Predictive Objects improves the outcome of business and industrial processes through actionable customer and asset
intelligence. The platform leverages machine learning, AI and big data to automate the creation and deployment of predictive models,
augmenting business experts for faster and more accurate predictions that run where the data is produced and where
decisions need to be made.
Predictive Objects can be deployed and run anywhere: inside connected objects or machines, at the edge of the network,
in industrial IoT platforms, in public or private clouds - bringing the decision to the object and to the data in real time.
Predictive Objects removes the data selection bias by providing access to all
data available within the organization - internal or external. This enables Predictive Objects
to process the entire Gaussian distribution of data, and detect weak signals that would
otherwise be ignored. The process is greatly accelerated through massive automation,
shortening projects from months to hours
Ingest any structured data of any type, internal or exogenous data (open data, web), as batches and/or streams.
Check and cleanse all data, homogenize and normalize sources, verify integrity.
Anonymize data before processing through record transposition for maximum protection.
Raw data produced by sensors or other assets are not directly usable by machine learning algorithms. Predictive Objects automates or facilitates the preparation of the datasets, and augments data with information that will make predictions more valuable for the business.
Turn raw data into temporal sequences of information based on the data type and density.
Add business-related information to the existing data: asset identification and classification, cost or revenue information.
Predictive models that do not account for business objectives are difficult to use by business users,
and model selection can be severely impacted. Predictive Objects lets the business expert define
a cost matrix that targets operational costs or gains, and computes the business value of each model.
When comparing models, Predictive Objects does not focus on technical parameters such as f-measure or recall, but creates a confusion matrix that integrates the cost dimension. Models that are deployed are not necessarily the best ones from a mathematical standpoint, but the ones that generate the highest ROI.
Predictive models created by Predictive Objects not only clearly measure the global value of predictions, they also account for resource availability, cost or practicality of operations. It is therefore possible for business users to target and understand operational costs and gains of these models.
Deployment to production of predictive models is immediate
to any targeted asset: object, device, platform, cloud.
There is no need for coding, dramatically shortening the
time it takes for projects to deliver value.
The Meta Active Machine Learning, Predictive Objects’ wisdom that is shared across projects and enriched at each model computation, dramatically accelerates the search for the best model, shortening to hours a process that used to take months of exploration by data science teams.
Find quickly if a model work, test hypothesis, and ensure ongoing efficiency: the performance of models is continuously assessed to test their accuracy, precision and recall. Production and deployment of new and revised models is ongoing.
Business experts can easily build models, test hypotheses, prove or disprove assumptions, and hence get value instantly. Through the use of symbolic algorithms, models are driven by variables and based on a decision tree that explains not only the prediction but also the impact of each variable on this prediction.
Predictive Objects run seamlessly and consistently on all types of assets, regardless of their level of computing power or their state of connectivity:
Thanks to their ability to get deployed and embedded on any platform, in any runtime, Predictive Objects packages intelligence with every asset, providing a combined view of the asset, its performance, and its evolution over time (predicted behavior, deviance from the model and therefore associated risk).
Through the use of symbolic algorithms, models generated by Predictive Objects are driven by variables and based on a decision tree that can be visualized by business experts, explaining not only the prediction but also the impact of each variable on this prediction.
The simulation capabilities of Predictive Objects make it possible to test "what if" scenario for each individual asset or cluster of assets. By adjusting variables, in the past, in the present or in the future, business users can visually evaluate the impact that any actions would have had, or that they will have, on the future performance of the asset.
By embedding intelligence inside assets, Edge AI reduces the impact of
data transmission latency & costs as well as addresses security issues
implied by the need to transfer huge amounts of data through network
to a cloud-hosted application (IoT platform for example)
Predictive Objects can be deployed and run seamlessly inside any connected assets, at the edge of the network, or in industrial IoT platforms- placing the predictive model as close as possible to the system it impacts. This way the asset can make decision locally in near real time enabling offline operation event with intermittent connectivity.
Rather than sending all data from connected assets to a remote cloud or infrastructure for analytics purposes, the IIoT will rely on the capacity to collect and process the raw data as close as possible to the machines that generate them. Hence reducing network security issues and additional bandwidth requirement.
Interested in our Predictive Objects features, set-up, model, etc?
Fill in the following form to request a meeting.