Big data analytics has becoming one of the most important research areas of computer science and related fields.
In Emerging Network Technology (ENT) Lab., we are exploiting new trend of
big data analytics research areas through secure machine learning (ML)-as-a-Service to provide three types of analytics services,
i.e., correlation, prediction, and causal impact, in the multi-tenant public cloud. Possible applications are intelligent security analytics, Bitcoin/Blockchain analytics,
new policy evaluation, causal inference in economic and marketing, etc.
Public cloud platform provides multi-tenant software leased services. Therefore, various participants can use software agents to effectively conduct their own services in an automated machine
learning (AutoML) pipeline. We are exploiting on how to construct numerous software-as-a-service (SaaS) services on a Platform-as-a-Service (PaaS)
framework delivered by well-known cloud service providers, such as Amazon AWS and Google GCP, to achieve secure big data analytics on correlation discovery, trend predication,
and causal impact inference. Moreover, we are also interested in extending secure machine learning techniques to emerging field of security and
privacy of Bitcoin/Blockchain, and Bitcoin/Blockchain analytics services.
Given a secure public cloud provider, it is still considered as an honest-but-curious adversary. Similar assumption will also be applied to other participants,
including data brokers, data users, machine learning model builders and service users. These participants could leverage their software agents from SaaS perspectives
to provide services in an AutoML pipeline process. The greatest research challenge is how to integrate these types of SaaS, i.e. security-as-a-service (SecaaS),
machine learning-as-a-service (MLaaS) and data broker-as-a-service (DBaaS), to achieve secure big data analytics without violating the secuirty and privacy
prinicples in multi-tenant public cloud.
We are establishing three types of SaaS on the well-known multi-tenant public cloud computing platforms. We assume machine learning model software, including SaaS programs and software agents,
and datasets are all confidential so they should be well-protected while proceeding secure machine learning model’s offline training and online testing phases of data analytics.
We consider using machine learning algorithms and integrating them with secure data and program protection algorithms, such as fully homomorphic encryption (FHE),
indistinguishability obfuscation (iO), differential privacy, secure multi-party computation techniques, to enable a two-phase machine learning process.
Furthermore, we are exploiting on how to effectively empower types of SaaS into AutoML pipeline to achieve automated secure machine learning without too much human intervention in the
analytics service loop for correlation, prediction,
and even causal impact inference in the multi-tenant public cloud.
We are currently recruiting master and Ph.D. students for big data analytics of intelligent security for data and program protection research project,
research issues include automated security and privacy of secure machine learning on the public cloud, big data analytics for intelligent security, security and privacy of
Bitcoin/Blockchain analytics, big data analytics for causal impact inference, etc.
For more detailed information, please refer to the head of ENT Lab., Prof. Yuh-Jong Hu(Email: jong at cs.nccu.edu.tw).