Research

Current research

I am currently a Ph.D. student at Scuola Superiore Sant’Anna, within the TeCIP Institute, working in the area of Mobile and Cyber-Physical Systems.

My current research focuses on efficient AI deployment at the edge, with particular interest in the interaction between data, model compression, and target hardware platforms.

A central direction of my research is the combination of data-aware quantization and platform-aware design-space exploration (DSE). The goal is to understand how information contained in the data can be exploited to make better decisions about the representation and deployment of neural networks on resource-constrained systems.

In particular, I am interested in exploring the relationship between:

• data-dependent signals extracted from neural networks and their inputs;
• mixed-precision quantization and other model-compression techniques;
• hardware and platform characteristics;
• design-space exploration for efficient deployment;
• and system-level objectives such as accuracy, latency, energy consumption, memory usage, and robustness.

Research interests

My broader research interests include:

• Efficient and Edge AI
• Neural Network Quantization
• Data-aware Model Compression
• Design-Space Exploration
• Mobile and Cyber-Physical Systems
• Approximate Nearest Neighbour Search
• Information Retrieval
• AI Systems and Software Engineering
• Optimization and Operations Research

I am particularly interested in problems where machine learning, algorithms, optimization, and systems engineering meet.

Research approach

My research interests are driven by a broader question:

How can we make intelligent systems more efficient by taking into account both the information contained in the data and the characteristics of the platform on which the system will run?

Rather than treating model compression and hardware optimization as independent problems, I am interested in approaches that consider them jointly.

This perspective motivates a workflow ranging from data-aware, platform-independent analysis to platform-aware design-space exploration, followed by refinement on the target system.

The ultimate objective is not simply to reduce model size, but to identify configurations that provide the best trade-offs between performance, resource consumption, and model quality for a given deployment scenario.

Previous research

Before starting my Ph.D., I worked on several problems at the intersection of information retrieval, approximate nearest neighbour search, and machine learning systems.

My Master’s thesis explored Information Retrieval using approximate k-nearest-neighbour search, with a focus on the use of Faiss for efficient similarity search.

I subsequently worked on Variance-Aware Quantization (VAQ) and its implementation in Rust, using the kANNolo approximate nearest-neighbour library and CBLAS-based numerical routines.

These experiences contributed to my broader interest in the relationship between algorithms, data structures, numerical computation, and efficient AI systems.

Beyond a single research area

Although my current work is focused on efficient Edge AI, I tend to approach research problems from a broader computer-science perspective.

I am particularly interested in theoretical ideas that can lead to practical systems, and in understanding how algorithmic choices affect real-world computational constraints.

This is also why I maintain an interest in optimization and operations research, especially when these methods can be applied to complex engineering and AI systems.