fabien.coelho--at--mines-paristech.fr)
- François Irigoin (email : francois.irigoin--at--mines-paristech.fr).

Location: Research work will be performed at Centre de Recherche en Informatique, Département Mathématiques & Systèmes, MINES ParisTech, and SAGEM, under a CIFRE contract.

PhD Thesis 6 « Energy-Aware Compilation for Communicating Software Systems » :

PhD Thesis Offer, Funding Courtesy of Fondation Télécom Futur et Ruptures 2011 Program

Context : Energy management structures society, and the recent emergence of the GreenIT movement illustrates this concern among providers of network gear, computing systems or storage devices. Even if computing consumes as a whole only a few percents of the world electrical energy, the significant increase of energy prices over the last few years drives this ecological trend, in sync with the economical aspect leading companies to reduce their expenses, for instance via the use of devices exhibiting lower energy footprints.

Hardware approaches to the energy use minimization problem within computing systems are not the only game in town. For instance, since micro-controllers typically require one tenth of the energy needed by radio interfaces, it might make sense to look for efficient algorithms and techniques that would trade a few communication operations over the network for local processing operations (filtering, compression, ciphering, aggregation...). Such an approach would clearly provide a net gain and present new optimization opportunities.

Embedded communicating systems, which need to rely on batteries as sole energy source, are definitely sensitive devices with regard to electrical consumption, and would benefit from approaches that would extend their autonomy while shrinking their weight. This of course includes smartphones and interactive tactile tablets.

Summary : In the area of human interactions, the widespread emergence of ubiquitous and mobile computing devices (smartphones, netbooks, RFID) mandates to build hardware and software systems that take into account energy requirements from the very start of the design process, so as to offer autonomy and ease-of-use while preserving the environment.

Building upon an energy-centered model of such computing systems and of their associated costs (instruction, memory, data transfer), this thesis should strive to design compilation algorithms that minimize their energy consumption while preserving adequate performance in terms of computing time and memory usage. One might want to integrate these algorithms within the PIPS compilation platform developed at CRI in order to perform full-scale assessments of the impact of the suggested solutions on meaningful communicating application benchmarks.

Conditions : The Phd student will be on a French CDD work contract, registered at the Ecole doctorale EDITE and under the supervision of academics from both institutions:
- Gérard MEMMI, TELECOM ParisTech
- Fabien COELHO, MINES ParisTech
- Pierre JOUVELOT, MINES ParisTech

Contact : Fabien Coelho (fabien.coelho--at--mines-paristech.fr)
Send your resume, your motivation letter, your grades of the Master-level courses attended to and the names of two reference.

Location:Work is located in Paris, with periodical stays in Fontainebleau.

PhD Thesis 7 « Formal Analysis and Implementation of the Faust Programming Language » : annonce5
The Faust programming language (Functional Audio Stream), designed at Grame, a Centre national de création musicale located in Lyon, focuses on the definition of synchronous digital signal processes; Faust programming paradigm is strictly functional. This language, used all over the world, can be applied in various domains, from advanced audio filter design to innovative contemporary music installations.

A first formal analysis of the core of Faust has been developed at CRI, the MINES ParisTech Centre de recherche en informatique, CRI, as part of the ANR Astrée project. This analysis has led to (1) the definition of the precise semantics of the language core instruction set and (2) a first attempt at characterizing the language key semantic properties (typing, synchronicity).

The goal of the proposed PhD research is to:
  • extend the existing formal definitions of the language core to the whole language;
  • extend and generalize the existing theorems regarding Faust mathematical properties;
  • implement within the Faust compiler the analyses developed above in order to provide even more performant implementations (sequential, parallel) of Faust programs.

Requirements:
A good understanding of the theoretical and mathematical principles of programming languages, in particular functional ones, and some notions of advanced compilation techniques are required. An interest for music and real-time sound processing algorithms would be a plus. Some practice of object-oriented programming (C++ or Java) would be an asset.

Contact : Pierre Jouvelot (email : pierre.jouvelot--at--mines-paristech.fr)
Send your CV, a motivation letter, your course grades and ranking for the Master program you attended to and one recommendation letter to the postal address of CRI, or via email.

