Marco Giordano

Moritz Imfeld, Jacopo Graldi, Marco Giordano et al. · eth-zurich

Fusion is a technique for merging multiple independently-trained neural networks in order to combine their capabilities. Past attempts have been restricted to the case of fully-connected, convolutional, and residual networks. This paper presents a systematic approach for fusing two or more transformer-based networks exploiting Optimal Transport to (soft-)align the various architectural components. We flesh out an abstraction for layer alignment, that can generalize to arbitrary architectures - in principle - and we apply this to the key ingredients of Transformers such as multi-head self-attention, layer-normalization, and residual connections, and we discuss how to handle them via various ablation studies. Furthermore, our method allows the fusion of models of different sizes (heterogeneous fusion), providing a new and efficient way to compress Transformers. The proposed approach is evaluated on both image classification tasks via Vision Transformer and natural language modeling tasks using BERT. Our approach consistently outperforms vanilla fusion, and, after a surprisingly short finetuning, also outperforms the individual converged parent models. In our analysis, we uncover intriguing insights about the significant role of soft alignment in the case of Transformers. Our results showcase the potential of fusing multiple Transformers, thus compounding their expertise, in the budding paradigm of model fusion and recombination. Code is available at https://github.com/graldij/transformer-fusion.

Marco Giordano, Kanika Dheman, Michele Magno

Sepsis is a lethal syndrome of organ dysfunction that is triggered by an infection and claims 11 million lives per year globally. Prognostic algorithms based on deep learning have shown promise in detecting the onset of sepsis hours before the actual event but use a large number of bio-markers, including vital signs and laboratory tests. The latter makes the deployment of such systems outside hospitals or in resource-limited environments extremely challenging. This paper introduces SepAl, an energy-efficient and lightweight neural network, using only data from low-power wearable sensors, such as photoplethysmography (PPG), inertial measurement units (IMU), and body temperature sensors, designed to deliver alerts in real-time. SepAl leverages only six digitally acquirable vital signs and tiny machine learning algorithms, enabling on-device real-time sepsis prediction. SepAl uses a lightweight temporal convolution neural network capable of providing sepsis alerts with a median predicted time to sepsis of 9.8 hours. The model has been fully quantized, being able to be deployed on any low-power processors, and evaluated on an ARM Cortex-M33 core. Experimental evaluations show an inference efficiency of 0.11MAC/Cycle and a latency of 143ms, with an energy per inference of 2.68mJ. This work aims at paving the way toward accurate disease prediction, deployable in a long-lasting multi-vital sign wearable device, suitable for providing sepsis onset alerts at the point of care. The code used in this work has been open-sourced and is available at https://github.com/mgiordy/sepsis-prediction

Marco Giordano, Stefano Giacomelli, Claudia Rinaldi et al.

We present a full-stack emergency vehicle (EV) siren detection system designed for real-time deployment on embedded hardware. The proposed approach is based on E2PANNs, a fine-tuned convolutional neural network derived from EPANNs, and optimized for binary sound event detection under urban acoustic conditions. A key contribution is the creation of curated and semantically structured datasets - AudioSet-EV, AudioSet-EV Augmented, and Unified-EV - developed using a custom AudioSet-Tools framework to overcome the low reliability of standard AudioSet annotations. The system is deployed on a Raspberry Pi 5 equipped with a high-fidelity DAC+microphone board, implementing a multithreaded inference engine with adaptive frame sizing, probability smoothing, and a decision-state machine to control false positive activations. A remote WebSocket interface provides real-time monitoring and facilitates live demonstration capabilities. Performance is evaluated using both framewise and event-based metrics across multiple configurations. Results show the system achieves low-latency detection with improved robustness under realistic audio conditions. This work demonstrates the feasibility of deploying IoS-compatible SED solutions that can form distributed acoustic monitoring networks, enabling collaborative emergency vehicle tracking across smart city infrastructures through WebSocket connectivity on low-cost edge devices.

Stefano Giacomelli, Marco Giordano, Claudia Rinaldi et al.

Accurate recognition of Emergency Vehicle (EV) sirens is critical for the integration of intelligent transportation systems, smart city monitoring systems, and autonomous driving technologies. Modern automatic solutions are limited by the lack of large scale, curated datasets and by the computational demands of state of the art sound event detection models. This work introduces E2PANNs (Efficient Emergency Pre trained Audio Neural Networks), a lightweight Convolutional Neural Network architecture derived from the PANNs framework, specifically optimized for binary EV siren detection. Leveraging our dedicated subset of AudioSet (AudioSet EV) we fine-tune and evaluate E2PANNs across multiple reference datasets and test its viability on embedded hardware. The experimental campaign includes ablation studies, cross-domain benchmarking, and real-time inference deployment on edge device. Interpretability analyses exploiting Guided Backpropagation and ScoreCAM algorithms provide insights into the model internal representations and validate its ability to capture distinct spectrotemporal patterns associated with different types of EV sirens. Real time performance is assessed through frame wise and event based detection metrics, as well as a detailed analysis of false positive activations. Results demonstrate that E2PANNs establish a new state of the art in this research domain, with high computational efficiency, and suitability for edge-based audio monitoring and safety-critical applications.

Marco Giordano, Claudia Rinaldi

Spoken language datasets are vital for advancing linguistic research, Natural Language Processing, and speech technology. However, resources dedicated to Italian, a linguistically rich and diverse Romance language, remain underexplored compared to major languages like English or Mandarin. This survey provides a comprehensive analysis of 66 spoken Italian datasets, highlighting their characteristics, methodologies, and applications. The datasets are categorized by speech type, source and context, and demographic and linguistic features, with a focus on their utility in fields such as Automatic Speech Recognition, emotion detection, and education. Challenges related to dataset scarcity, representativeness, and accessibility are discussed alongside recommendations for enhancing dataset creation and utilization. The full dataset inventory is publicly accessible via GitHub and archived on Zenodo, serving as a valuable resource for researchers and developers. By addressing current gaps and proposing future directions, this work aims to support the advancement of Italian speech technologies and linguistic research.

Julian Moosmann, Marco Giordano, Christian Vogt et al.

This paper introduces a highly flexible, quantized, memory-efficient, and ultra-lightweight object detection network, called TinyissimoYOLO. It aims to enable object detection on microcontrollers in the power domain of milliwatts, with less than 0.5MB memory available for storing convolutional neural network (CNN) weights. The proposed quantized network architecture with 422k parameters, enables real-time object detection on embedded microcontrollers, and it has been evaluated to exploit CNN accelerators. In particular, the proposed network has been deployed on the MAX78000 microcontroller achieving high frame-rate of up to 180fps and an ultra-low energy consumption of only 196μJ per inference with an inference efficiency of more than 106 MAC/Cycle. TinyissimoYOLO can be trained for any multi-object detection. However, considering the small network size, adding object detection classes will increase the size and memory consumption of the network, thus object detection with up to 3 classes is demonstrated. Furthermore, the network is trained using quantization-aware training and deployed with 8-bit quantization on different microcontrollers, such as STM32H7A3, STM32L4R9, Apollo4b and on the MAX78000's CNN accelerator. Performance evaluations are presented in this paper.