A5049269493- Domain Adaptation and Few-Shot Learning
- Anomaly Detection Techniques and Applications
- Video Surveillance and Tracking Methods
- Human Pose and Action Recognition
- Multimodal Machine Learning Applications
- Advanced Image and Video Retrieval Techniques
- COVID-19 diagnosis using AI
- Viral Infections and Vectors
- Complex Network Analysis Techniques
- Data-Driven Disease Surveillance
- Mosquito-borne diseases and control
- Vector-Borne Animal Diseases
- Face and Expression Recognition
- Network Security and Intrusion Detection
- Time Series Analysis and Forecasting
- Image Retrieval and Classification Techniques
- Machine Learning and ELM
- Advanced Graph Neural Networks
- Food Supply Chain Traceability
- Animal Disease Management and Epidemiology
- COVID-19 epidemiological studies
- Remote-Sensing Image Classification
- Autonomous Vehicle Technology and Safety
- Advanced Neural Network Applications
- Text and Document Classification Technologies
University of Modena and Reggio Emilia
2018-2024
Ferrari (Italy)
2020-2022
Novelty detection is commonly referred as the discrimination of observations that do not conform to a learned model regularity. Despite its importance in different application settings, designing novelty detector utterly complex due unpredictable nature novelties and inaccessibility during training procedure, factors which expose unsupervised problem. In our proposal, we design general framework where equip deep autoencoder with parametric density estimator learns probability distribution...
Continual Learning has inspired a plethora of approaches and evaluation settings; however, the majority them overlooks properties practical scenario, where data stream cannot be shaped as sequence tasks offline training is not viable. We work towards General (GCL), task boundaries blur domain class distributions shift either gradually or suddenly. address it through mixing rehearsal with knowledge distillation regularization; our simple baseline, Dark Experience Replay, matches network's...
In Continual Learning, a Neural Network is trained on stream of data whose distribution shifts over time. Under these assumptions, it especially challenging to improve classes appearing later in the while remaining accurate previous ones. This due infamous problem catastrophic forgetting, which causes quick performance degradation when classifier focuses learning new categories. Recent literature proposed various approaches tackle this issue, often resorting very sophisticated techniques....
The staple of human intelligence is the capability acquiring knowledge in a continuous fashion. In stark contrast, Deep Networks forget catastrophically and, for this reason, sub-field Class-Incremental Continual Learning fosters methods that learn sequence tasks incrementally, blending sequentially-gained into comprehensive prediction. This work aims at assessing and overcoming pitfalls our previous proposal Dark Experience Replay (DER), simple effective approach combines rehearsal...
The recent growth in the number of satellite images fosters development effective deep-learning techniques for Remote Sensing (RS). However, their full potential is untapped due to lack large annotated datasets. Such a problem usually countered by fine-tuning feature extractor that previously trained on ImageNet dataset. Unfortunately, domain natural differs from RS one, which hinders final performance. In this work, we propose learn meaningful representations imagery, leveraging its...
Accurate prediction of future human positions is an essential task for modern video-surveillance systems. Current state-of-the-art models usually rely on a "history" past tracked locations (e.g., 3 to 5 seconds) predict plausible sequence up the next seconds). We feel that this common schema neglects critical traits realistic applications: as collection input trajectories involves machine perception (i.e., detection and tracking), incorrect fragmentation errors may accumulate in crowded...
Continual Learning (CL) investigates how to train Deep Networks on a stream of tasks without incurring forgetting. CL settings proposed in literature assume that every incoming example is paired with ground-truth annotations. However, this clashes many real-world applications: gathering labeled data, which itself tedious and expensive, becomes infeasible when data flow as stream. This work explores Semi-Supervised (CSSL): here, only small fraction input examples are shown the learner. We...
Denoising Diffusion Probabilistic Models have shown an impressive generation quality, although their long sampling chain leads to high computational costs. In this paper, we observe that a also error accumulation phenomenon, which is similar the exposure bias problem in autoregressive text generation. Specifically, note there discrepancy between training and testing, since former conditioned on ground truth samples, while latter previously generated results. To alleviate problem, propose...
People re-identification task has seen enormous improvements in the latest years, mainly due to development of better image features extraction from deep Convolutional Neural Networks (CNN) and availability large datasets. However, little research been conducted on animal identification re-identification, even if this knowledge may be useful a rich variety different scenarios. Here, we tackle cattle exploiting CNN show how is poorly related with human one, presenting unique challenges that...
Rehearsal approaches enjoy immense popularity with Continual Learning (CL) practitioners. These methods collect samples from previously encountered data distributions in a small memory buffer; subsequently, they repeatedly optimize on the latter to prevent catastrophic forgetting. This work draws attention hidden pitfall of this widespread practice: repeated optimization pool inevitably leads tight and unstable decision boundaries, which are major hindrance generalization. To address issue,...
In this work, we propose a Self-Supervised training strategy specifically designed for combinatorial problems. One of the main obstacles in applying supervised paradigms to such problems is requirement expensive target solutions as ground-truth, often produced with costly exact solvers. Inspired by Semi- and learning, show that it possible easily train generative models sampling multiple using best one according problem objective pseudo-label. way, iteratively improve model generation...
Culex pipiens , an important vector of many borne diseases, is a species capable to feeding on wide variety hosts and adapting different environments. To predict the potential distribution Cx. in central Italy, this study integrated presence/absence data from four-year entomological survey (2019–2022) carried out Abruzzo Molise regions, with datacube spectral bands acquired by Sentinel-2 satellites, as patches 224 × pixels 20 meters spatial resolution around each site for satellite revisit...
Abstract Diseases of the respiratory system are known to negatively impact profitability pig industry, worldwide. Considering relatively short lifespan pigs, lesions can be still evident at slaughter, where they usefully recorded and scored. Therefore, slaughterhouse represents a key check-point assess health status providing unique valuable feedback farm, as well an important source data for epidemiological studies. Although relevant, scoring in slaughtered pigs very time-consuming costly...
Artificial-intelligence-based methods are regularly used in the biomedical sciences, mainly field of diagnostic imaging. Recently, convolutional neural networks have been trained to score pleurisy and pneumonia slaughtered pigs. The aim this study is further evaluate performance a network when compared with gold standard (i.e., scores provided by skilled operator along slaughter chain through visual inspection palpation). In total, 441 lungs (180 healthy 261 diseased) included study. Each...
The occurrence of West Nile Virus (WNV) represents one the most common mosquito-borne zoonosis viral infections. Its circulation is usually associated with climatic and environmental conditions suitable for vector proliferation virus replication. On top that, several statistical models have been developed to shape forecast WNV circulation: in particular, recent massive availability Earth Observation (EO) data, coupled continuous advances field Artificial Intelligence, offer valuable...
The field of multi-object tracking has recently seen a renewed interest in the good old schema tracking-by-detection, as its simplicity and strong priors spare it from complex design painful babysitting tracking-by-attention approaches. In view this, we aim at extending tracking-by-detection to multi-modal settings, where comprehensive cost be computed heterogeneous information e.g., 2D motion cues, visual appearance, pose estimates. More precisely, follow case study rough estimate 3D is...