New attacker models and emerging threats for AI

Profile:

The ideal candidate must be expert about the security of machine learning and artificial intelligence based systems. In particular, the candidate must possess a good knowledge of adversarial AI attacks such as backdoors, model and data poisoning and inference. 
Furthermore, the candidate must have basic skills on at least one mainstream programming language between Python, Java or C/C++.
Finally, the candidate must have some expertise on some major AI libraries such as TensorFlow, PyTorch and scikit-learn.

Activity:

The candidate will carry out a research on novel black-box attacks agains ML classifiers and, in general, artificial intelligence systems. More precisely, the activity will require to investigate the state of the art of adversarial techniques allowing an attacker to infer information from a ML model without direct access to its internal data (black-box). Then, based on these techniques, the candidate will design and develop an algorithm to induce backdoor tasks by taking advantage of pre-installed biases that naturally exist in models.

Formal requirements:

• master’s degree (Laurea Magistrale) in Computer Science, Computer Engineering, Mathematics or Law or equivalent qualification obtained abroad;

   

Specific requirements:

• Good knowledge of written and spoken English;
• Research experiences in cybersecurity, vulnerability detection and penetration testing;
• Experience in computer programming (Python, Java, C/C++), use/development of software libraries for scientific computing and AI.

Duration:

12 months

Gross amount:

€ 16.243,00

Project:

The scholarship referred to in this announcement is funded by the project "SERICS - SEcurity and RIghts in the CyberSpace PE0000014”, codice P0275, CUP: D67G22000340001, finanziato con fondi dell’Unione Europea - Next Generation EU nell’ambito del Piano Nazionale di Ripresa e Resilienza (PNRR), Missione 4 Componente 2, Investimento 1.3

Application:

Applications must be submit through PICA website only using the link at the bottom of this page.
Please note that for the submission of applications on PICA website, login via SPID IS NOT AVAILABLE at the moment - Please register using the alternative methods.

Si segnala che per la presentazione delle candidature sul sito PICA al momento NON E' DISPONIBILE il login tramite SPID - Si prega di effettuare la registrazione secondo le modalità alternative.