Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

From the late Sixties on, industrial development in Italy evolved through the spread of small and medium sized firms, aggregated in district networks, with an elevated propensity to enterprise and the marked presence of owner-families. Installed within the local systems, the industrial districts tended to simulate large-scale industry exploiting lower costs generated by factors that were not only economic. The districts are characterised in terms of territorial location (above all the thriving areas of the North-east and Centre) and sector, since they are concentrated in the "4 As" (clothing-fashion, home-decor, agri-foodstuffs, automation-mechanics), with some overlapping with "Made in Italy". How can this model be assessed? This is the crucial question in the debate on the condition and prospects of the Italian productive system between the supporters of its capacity to adapt and the critics of economic dwarfism. A dispassionate judgement suggests that the prospects of "small is beautiful" have been superseded, but that the "declinist" view, that sees only the dangers of globalisation and the IT revolution for our SMEs is risky. The concept of irreversible crisis that prevails at present is limiting, both because it is not easy either to "invent", or to copy, a model of industrialisation, and because there is space for a strategic repositioning of the district enterprises. The book develops considerations in this direction, showing how an evolution of the district model is possible, focusing on: gains in productivity, scope economies (through diversification and expansion of the range of products), flexibility of organisation, capacity to meld tradition and innovation aiming at product quality, dimensional growth of the enterprises, new forms of financing, active presence on the international markets and valorisation of the resources of the territory. It is hence necessary to reactivate the behavioural functions of the entrepreneurs.

Economics --- Economics of industrial organisation --- Business & management --- Economia --- Imprese --- PMI --- Localismi d'impresa --- Globalizzazione

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Economics --- Economists --- History

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Deepfakes, synthetic media created using machine learning, commonly used to recreate another’s identity, are becoming increasingly realistic and accessible, raising concerns about their misuse for misinformation, fraud, and other malicious activities. While significant progress has been made in developing tools to detect deepfakes, these detection systems remain highly vulnerable to adversarial attack, which are small manipulations of the input sample that can deceive the detectors into making incorrect classifications. The field of Explainable AI (XAI) aims to make AI systems more transparent and interpretable and has shown promise in helping users recognize adversarial manipulations by highlighting the decision-making process of machine learning models. This thesis explores the use of an XAI for post-hoc adversarial attack detection, evaluating its ability to identify adversarial manipulations in deepfake detection systems. Additionally, in this study, we investigate how adversarial attacks distort the explanations generated by XAI, the system’s ability to generalize to new types of adversarial attacks, and how this approach compares to the widely used defense technique, adversarial training. The findings of our work show that XAI-based methods are effective at identifying manipulations when they are similar to those included in the training process, offering both interpretability and reliable classification. However, their ability to detect previously unseen types of adversarial manipulations remains limited, with certain attacks successfully bypassing detection. This limitation may be linked to overfitting in the training process, caused by the use of a dataset with limited diversity in manipulation techniques. This thesis compares our XAI-base classification to and adversarially trained model and shows that while XAI-based methods demonstrate unique advantages in terms of interpretability, adversarial training outperforms XAI-based detection in terms of overall reliability and robustness. Despite this, XAI-based detection shows potential as a complementary tool for understanding and addressing vulnerabilities in detection systems.