Introduction The objective of the research is to develop a real-time, resource-aware, and quality-adaptive monitoring and decision layer to support Complex Financial Patterns (CFP) based Real-Time Services Delivery called fSense. fSense will provide the basic sensing service interface to support distributed pattern and anomaly detection and decision making processes for financial analysts, executives and fraud managers. It will support heterogeneous data ingesting for various types of financial and social media data feeds, fusing data streams to extract, filter, aggregate, and validate financial patters and outliers communicating fused and filtered pattern streams onto services access and composition points fSense will be composed of a run-time environment that will provide various resource-aware and qualityadaptive sensing and patterning services (Figure 1(b)) The input paramaters to these service calls include expected functionality descriptions and the required Quality of Service (QoS). The data collected at individual input streams are valuable when the corresponding spatio-temporal, social, behavioral and financial contexts are taken into consideration. Therefore, the fSense model will account for the timevarying, as well as the contextual nature of the data streams in order to continously identify novelties and outliers. The requirements for this Complex Financial Patterns (CFP) based Real-Time Services Delivery network will be developed in pertnership with Early Warning . Early Warning offers a broad range of fraud prevention and risk management services to help fight fraud in the financial system. Through industry-wide collaboration Early Warning helps address such issues as deposit fraud, payment fraud and identity fraud. The outcomes of the research will be evaluated by Early Warning. Technical Aproach We see that the objective of the proposed research is complicated by (1) the ad-hoc and decentralized nature of the services architectures; (2) imprecise and redundant nature of the collected data; and (3) the limited computation, storage, power, and communication resources of the services infrastructure. The sparcity of the data requires algorithms for data prediction. The volume of the real-time data that has to flow through the services infrastructure requires aggregation and filtering of the streaming data at the intermediary comupute nodes. Imprecise nature of the sensed data requires clustering based computation to benefit from spatio-temporal and social redundancy for data boosting. Resource limitations force the services processing nodes to adapt their QoS to varying conditions. The various resource constraints on the service infrastructure require a data architecture optimized on specific tasks and features to save resources. Therefore development of the fSense autonomous and adaptive monitoring, pattern and anomaly detection and decision making services layer involves following technical challenges (Figure 2(a)): Specification: describing various meta-data and high-level workflow descriptions with QoS guarantees, Scheduling: context-aware activation of complex pattern discovery and matching nodes by routing required tasks and QoS parameters within the data infrastructure, Calibration: creating, calibrating, and maintaining multi-granularity data stream descriptors to support real-time integration and contextual pattern based inferences, Integration: fusing redundant financial data-streams within the data and sevrvices network, Alerts and Triggers: filtering and detection of novel complex and multi-stream financial event patterns from integrated data-streams to within the required QoS guarantees, and utilization of these complex financial patterns for alert and triggers within the services infrastructure Autonomous services adaptation: ability to maintain QoS guarentees by decenteralized health (resource and pre- cision) monitoring and network recalibration in response to dynamic sensor resource availabilities.
|Effective start/end date||8/1/14 → 7/31/19|
- National Science Foundation (NSF): $805,129.00