Join Our Team

PhD Position — Reliable generative AI systems for high-stake decision making

Start: Autumn 2026   |   Application deadline: 16th June 2026. (11:59 AM Italian time)

We invite applications for a funded Ph.D. position in “Reliable generative AI systems for high-stake decision making” at the University of Bergamo, Italy, with a particular emphasis on “uncertainty quantification for small language models”. The project focuses on developing reliable and computationally efficient methods for confidence estimation, calibration, and uncertainty-aware decision-making in compact generative models deployed in edge and agentic AI settings.

Project Description:

The emergence of small language models (SLMs) [1] is enabling a new generation of efficient, deployable, and autonomous AI systems operating on edge devices and within agentic AI frameworks [2]. Yet, these models frequently exhibit unreliable confidence estimates, hallucinations, and poor calibration [3]. Current uncertainty quantification (UQ) approaches for large language models (LLMs) are often computationally prohibitive and insufficiently adapted to compact generative architectures.

This Ph.D. project will investigate uncertainty quantification for white-box SLMs (1 - 7B parameter models), with the goal of developing efficient, theoretically grounded, and practically deployable methods for estimating and utilizing uncertainty in generative AI systems. Research topics may include calibration under quantization, retrieval-augmented generation, epistemic uncertainty estimation, uncertainty-aware reasoning and routing, selective generation, and robust evaluation methodologies for generative models. Applications include edge AI, tool-using autonomous agents, and low-resource deployment scenarios where reliable confidence estimation is critical for safe and trustworthy operation (e.g., time-series forecasting, medical report generation).

The project lies at the intersection of trustworthy machine learning, natural language processing, Bayesian methods, and efficient generative AI systems, and aims to advance both the theoretical understanding and practical reliability of SLMs.

Bibliography:

1. Subramanian, Shreyas, et al. "Small language models (slms) can still pack a punch: A survey." arXiv preprint arXiv:2501.05465 (2025).
2. Belcak, Peter, et al. "Small language models are the future of agentic ai." arXiv preprint arXiv:2506.02153 (2025).
3. Chuang, Yu-Neng, et al. "Confident or seek stronger: Exploring uncertainty-based on-device llm routing from benchmarking to generalization." arXiv preprint arXiv:2502.04428 (2025).

Eligibility criteria:

We are looking for motivated candidates who have:

• Completed a Masters (or MSc) degree from an accredited educational institution
• A strong background in machine learning, statistics, mathematics, physics or a related field
• Experience with deep learning frameworks (e.g., PyTorch) and programming languages (e.g., Python, C++)
• Strong mathematical and analytical skills
• Familiarity with Transformer architectures and modern NLP methods
• Interest in trustworthy, reliable, and efficient AI systems

Desirable qualifications:

• Experience with uncertainty estimation or Bayesian machine learning
• Research experience in generative AI or large language models
• Prior publication record in peer-reviewed venues (e.g., workshops, conferences, or journals in machine learning, NLP, or AI), research projects, or open-source contributions

Research Environment:

The student will be enrolled in the University of Bergamo (UniBG), Italy, in the Manufacturing Ecosystems Enabled by Intelligent Technologies (MEET) PhD Program, and will conduct the research in joint collaboration with the research center Consiglio Nazionale delle Ricerche (CNR), Milan. The project will provide opportunities for interdisciplinary collaboration across trustworthy AI, generative modeling, uncertainty quantification, with applications to edge intelligence and autonomous AI agents. The student will have access to computational resources from both UniBG and CNR.

The PhD research will be supervised by:

• Dr. Alessandro Brusaferri (Staff researcher at CNR-STIIMA, Milan)
• Dr. Subhankar Roy (Assistant Professor at UniBG)
• Prof. Angelo Gargantini (Professor at UniBG)

The student will be encouraged to publish in leading international venues such as NeurIPS, ICLR, ICML, ACL, EMNLP, and UAI.

While the project focuses broadly on uncertainty quantification for small language models, the exact research direction can be tailored based on the student’s interests and background.

Application:

Applications are currently open. Please refer to the official webpage (link) of University of Bergamo to know more about the PhD program, working conditions, stipends, exact access requirements, and how to apply. This particular position has been listed on page 12 of this (in ENG) or this (in ITA) document.

Deadline: 11:59 a.m. (Italian time) on 16th June, 2026.

Contact:

For informal inquiries or expressions of interest, please contact any of the following supervisors. Before sending an email please read the guidelines on the "Contact" page.

• Dr. Alessandro Brusaferri (email: Alessandro.Brusaferri@stiima.cnr.it)
• Dr. Subhankar Roy (email: subhankar.roy@unibg.it)
• Prof. Angelo Gargantini (email: angelo.gargantini@unibg.it)