Giving insurers an Edge in data farming
A combination of autonomous driving and real time data processing has highlighted the shortfalls in cloud based storage, in particular with latency issues. It’s now clear that carmakers need to develop an effective strategy for the connected car that isn’t solely reliant on the cloud.
“While cloud based storage and processing is continuously improving, there is a cost associated with transferring data (for example if the data is transferred using internet) and storing data in the cloud,” says Anupam Malhotra, director, connected vehicles and data at Audi of America. “Moreover, entirely relying on cloud for data processing and sending all the data to cloud increases the network traffic and increases latency which can be critical to the vehicle’s performance in certain types of traffic situations where an immediate reaction is necessary.” Creating a network of edge computing seems to be the most tangible solution to this, says Malhotra. “For all these reasons, edge computing techniques can be an effective way to process and analyse the data at the device level, gather the necessary insights and transfer only the required data to the cloud.”
Offering a different view is Sven L. Andén, chairman & CSO, cyber security consultant at Silent Cyber, Sandab Group, saying: “Sending all data to the cloud is a fool’s errand. I would advocate that the data collection dongles should be designed with enough internal intelligence to record the data and forward relevant derived data to the cloud to be mined for significance. This would be with the provision that within a period of time after a hit is achieved from the mining, a full dump of preconditions, event, and post conditions can be requested for transmission in compressed form during low bandwidth times. Data logging for ADAS development, for example, has already been working on solving these issues for high bandwidth recorded sensor data with vehicle CAN information for algorithm testing efforts.”
Increasing capacity and processing power
Using a hybrid edge/cloud data model to enable distributed data processing based on a priority analytics model should drastically increase capacity and processing power, say our experts. Malhotra explains: “We understand the auto insurance industry is exploring ways to leverage vehicle and driving behaviour data to better assess risk profiles and set insurance premiums reflective of driving style/behaviour. In order to achieve this, however, insurance providers would need a wealth of personalised driver information such as second-by-second GPS data, braking data, accident data.
“While all the required sensitive data can be transferred to the cloud to perform the analysis, a potentially big hindrance is convincing customers of the value of such services and getting agreement to share data. Additionally, the industry also needs to address how to preserve sensitive data in the cloud in a way that ensures all privacy requirements are met. The hybrid edge/cloud data model could address this issue by pre-processing highly sensitive data and sending to the cloud only aggregated privacy compliant data for further analysis which could increase the chances of receiving customer consent.”
Andén is also working on similar priority based technology: “We are working with these issues in our soon to be released Sentry product that is able to forward cyber security incident data to the cloud for data mining efforts, yet still provide both system log capabilities for devices on the internal vehicle networks and valuable forensic information.”
Data efficiency
To maximise data processing the auto insurance industry will need to focus on AI and machine learning models to employ a combination of edge computing, the cloud and in-car technology. With an increase in the number of connected vehicles, the auto industry will have access to a growing wealth of data about their customers, says Malhotra.
“As a result, the auto insurance industry business model will likely need to evolve rather quickly. This could mean using artificial intelligence and machine learning algorithms to create complete risk profiles for customers based not just on traditional data sources (such as age, demographics, location) but also on real-time or near real-time driving data. In this effort, the insurance industry needs to be open to collaboration with automakers to leverage technologies like machine learning, AI, and edge computing in the vehicle and in the cloud. Further, there should be a discussion on the potential value of allowing insurance industry algorithms for risk scoring, etc. to run within automaker computing environments instead of requiring all data be transferred over.”
Asthe insurance industry begins to look at cyber security incidents in vehicles to quantify the risk presented by cyber intrusions, ever more data will need to be accumulated and stored, Andén concludes. “The efficient and timely generation of information from that data, will require more research. This research will continue across all links of the data gathering and processing chain: accumulating data in vehicle, forwarding derived information to the cloud, real time monitoring of the derived data in the cloud for indicators that a full log is relevant, downloading a full log when possible, then aggregating the data from the logs through time and across market segments, and creating information from the data in formats that identify trends, or relevant deviations from those trends.”
[Ins.Weeks.2017.10.23]
