Julien Mairal

Host institution: INRIA

Laboratory: LJK

Call for projects: Start-up (PE6)

Project name: SOLARIS – Large Scale Learning with Deep Core Machines

Amount: €1.49 million

Description :

Machine learning has become a key part of scientific fields that produce massive amounts of data and urgently need scalable tools to automatically make sense of it.

Unfortunately, classical statistical modeling has often become impractical due to recent changes in the amount of data to be processed, as well as the high complexity and large size of models capable of taking advantage of big data.

The promise of SOLARIS is to invent a new generation of machine learning models that meets today’s large-scale data analysis needs: high scalability, ability to process high-dimensional models, rapid learning, ease of use and adaptability to various data. structures.

To achieve the expected breakthroughs, our angle of attack consists of new optimization techniques to solve large-scale problems and a new learning paradigm called deep-core machine. This paradigm marries two schools of thought that have so far been missed as having little overlap: kernel methods and deep learning.

The former is associated with a well-understanding theory and methodology but lacks scalability, while the latter has achieved significant success on large-scale prediction problems, especially in computer vision.

Deep-core machines will lead to theoretical and practical breakthroughs in machine learning and related fields.

For example, convolutional neural networks were invented more than two decades ago and are today at the cutting edge of technology for image classification.

Yet the theoretical underpinnings and principle-based methodology of these deep networks are elusive.

The project addressed these fundamental questions and its results made deep networks simpler to define, easier to use and faster to train.

It will also take advantage of kernels’ ability to model invariance and work with a large class of rendered data such as graphs and sequences, leading to a wide range of applications with potentially revolutionary advances in various scientific fields.