Student Internship in Energy-efficient training algorithms

작업 내용

Student Internship in Energy-efficient training algorithms

Our newly founded Fundamental Theory Lab part of Belgium Research Center (Leuven, Belgium) will play a leading role in researching new computational technologies for next generation networks. Research activities range from mathematical and computational optimization up to their application to sequential decision processes, online optimization for scheduling, predictive network modeling, automated pattern detection/extraction, automated data labeling and as well as fast converging solvers.

Description: this internship is situated in the domain of computational methods for neural networks; the training of neural networks by backpropagation does (often) not account for energy consumption considerations and the resulting constraints on the properties and usage of the learned model. However, these constraints are of fundamental importance in enabling autonomy of devices executing the learned model(s).

Goal: the main goal of this internship is to formulate, design and evaluate optimization methods and algorithms that minimize the neural network loss/error function (associated to simple regression/ classification tasks) under energy constraints.

Tasks: the tasks assigned to the intern -under the supervision of a Post-Doc level promotor- include but are not limited to:

• Development and evaluation of energy consumption models
• Design and numerically evaluation of first order iterative optimization algorithms for solving of nonlinearly constrained problems for the training of multi-layer neural networks
• Comparison against non-constrained training algorithms (accuracy vs. complexity vs. energy consumption)

Duration:

• Short duration (from 3 to 6 months): suited for MSc curriculum course or other voluntary activity conducted in academic context.
• Long duration (up to 12 months): suited for MSc thesis internship
• Note well: following Belgian Law, being a student internship (no working contract) only direct costs may be reimbursed.

Candidate profile and Position requirements:

• At least : last year of MSc degree in Applied mathematics, Mathematical engineering, Numerical Mathematics but also Theoretical computer science, Computer science engineering.
• Note well: candidate must be enrolled or already obtained their University degree from an academic institution of one of the EU country.
• Experience in either
• Continuous optimization and nonlinear programming together with good knowledge of iterative methods for the solving of constrained nonconvex optimization problems
• Stochastic optimization and related methods including stochastic approximation, sample average approximation, stochastic splitting methods etc.
• Experience in energy-/power consumption models for computing systems is considered as a plus.
• Good knowledge of programming language (C/C++, FORTRAN), functional programming language (LISP, Julia, etc.) or numerical programming languages (such as Octave) is considered as a plus.
• Excellent Communication skills - Presentation, oral and writing skills.

Application requirements

• Certified copy of the MSc diploma/certificate shall be included in annex of the CV.
• The CV shall indicate the detailed coordinates of the current or eventually the last academic institution and department of the candidate.
• The CV shall include a detailed list of publications/achievements in relation to the job description.
• Note well: the candidate must follow or have obtained his/her MSc degree from an academic institution of one of the EU countries.

Starting date:

September, 1st 2022 (earliest) – December, 1st 2022 (latest)