Selected work

Representative Papers

The Drift Burst Hypothesis

Sep 27, 2016Journal of Econometrics

This study investigates the existence, market prevalence, and underlying mechanisms of transient, localized “drift bursts” in financial asset prices. To this end, we incorporate drift bursts into a continuous-time Itô semimartingale framework and develop a theoretical model under no-arbitrage conditions, alongside a nonparametric test statistic designed to reliably detect such events from high-frequency data contaminated by noise. Our work is the first to formally model drift bursts as a regular feature of financial markets, uncovering their intrinsic links to liquidity shocks and price reversals. Empirical analysis reveals that drift bursts are pervasive across equity, bond, foreign exchange, and commodity markets, occurring on average once per week; notably, negative bursts accompanied by high trading volume are more likely to trigger significant price reversals.

64 citations10 influentialRead paper

On Constrained Intersection Representations of Graphs and Digraphs

Apr 25, 2025International Symposium on Algorithms and Computation

This paper investigates the Optimal Directed Intersection Representation problem for directed acyclic graphs (DAGs): assign color sets to vertices such that a directed edge $u o v$ exists if and only if the sets intersect and $v$’s color set is strictly larger than $u$’s, minimizing the total number of distinct colors. We establish, for the first time, that this NP-hard problem admits a polynomial-time exact algorithm on the subclass of triangle-free Hamiltonian DAGs. We present the first combinatorial algorithm for this constrained setting and prove tight equivalences and bounds linking it to classical models—including poset dimension, interval graph embeddings, and other intersection-based representations. Our analysis systematically uncovers deep connections between directed intersection representations and broader theories of graph and digraph representation, thereby unifying and extending prior work in structural and algorithmic graph theory.

1 citationsRead paper

FF-SRL: High Performance GPU-Based Surgical Simulation For Robot Learning

Oct 14, 2024IEEE/RJS International Conference on Intelligent RObots and Systems

To address the safety and efficiency bottlenecks arising from heavy reliance on real-world interactions in reinforcement learning (RL) training for robotic surgery, this paper introduces the first full-stack, GPU-native simulation learning framework. It unifies soft-tissue physics simulation (CUDA-accelerated), RL policy training, and lightweight high-fidelity scene modeling entirely on a single GPU—eliminating CPU-GPU data transfer overhead. The architecture enables millisecond-scale closed-loop interaction and minute-scale training for complex surgical tasks, reducing training time for representative tissue manipulation tasks to just 2–3 minutes—over 10× faster than state-of-the-art CPU/GPU hybrid approaches. All code is open-sourced, establishing a new paradigm for safe, efficient, and scalable autonomous surgical skill acquisition.

1 citationsRead paper

FineHOI: Part-Aware Dense Representations for Zero-Shot Human-Object Interaction Detection

Sep 05, 2026

Human-Object Interaction (HOI) detection aims to localize humans and objects in images and classify their interactions. Zero-shot HOI focuses on recognizing interactions that are not observed during training, requiring models to generalize beyond seen verb-object compositions. Recent approaches leverage Vision-Language Models (VLMs), benefiting from rich semantic representations. However, they often rely on global or detector-centric features that compress interaction cues and hinder fine-grained spatial reasoning. To overcome this limitation, we propose FineHOI, a zero-shot HOI framework that explicitly models interactions from dense patch-level features. Our approach is motivated by the observation that human-object interactions are defined by localized spatial relationships, which are not preserved by global and detector-centric representations. To this end, we introduce an Adaptive Part-Level Attention module that decomposes humans and objects into semantically coherent parts via unsupervised clustering, and re-weights them based on their interaction relevance. These representations are then integrated through a Region-Aware Interaction Transformer that integrates part-aware and global features and produces the final HOI embedding. Extensive experiments demonstrate that FineHOI consistently outperforms existing zero-shot HOI methods, achieving particularly strong gains on unseen interactions. Code is available at https://github.com/francescotonini/fine-hoi.

