Beyond Signature-based Defense: Deep Learning Frameworks for Concealed Cyber Exploit Detection

The emergence of intelligent manufacturing due to Industry 4.0 and 5.0 frameworks increased the exposure to dependence on deep learning systems, which are ideal targets of advanced cyber-attacks. Traditional cybersecurity techniques are becoming insufficient to deal with the emerging trends, including covert attacks like backdoor attacks that interfere with the model integrity during the training of the model and interfere with outputs. A new, multi-layered smart cyber defence framework demanding a combination of threat modelling, data protection, and model protection has been presented in this paper to protect deep learning concept. The framework model is a permutation of both statistical and intelligent methods to avoid data privacy and confidentiality and extends structural measures of security management in terms of policies and procedures. It also presents ways of improving the sturdiness of models alongside attacks with the help of an adaptive threat model to identify the vulnerable points and implement countermeasures efficiently. An overview of back door attacks and defence mechanisms to back door attacks is given and a qualitative comparison of strategies in existence has been made. Last but not least, the paper identifies the future research directions to enhance exploit detection in deep learning by providing detailed suggestions on how the research can be taken further to promote research in resilient cyber defence towards the smart manufacturing systems.