[bull-ia] Post-Doctoral Position at Université Nice-Sophia-Antipolis, France on Multi-objective Large-scale Evolutionary Tuning of an ISP

Post-doctoral position in Computer Science – Université Nice Sophia-Antipolis, France on Multi-objective Large-scale Evolutionary Tuning of an Image Signal Processor (ISP)

Keywords : Operational research (OR), evolutionary computation (EC), multi- and many-objective optimization, large-scale optimization, image processing

Employer Profile : Université Nice Sophia-Antipolis, public technical and scientific institution, belongs to Université Côte d’Azur (UCA) which is a Community of Universities and Higher Education Institutions (ComUE) created in 2015 with a research mission and consists of 13 members and more than 30,000 students. It is situated in Alpes-Maritimes region of France, the French Riviera.

Job Description : Image processing can be thought of as a transformation of an image into a new enhanced image. The transformation is ensured by an image signal processor (ISP) that is often coupled with digital cameras and embedded in small devices like smartphones, tablets… The ISP can apply many digital tasks like noise reduction, choosing the correct white balance or color correction for instance to enhance the image quality of pictures. However, all previous tasks are based on methods that must be parametrized. When enhancing only one raw image, we are facing many parameters (up to 10000) and quality metrics, and each of them could be float-valued. The problem we are facing is challenging because of its size and its complexity: quantity of parameters to optimize is not common, and several potentially time-consuming criteria must be optimized simultaneously.

Moreover, each parameter could be constrained. The problem to be solved combines several fundamental issues like very large-scale optimization, multi- or many- objective optimization, constrained optimization, expensive and/or noisy optimization. As a consequence, evaluating each combination of parameters is impracticable in time even in parallel conditions. In addition, a gradient-based search is also not usable because problem to optimize is a set of procedures that are not differentiable.

The main goal of this work is to develop and program a proof-of-concept able to find good parametrizations of an ISP considering several kinds of images as landscapes, pictures by night or macro-photography to name a few. For all exposed reasons, evolutionary algorithms (EA), which are generic population-based metaheuristic optimization algorithms, are envisaged for their ability to deal with previous identified issues. Such algorithms are also capable of optimizing many objectives, subjective and noisy criteria when considering interactions with users, combining local and global search…

In this context, the goal of the postdoctoral position is to address the following challenges:
– Program a proof-of-concept able to couple an ISP with an optimization algorithm able to find good parametrizations,
– Make a brief survey of the literature on the treated problem,
– Propose a modified and adapted EA able to deal with several issues presented before.


Bartz‐Beielstein Thomas, Branke Jürgen, Mehnen Jörn, Mersmann Olaf. Evolutionary Algorithms. WIREs Data Mining and Knowledge Discovery 2014, 4: 178-195.
R. Cheng, Y. Jin, M. Olhofer and B. sendhoff, “Test Problems for Large-Scale Multiobjective and Many-Objective Optimization,” in IEEE Transactions on Cybernetics, vol. 47, no. 12, pp. 4108-4121, Dec. 2017.
X. Zhang, Y. Tian, R. Cheng and Y. Jin, “A Decision Variable Clustering-Based Evolutionary Algorithm for Large-Scale Many-Objective Optimization,” in IEEE Transactions on Evolutionary Computation, vol. 22, no. 1, pp. 97-112, Feb. 2018.

Candidate Profile

The candidate must hold a Ph.D. thesis in Computer Science (obtained less than five years ago), with a specialization on the evolutionary computation field. Knowledge on image processing field might also help. Finally, he must have good English skills in writing and communication. French is not mandatory. Programming skills in Java, Python or Matlab is necessary.

Specific Conditions

Starting date: November 2018
Duration: 12 months
Salary: around 2400 euros/month (gross salary) according experience
Location: I3S laboratory (https://www.i3s.unice.fr/en) and SPARKS team, SophiaTech Campus, Sophia Antipolis, France.


Send curriculum vitae with a motivation letter to Dr. Denis Pallez (denis.pallez@unice.fr)
Deadline for applications: September, 20th 2018

            (' O-O ')
Denis PALLEZ           denis.pallez@unice.fr
Maitre de Conférences / Associate Professor
Université Côte d’Azur (IUT), CNRS, I3S
(+33) 4 89 15 42 85