0 citationsRead paper

TokenMatch: 3D Mesh Correspondence Transformer with Curvature-Guided Tokenisation

Sep 03, 2026

为解决3D形状在部分观察和非等距变形下的匹配难题，本文提出TokenMatch，一种基于曲率引导的Transformer模型，通过自注意力和交叉注意力机制学习形状间的密集对应关系。

0 citationsRead paper

Recent publications

Latest Papers

FineHOI: Part-Aware Dense Representations for Zero-Shot Human-Object Interaction Detection

Sep 05, 2026

Human-Object Interaction (HOI) detection aims to localize humans and objects in images and classify their interactions. Zero-shot HOI focuses on recognizing interactions that are not observed during training, requiring models to generalize beyond seen verb-object compositions. Recent approaches leverage Vision-Language Models (VLMs), benefiting from rich semantic representations. However, they often rely on global or detector-centric features that compress interaction cues and hinder fine-grained spatial reasoning. To overcome this limitation, we propose FineHOI, a zero-shot HOI framework that explicitly models interactions from dense patch-level features. Our approach is motivated by the observation that human-object interactions are defined by localized spatial relationships, which are not preserved by global and detector-centric representations. To this end, we introduce an Adaptive Part-Level Attention module that decomposes humans and objects into semantically coherent parts via unsupervised clustering, and re-weights them based on their interaction relevance. These representations are then integrated through a Region-Aware Interaction Transformer that integrates part-aware and global features and produces the final HOI embedding. Extensive experiments demonstrate that FineHOI consistently outperforms existing zero-shot HOI methods, achieving particularly strong gains on unseen interactions. Code is available at https://github.com/francescotonini/fine-hoi.

0 citationsRead paper

TokenMatch: 3D Mesh Correspondence Transformer with Curvature-Guided Tokenisation

Sep 03, 2026

为解决3D形状在部分观察和非等距变形下的匹配难题，本文提出TokenMatch，一种基于曲率引导的Transformer模型，通过自注意力和交叉注意力机制学习形状间的密集对应关系。

0 citationsRead paper

VERAGMIL: Virtual Environment for Scooping Granular Foods with Imitation Learning Models

Aug 18, 2026

为了解决机器人辅助喂食系统处理颗粒状食物的难题，VERAGMIL框架结合了高保真模拟器和虚拟现实界面，通过模仿学习方法训练模型，显著提高了任务成功率并减少了洒落。

0 citationsRead paper

An Optimal Energy Production Problem with Energy Source Switching and Load Following Nuclear Power Plants

Aug 10, 2026

This study addresses the challenge of optimally dispatching nuclear power units to track stochastic net load and switch operating modes in power systems with high renewable penetration. It formulates the problem, for the first time, as a finite-horizon optimal switching problem with stochastic demand, explicitly incorporating ramping constraints and mode transition costs. The value function is characterized via a Hamilton–Jacobi–Bellman quasi-variational inequality, and the optimal policy is computed using a monotone semi-Lagrangian scheme. The analysis quantifies how factors such as loss-of-load cost, switching cost, ramping capability, and market access influence nuclear flexibility, revealing its distinct economic value in both regulated and liberalized electricity markets. These findings provide theoretical foundations for designing compensation mechanisms and market structures that support low-carbon flexible resources.

0 citationsRead paper

SLAP: Selective Local Vision-Language Alignment for Fish Re-Identification via Partial Optimal Transport

Aug 09, 2026

This work addresses the limitations of existing CLIP-based methods in fish re-identification, which rely on global image-text alignment and are thus susceptible to background clutter and non-discriminative regions. To overcome this, the authors propose a selective local visual-language alignment framework that, for the first time, incorporates partial optimal transport (POT) into fine-grained fish re-identification. By leveraging POT, the method establishes selective correspondences between local image patch embeddings and multiple identity-aware textual prompts, thereby enhancing strongly relevant cross-modal associations while suppressing interference from irrelevant regions. Built upon the CLIP architecture and integrating local embedding, multi-prompt encoding, and end-to-end training, the proposed approach significantly outperforms existing CLIP-based methods on the Symphodus melops dataset and demonstrates superior generalization across multiple marine species re-identification benchmarks.

0 citationsRead paper