Alessio Borgi - About
👋 About Me
🚀 Open to Collaborate!
I'm eager to work with anyone who has great ideas, wants to learn and share their experience. Don't hesitate to reach out if you'd like to collaborate on a project, research, paper idea, or a moonshot you're cooking up.
🗺️ Places I've Been
📄 Latest Publications
Heterogeneous Sheaf Neural Networks
Published in NeurReps Workshop @ NeurIPS 2026 - Symmetry and Geometry in Neural Representations, (Sydney, Australia 🇦🇺), 2026
Accepted at the NeurReps Workshop @ NeurIPS 2026. HetSheaf is a cellular-sheaf framework for heterogeneous graphs that encodes node and edge types through type-aware local feature spaces and learned restriction maps, without specialised architectural components. The companion SheafPool readout is invariant to basis changes and enables graph-level prediction. Gains of up to +2 pp on the Heterogeneous Graph Benchmark with up to 10× fewer parameters.
LocaLeJEPA: Maximum Entropy Representations under Geometric Priors
Published in arXiv preprint arXiv:2610.28458, 2026
LocaLeJEPA replaces a global Gaussian constraint with centred, local uniform-ball matching for self-supervised learning. Its entropy-inspired regulariser leaves neighbourhood locations free and improves frozen-feature recognition and transfer across Euclidean, hyperbolic and equivariant encoders.
Oscillatory Neural Dynamics over Sheaves
Published in arXiv preprint arXiv:2610.10018, 2026
ONDA combines learned matrix-valued sheaf transport with second-order oscillatory dynamics for long-range graph learning. A stalk-wise sensitivity analysis characterises persistent influence in the conservative setting, while experiments demonstrate strong performance on long-range propagation, severe graph bottlenecks, and graph transfer, alongside competitive results on heterophilic benchmarks.
Let the Heads Talk: Beyond Diagonal Graph Attention
Published in arXiv preprint arXiv:2610.01494, 2026
A bridge between attention and sheaf neural networks: Top-A shows that multi-head attention is diagonal matrix-valued transport on its head space and adds edge-conditioned off-diagonal routes, letting each interaction move information across heads before aggregation. It recovers vanilla attention exactly when routing vanishes and improves relational, heterogeneous and algorithmic reasoning.
Program Graph Learning for Software Vulnerability Analysis: A Survey
Published in Transactions on Graph Intelligence and Network Applications (TGINA), Scilight, 2026, 2026
A journal survey of graph-driven vulnerability intelligence: how software vulnerability detection, classification, localization and repair become graph learning problems over program entities (ASTs, CFGs, DFGs, PDGs, CPGs). The survey covers graph-based, multimodal and LLM-assisted methods, the dataset/tool/framework ecosystem, and five open challenges — from label noise to the new attack surfaces of AI-generated code and agents.
🧠 Latest ML Blog Posts
💼 Latest LinkedIn Post
🐙 GitHub Activity
A clean PyTorch / PyG implementation library of Sheaf Neural Networks as Message Passing comprising all variants and a benchmark suite with 14+ Datasets.
Simulation of a Cleaning Robot with the capability of performing SLAM of the environment the robot is navigating, Planning Trajectories to calculate optimal paths considering static and dynamic obstacles, and Dynamic Obstacle Avoidance to detect and navigate around obstacles.
RealTime-VLM brings real-time VLM inference to the browser. It continuously captures webcam frames, sends image+text to an OpenAI-compatible API, and displays responses with sub-second latency. Works with local or hosted VLMs.
Improving hyperthermia treatment by controlling temperature distribution in both 1D and 2D domains and thermal energy applied to cutaneous and subcutaneous tissues, through Bio-Heat equation with Physics-Informed Neural Networks (PINNs).
Autonomous robot for lunar navigation and object interaction, developed during TESP '25 at the Space Robotics Lab (Tohoku University). Features custom robot design, Dijkstra-based path planning, object detection with vision, and gripper control.
A comprehensive MATLAB library for solving a wide range of robotics tasks, providing tools and functions for robotic simulations, control systems, kinematics, and path planning.
[CVPR 2025] Z-SASLM is a zero-shot framework for multi-style image synthesis leveraging Spherical Linear Interpolation (SLI) to achieve smooth, coherent blending—without any fine-tuning.
Explainable Graph Neural Networks (XGNNs) and Reinforcement Learning (RL) techniques for graph generation and optimization tasks. The project aims to explore and validate the use of explainable AI for creating interpretable graph structures.
Adaptive Vision Transformer for efficient image classification, implementing dynamic token sparsification to reduce computational costs while maintaining accuracy.
Novel framework for Zero-Shot Style Alignment in Text-to-Image generation, incorporating Multi-Modal Context-Awareness and Multi-Reference Style Alignment, using minimal attention sharing, ensuring consistent style transfer without fine-tuning.
LocaLeJEPA: JEPA self-supervised learning that matches the embedding distribution locally, in k-means neighborhoods, instead of globally, giving stronger features across Euclidean, hyperbolic and equivariant encoders.
A deep learning-based application for skin disease detection and classification.