PhD Thesis 8 « Design and Implementation of Innovative Musical Applications for New Multimedia Technologies » : annonce5

The MINES ParisTech Centre de recherche en informatique, CRI, has been working over the last few years on designing and implementing innovative applications for new multimedia technologies, in particular  related to music. These research activities have always been initiated in close collaboration with experts from the application domains, and are intended to lead to the development and validation of prototypes. Up to now, these systems have targeted education (in music theory), instrument practice (for jazz improvisation) and music therapy (for ADHD and Alzheimer patients).
These developments linking music and technological applications are based on both theoretical and technical foundations. Conceptual notions used in the past included motivation theories or metapsychological analytical frameworks; computing-related technologies used in these applications were online software tools (such as Java applets) or videogame artifacts (Wiimotes, game engines). CRI is one of the first French research institutions to have worked on Serious Games, videogames that focus not only on fun but also useful outcomes.

The goal of the proposed PhD research is to:

  • extend the exploration of application domains already addressed in previous projects, either with existing partners or new ones;
  • look at new promising application domains that could profit from the use of new technologies for music or sound.
The research project should ideally strive to address both theoretical (new domain or in-depth exploration of previous ones, formal frameworks providing insight on specific issues and linking music and application,…) and technical (design and implementation of software environments, validation tests,…).

Requirements :
The PhD candidate must have a good understanding of multimedia computer science technologies, and a strong interest for music applications or sound processing. Practical knowledge of software development is required, as is a curious mind challenged by issues linking multiple domains. An appetite for videogames would be an asset.

Contact : Pierre Jouvelot (email : pierre.jouvelot--at--mines-paristech.fr)
Send your CV, a motivation letter, your course grades and ranking for the Master program you attended to and one recommendation letter to the postal address of CRI, or via email.

PhD Thesis 9 « Movement-Based User Interface Language Descriptions: Application to Music Therapy for Physically-Challenged Users » : annonce5

The MINES ParisTech Centre de recherche en informatique, CRI, has been working the last 15 years on innovative technological applications related to music (education, Serious Games, health…). Music Therapy has been shown to be a rich subject matter for such an approach. MAWii, developed at CRI by Samuel Benveniste, uses Nintendo-based Wii technology to provide to music therapists new interaction modalities for music synthesis. Tested in collaboration with researchers from the Institut de psychologie at Paris Descartes University, MAWii is used in hospitals with teenagers experiencing ADHD-like, behavior disorders.

Wiimotes, driven via MAWii, present wide configuration spaces and adaptation opportunities for patient-specific traits. The goal of this thesis is to build upon such capabilities to offer even larger adaptation features for music therapy targeting in particular patients with limited motor abilities (aging persons, handicapped patients, road-injured adults….). Specifically, one might want to look at:

  • defining new interaction modalities specific to particular pathologies, and even specific to each patient;
  • designing a visual and/or textual DSL language for movement specification, for instance using existing domain-specific languages (dance, aerobatics, language for deaf/dumb persons…), and providing its specific syntax and semantics;
  • studying automatic "parsing" techniques from actual 3D or 2D movements to language constructs, in order to provide an effective learning mechanism of various simple choreographies and link them to music productions;
  • integrating this language into MAWii user environment (movement editor, sound configuration, user profile management…);
  • asserting with practitioners that this system is both intuitive and fast enough to easily adapt MAWii to patient-specific traits;
  • assessing possible therapeutic gains brought by this approach compared to the original version of MAWii.

Requirements :
The PhD candidate must have a good understanding of multimedia computer science technologies, and a strong interest for music applications, sound processing or health issues. Practical knowledge of software development is required, as is a curious mind challenged by issues linking multiple domains. An appetite for videogames would be an asset.

Contact : Pierre Jouvelot (email : pierre.jouvelot--at--mines-paristech.fr)
Send your CV, a motivation letter, your course grades and ranking for the Master program you attended to and one recommendation letter to the postal address of CRI, or via email.