Stage : Sécurité pour l'Apprentissage Machine Distribué F/H

SAP a vu le jour en 1972, sur l'initiative d'une équipe de cinq collaborateurs désireux de créer quelque chose de nouveau. Ensemble, ils ont révolutionné le secteur des logiciels d'entreprise et réinventé les méthodes de gestion. Aujourd'hui, SAP se positionne en leader sur le marché des logiciels d'entreprise, mais nous restons fidèles à nos racines. C'est pourquoi nous concevons des solutions qui nourrissent l'innovation, favorisent l'égalité et multiplient les opportunités pour nos employés et nos clients, au-delà des frontières et des cultures.
SAP apprécie l'esprit d'entreprise, encourageant la créativité et établant des relations durables avec nos employés. Notre personnel réfléchit la diversité et l'intégration: nous sommes conscients qu'il s'agit d'une richesse source de compétitivité à long terme et d'opportunités pour tous. Nous pensons qu'ensemble, nous pouvons transformer les secteurs d'activité, faire croître l'économie, faire évoluer les sociétés et préserver l'environnement, car seules les meilleures entreprises sont à même d'améliorer la société et le quotidien des individus.

OBJECTIVES
The objective of this internship is to set up a demonstrator for the security of distributed machine learning.

EXPECTATIONS AND TASKS
As part of this internship, the student will contribute to:

Study the state of the art in security for distributed machine learning;

Design of a new approach for the protection of AI-based software data and the safeguarding of intellectual property;

Implementation of a proof of concept demonstrating the feasibility of such an approach on an industrial use case.

We expect that 60% of the time will be spent on development and 40% on research activities.

MISSIONS AND RESPONSIBILITIES
This internship is based in the SAP Labs France research laboratory, in Sophia-Antipolis. The work will be carried out as part of the "Security & Trust" research program, and focus on the secure integration of the Internet of Things with SAP HANA applications. The Internet of Things (IoT) is expected to reach 50 billion connected devices and a value of $ 14.4 trillion at stake through 2020. SAP is harnessing this trend and is focusing its IoT development on the IoT service of the SAP HANA cloud platform. 

Until now, background deployments (on-prem and cloud) have been viewed as the single source of truth and single point of access when it comes to enterprise systems (ES). Nevertheless, a paradigm shift has recently been observed, by the deployment of ES assets towards the cutting edge sectors of the landscapes; by distributing data, decentralizing applications, de-abstracting technology, and integrating advanced components seamlessly into central back-end systems. By capitalizing on recent advances in high performance computing as well as the increasing amounts of publicly available labeled data, Deep Neural Networks (DNN), as an implementation of AI, 

AI capabilities deployed in a distributed manner will support the transition mentioned above. As Deloitte reports, "... companies are integrating artificial intelligence, especially machine learning, into their Internet of Things applications and seeing their capabilities grow, including improving operational efficiency and avoiding unplanned downtime "[Schatsky et al., 2017].

The deployment of data processing capacities in distributed enterprise systems poses several security problems related to the protection of input and output data [Parliament and Council, 2016] as well as to software resources. In the specific context of distributed intelligence, DNN-based / enhanced software will represent key investments in infrastructure, skills and governance, as well as in data and talent acquisition. The software industry therefore has a direct need to safeguard these strategic investments by enforcing the protection of this new form of intellectual property. In addition, in the wake of data protection regulations (PD) such as the European data protection regulation (EU-GDPR) [Parliament and Council, 2016], independent software vendors (ISVs) have the non-transferable obligation to comply with it. Therefore, ISVs aim to protect both the data and intellectual property of their AI-based software, deployed on potentially insecure hardware and peripheral platforms [Goodfellow, 2018].

The absence of solutions for the protection of intellectual property exposes the owners of NNs trained to reverse engineering on their DL models [Tramèr et al., 2016]. As reported in [Augasta and Kathirvalavakumar, 2012] [Floares, 2008], attackers can steal trained NN models. In this new coding paradigm, where design patterns are applied in known and legacy implementations, the issue of intellectual property is at stake. The issue is not so much how to protect the DNN architecture (since most architectures are based on well-known research), but rather how to protect the trained DNN model.

Reference
Automatic smart contract generation using controlled natural language and template1
Secure Smart Contract Generation based on Petri Nets2
Designing Secure Ethereum Smart Contracts: a Finite State Machine Based Approach3
Auto-Generation of Smart Contracts from Domain-Specific Ontologies and Semantic Rules4

TRAINING / SKILLS AND ABILITIES

University level: Last year of MSc in computer science or beyond

C, Python, Solidity

Experience on smart contracts, blockchain, machine learning, cybersecurity

Fluency in English (working language)

Ability to organize meetings and contact people

Good oral and written communication skills

Ability to write documents in English, ability to summarize

PROFESSIONAL EXPERIENCE

No professional experience required