Indro Spinelli

Automating multi‐stage decision workflows in regulated industrial domains demands intelligent systems capable of orchestrating heterogeneous evidence sources, such as uncontrolled photographs, free‐text narratives and structured forms, under strict operational and compliance constraints. We present FADES, a production‐grade expert system that integrates conversational AI with computer vision to automate end‐to‐end claim evidence interpretation and vehicle damage assessment. The system addresses three core challenges: structured data acquisition via state‐machine‐controlled conversational agents powered by tool‐enabled Large Language Models (LLMs); robust damage localization through modern object detection; and deterministic severity estimation via panel‐aware geometric reasoning and rule‐based logic. Our methodology combines backend‐enforced slot‐filling for First Notice of Loss (FNOL) data collection, ensuring compliance with the Italian CAI standard, with a multi‐stage damage analysis pipeline that reconciles detection outputs with vehicle panel segmentation. We systematically evaluated diffusion‐based (DiffusionDet) and transformer‐based (RF‐DETR, D‐FINE) detectors on 13,353 real accident images comprising 41,057 annotations across four damage classes. RF‐DETR Large achieves ( over DiffusionDet), () and , while satisfying real‐time processing requirements. Operational validation demonstrates that the integrated system reduces claim processing time from 30 to 12 min, achieving time‐to‐completion under 7 min at a sustained throughput of 45 cycles per day. Cross‐dataset experiments reveal substantial domain gaps ( on CarDD), confirming the necessity of insurance‐specific training data. Micro‐ablation analysis shows full pipeline integration outperforms partial implementations by in end‐to‐end efficiency. These results provide empirical evidence that hybrid deterministic‐generative architectures deliver the reliability and auditability required for regulated industrial applications.

The NA62 experiment at the Super Proton Synchrotron at CERN aims to measure the branching ratio of the ultra-rare channel 𝐾 + → 𝜋+ 𝜈𝜈¯ with a predicted BR = (8.6 ± 0.42) × 10^ −11. The GigaTracker, a four-layer silicon-pixel tracker, measures the momentum and direction of incoming hadron beam particles at a rate of 750 MHz. The experiment’s success relies on effectively managing pile-up, temporal matching and vertex reconstruction between beam kaons and decay pions, a challenging task due to the current combinatorial track-building approach. This research integrates machine learning best practices in the current modeling strategies applied in the NA62 experiment. In this study, we extensively evaluate different deep learning architectures for particle tracking, showing the most effective approaches. In particular, we designed and implemented three distinct strategies based on Multi-Layer Perceptron, Transformer and Graph Neural Network. All methods were evaluated using efficiency, purity and fake-track rate. The most promising ones were further assessed in an aggregation-free setup, where clustering based on ground truth information was removed to test their robustness in a real-world scenario.

Objective. Deep learning tools applied to high-resolution neurophysiological data have significantly progressed, offering enhanced decoding, real-time processing, and readability for practical applications. However, the design of artificial neural networks to analyze neural activity in vivo remains a challenge, requiring a delicate balance between efficiency in low-data regimes and the interpretability of the results. Approach. To address this challenge, we introduce a novel specialized transformer architecture to analyze single-neuron spiking activity. The model is tested on multi-electrode recordings from the dorsal premotor cortex of non-human primates performing a motor inhibition task. Main results. The proposed architecture provides an early prediction of the correct movement direction, achieving accurate results no later than 230 ms after the Go signal presentation across animals. Additionally, the model can forecast whether the movement will be generated or withheld before a stop signal, unattended, is actually presented. To further understand the internal dynamics of the model, we compute the predicted correlations between time steps and between neurons at successive layers of the architecture, with the evolution of these correlations mirrors findings from previous theoretical analyses. Significance. Overall, our framework provides a comprehensive use case for the practical implementation of deep learning tools in motor control research, highlighting both the predictive capabilities and interpretability of the proposed architecture.

Event cameras, known for low-latency operation and superior performance in challenging lighting conditions, are suitable for sensitive computer vision tasks such as semantic segmentation in autonomous driving. However, challenges arise due to limited event-based data and the absence of large-scale segmentation benchmarks. Current works are confined to closed-set semantic segmentation, limiting their adaptability to other applications. In this paper, we introduce OVOSE, the first Open-Vocabulary Semantic Segmentation algorithm for Event cameras. OVOSE leverages synthetic event data and knowledge distillation from a pre-trained image-based foundation model to an event-based counterpart, effectively preserving spatial context and transferring open-vocabulary semantic segmentation capabilities. We evaluate the performance of OVOSE on two driving semantic segmentation datasets DDD17, and DSEC-Semantic, comparing it with existing conventional image open-vocabulary models adapted for event-based data. Similarly, we compare OVOSE with state-of-the-art methods designed for closed-set settings in unsupervised domain adaptation for event-based semantic segmentation. OVOSE demonstrates superior performance, showcasing its potential for real-world applications. The code is available at https://github.com/ram95d/OVOSE.

The target duration of a synthesized human motion is a critical attribute that requires modeling control over the motion dynamics and style. Speeding up an action performance is not merely fast-forwarding it. However, state-of-the-art techniques for human behavior synthesis have limited control over the target sequence length. We introduce the problem of generating length-aware 3D human motion sequences from textual descriptors, and we propose a novel model to synthesize motions of variable target lengths, which we dub "Length-Aware Latent Diffusion" (LADiff). LADiff consists of two new modules: 1) a length-aware variational auto-encoder to learn motion representations with length-dependent latent codes; 2) a length-conforming latent diffusion model to generate motions with a richness of details that increases with the required target sequence length. LADiff significantly improves over the state-of-the-art across most of the existing motion synthesis metrics on the two established benchmarks of HumanML3D and KIT-ML.

Deep learning tools applied to high-resolution neurophysiological data have significantly progressed, offering enhanced decoding, real-time processing, and readability for practical applications. However, the design of artificial neural networks to analyze neural activity remains a challenge, requiring a delicate balance between efficiency in low-data regimes and the interpretability of the results. To this end, we introduce a novel specialized transformer architecture to analyze single-neuron spiking activity. We test our model on multi electrodes recordings from the dorsal premotor cortex (PMd) of non-human primates while performing a motor inhibition task. The proposed architecture provides a very early prediction of the correct movement direction - no later than 230ms after the Go signal presentation across animals - and can accurately forecast whether the movement will be generated or withheld before a Stop signal, unattended, is actually presented. We also analyze the internal dynamics of the model by computing the predicted correlations between time steps and between neurons at successive layers of the architecture. We find that their evolution mirrors previous theoretical analyses. Overall, our framework provides a comprehensive use case for the practical implementation of deep learning tools in motor control research.