With its systematic multi-channel validation, Blender Logic’s C.O.R.E. provides a principled approach to AIML orchestration problems.
C.O.R.E.’s ontologically typed language module is comprised of two layers. The base layer is a collection of primitive computational classes. Such as Booleans, Strings, Integers, Hierarchies and Graphs. All modeling, all querying, all computation, reduces to some composition of these classes. The primitive computational classes are realized in some physical system such as SAP’s HANA or open source PostGresSQL.
C.O.R.E.’s next layer is a compact set of ontological basis vectors (e.g., for Space, Time, Objects and processes) with a mathematical-logic grammar for composing them into structs and expressions. Each vector can have many partitions (e.g., units or subclasses). Individual vector partitions are linked to specific primitive computational classes. E.g., the Year partition of Time may be linked to the computational class Integer while the Month-of-year partition of Time may be linked to the computational class Integer Modulo 12. Ontological basis vectors can be composed into structs (e.g., combining the partitions year, month-of-year and day-of-month into the composite type Date) and expressions (e.g., Time(event 1) before Time(Event 2)). Common modeling concepts such as ‘attribute’ and ‘relation’ are also composed of ontological basis vectors.
Blender Logic’s ontologically typed computing language guarantees cross domain comparability as AIML applications are integrated. Every ontologically typed expression OTE is composed of at least one element of each ontological basis vector. Every OTE purports to assert, deny, query or command some portion of reality. OTEs can specify both type definitions (sometimes called Tbox statements) and facts (sometimes called Abox statements). OTEs can depict classes, records, schemas, laws, even terms.
provide a useful mechanism for describing the functional architecture of BI systems and AIML processes. They include sensor classes both structured and unstructured, world description classes, explanatory classes, predictive classes, planning classes, action classes and monitoring classes. All ML algorithms have as a part of their signature the channel class of their inputs and outputs.
capture the relative belief rankings that are critical for cross channel validation. For example, if one image classifier detects a person while another detects that the space around the object detected as a person is less than 50 CM3 the channel weightings, if the space detector has higher confidence than the image detector, the volume detected in the one channel will be weighted more heavily than the image detected in the other and the cross channel validator will determine that the object can not be a ‘real’ person, but perhaps a figurine or an image of a person. Channel weightings are initialized when C.O.R.E. is run. But they can vary over time as the system learns.
define the objective function and looping mechanism by which C.O.R.E. compares the outputs of different channels. For example, if C.O.R.E.’s objective function is to maximize confidence, C.O.R.E. may try to compare the output of one channel with data already output from another channel (do two different image detectors at different positions relative to a field of interest agree?). And if no comparable data exists, C.O.R.E. may query whether there exists and algorithm that could be run on the same or different data to produce comparable output. And if not C.O.R.E. may query whether there exists an algorithm which could be trained to run on the same or different data and which if run would produce comparable data.
are composed of ontological types used to index both the inputs and outputs of AIML processes and whatever data or knowledge exists inside or outside the user’s system that may be considered relevant.
C.O.R.E. can be used in conjunction with any AIML environments like Tensor Flow or PyTorch and/or with existing orchestrators to slash the time and effort required to build train deploy and adapt large scale AIML implementations in a principled and clearly tractable fashion. C.O.R.E. empowers you to make your large scale AIML deployments more confident, more performant, more fault tolerant, more adaptive and more explainable than ever before.
Real meaning is an element of a world model. The thousands, tens or even hundreds of thousands of fields, columns, headers, graphs, and polygon classes etc. that make up the data dictionary or catalog of a large enterprise are labels for data about the world that is relevant to the enterprise: its customers, its products, its employees, its manufacturing processes, its operating regions and so on.
There is no guarantee simply given a collection of data labels and meta tags (even if fully joined in the database sense of the term), that they refer to a single model of the world that directly supports logical reasoning, machine learning and statistical models of all kinds i.e., the kind of inferencing that needs to occur on the data referenced by the catalog or dictionary.
That guarantee can only come by design. Critical design elements include