Application domains

Multi-Linear Algebra and Solvers
Training and Inference for DNNs

Research axes

Use of runtime systems
Design of compression techniques
Energy minimization

Tools and Optimization for high Performance Applications and Learning

We propose to structure our research around two main application fields: linear multi-dimensional algebra and solvers on the one hand, and training in particular of deep learning networks on the other hand. In these two domains, our contributions will be organized around three main research axes: the use of task based runtime systems (to provide robust solutions and to increase the portability in the context of heterogeneous large scale platforms), the use of compression (to limit memory footprint and data transfers) and the minimization of energy consumption and carbon impact (using an approach of rewriting algorithms and placement strategies to limit data movements).

Task-based Linear Algebra

Express high level algorithms in an elegant way while ensuring high performance

Multi-Linear Algebra and Tensor Computations

Parallelization of tensor decomposition and tensor contraction algorithms

Energy minimization in Linear Solvers

Find better trade-offs between energy consumption and high performance

Task-based Approaches for Deep Learning

Express high level algorithms in an elegant way while ensuring high performance

Tensor Compression for Inference

Ue tensor compression techniques and quantization to perform inference with little memory or low latency

Energy-Efficient Training

Reduce the resource consumption of deep network training: time, energy, memory