PhD studentship- The project, “Security of Foundation Models for Code,” overseen by Dr. Sergio Maffeis from Imperial College and in collaboration with IBM Research, invites applications for a fully funded PhD position. The opportunity includes coverage of tuition fees, a travel and expenses allowance, and an enhanced stipend starting at £23k (tax-free, subject to annual increases) for up to 4 years.
This research initiative aims to enhance the robustness and security of Large Language Models (LLMs), initially focusing on solutions for Natural Language Processing (NLP) before delving into the complexities of source code generation. While LLMs excel in generating computer programs and analyzing vulnerabilities, the structured nature of code poses unique challenges compared to natural language.
PhD studentship Key Objectives:
- Trained Model Verification: Research poisoning attacks and defenses for LLMs, crucial as users often download models from untrusted repositories. Develop defenses, potentially integrated during fine-tuning, to mitigate such threats.
- Deployment: Investigate prompt injection attacks on deployed LLMs, where malicious prompts can lead to harmful content or vulnerable code generation. Evaluate the cost of such attacks and devise strategies to bolster LLMs against them.
The project will be conducted within the esteemed Department of Computing at Imperial College London, renowned globally for its excellence in computer science research. Dr. Maffeis leads the Security and Machine Learning lab, focusing on enhancing machine learning systems’ robustness and addressing security challenges using machine learning techniques.
Furthermore, the PhD is co-funded by IBM Research, a pioneering organization in computer science and artificial intelligence research. The selected student will have the opportunity to spend at least three months at IBM Research, with expenses covered, and access their cutting-edge computing infrastructure.
Applicants should hold a First Class or Distinction Masters degree in a relevant scientific or technical discipline, such as computer science or mathematics, and demonstrate proficiency in English. Overseas applicants will be considered in exceptional cases, with the requirement to cover the difference between Home and Overseas tuition fees.
To apply, follow the guidelines at the provided link, specifying “Security of Foundation Models for Code” as the proposed research topic and “Sergio Maffeis” as the proposed research supervisor. Early applications are encouraged, and informal inquiries can be directed to sergio.maffeis@imperial.ac.uk.
The position will be based at the South Kensington campus in central London. Applicants are encouraged to visit the department’s website for general information on becoming a PhD student. The department is committed to equality, diversity, and inclusivity, and encourages applicants from underrepresented backgrounds to apply.
