Personal Information Hub (PIH) Architecture

We will provide a distributed, scalable, provenance-tracking Private Storage Container based on “lazy provenance” with social network hints. We will extend the I/O and encryption support in the Mirage OS to develop dedicated microkernel VMs that provide specialist storage services with minimal resource usage, written in type-safe, high level language code. We will provide a cloud based framework to support secure and user policy-compliant processing of aggregate private data from a large number of users. This includes a system where the data and the data processing is secured inside a trusted execution environment (data can thus be in the clear without privacy risks), which is isolated from the rest of the system, has very restrictive I/O channels, and incorporates multiple stages of verification to enforce user privacy policies. 

We will develop tools for automatic collection of users’ context. This includes new passive data collection methods to analyse individual applications' performance and identify in-browser applications, as well as extensions to existing tools to better capture user I/O activity (e.g. microphones, cameras) and additional information about the user network environment. We will extend the HostView framework by developing mechanisms that are less intrusive and at the same time can obtain feedback from user explicit input on content preference, user context, and activities. We will also develop methods to correlate these different data streams into datasets annotated with the user feedback. We will develop the data collectors that add semantics to the raw data and store broad range of various (user centric) data in common storage framework for the first time.

We will develop a set of user profiling techniques and models (to be used for service personalization), including Content consumption profiling techniques that embrace more detailed content characterization, such as domain knowledge and relationships between different content features including genres, language, topic, resolution, etc. Models of social influence on content consumption, capturing the influence circle around a user and characterizing the influence depending interests/topics. Application-agnostic predictors of users’ satisfaction of network performance. We will design new context-aware content recommendation algorithms that takes into account our enhanced content consumption profiles and social influence models. We will develop a framework with new content service optimizations and personalization mechanisms. Content pre-fetching to a device in the user vicinity based on social and content consumption profiles and user mobility pattern. User-centric content personalization on local STB devices o Heterogeneous network resource management mechanisms going beyond pure network resource optimizations by including on user context and preferences.

We will advance the state of the art in privacy preserving data aggregation by developing mechanisms to aggregate users’ data based on existing techniques such as homomorphic encryption and secure multi-party computation (SMC) in order to assure the correctness of the resulting aggregate even with the existence of some misbehaving nodes. We will extend privacy preserving data lookup by designing dedicated primitives for private information retrieval and searchable encryption techniques for distributed environments. In our system, an adversary will not be able to learn anything about queries or results, and the new privacy primitives will enable to verify the correctness of operations. We will extend differential privacy techniques to operate over a distributed environment which adds noise to disclosed query results so that it bounds the ability of an adversary to infer from any query output, whether a specific record was present in the input to the computation.

We will undertake a range of focused ethnographic studies of the UCN technologies deployed in real world settings. These will combine qualitative studies of use with captured information from the network to seek to analyse and expose the day in the life of these technologies in use. Extend our longitudinal study to investigate how network and service management mechanisms can become more user-centered and reflect human interaction subtlety.

Developing a novel UCN PoC prototype based on two new uses cases going beyond state of the art in terms of the underlying service features, e.g., privacy-preserving user context-aware content recommendation and personalized content delivery optimizations.

Based on their experience in user data handling and analytics as well as their recommender engine activities, Fraunhofer FOKUS will participate in WP2 in the collection of user behavioural data and in deriving personalized recommendations based on such data in WP3. Fraunhofer will also use its expertise in hybrid and multi-device applications to develop use case specifications participate in system integration and provide demonstrators for the UCN software.

Project details:

Partners: Technicolor Research and Innovation Paris, EURECOM, Fraunhofer FOKUS, Intamac, Martel, University Cambridge, University Nottingham, UPMC/LIP6, NICTA; Portugal Telecom