Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
  • G06N5/00—Computing arrangements using knowledge-based models
  • G06N5/04—Inference or reasoning models
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
  • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
  • G06F16/28—Databases characterised by their database models, e.g. relational or object models
  • G06F16/283—Multi-dimensional databases or data warehouses, e.g. MOLAP or ROLAP
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
  • G06N5/00—Computing arrangements using knowledge-based models
  • G06N5/02—Knowledge representation; Symbolic representation
  • G06N5/022—Knowledge engineering; Knowledge acquisition

Definitions

• the present invention relates to a data processing device designed for programming expert systems or artificial intelligence systems simulating human reasoning by being able to implement induction reasoning.
• a dedicated data processing device including statistical processing, data mining ("datamining" in the English terminology), the design of decision support tools, diagnostics, the prediction or approximation, the design of simulators, machine learning systems or learning aids and generally the design of situation analysis systems or situation analysis.
• a processor takes bits sequences into input, transforms them through series of logic gates, and derives another sequence of bits at the output. It is the series of logic gates that determines the calculation that the processor makes.
• This series predetermined in the program, depends on a user-impregnated model according to the rules of formal logic, and based on known situations.
• These models can take extremely varied forms, from quantitative models for market finance to mechanical models for engineering.
• the principle z, genera! consists in analyzing real phenomena to predict results from the application of one or more models at a given level of approximation.
• neural networks have been developed which have proven to be more efficient than conventional processor architectures, particularly in signal and information processing (telecommunications, finance, meteorology), statistical processing (marketing), ranking and recognition of form
• the present invention proposes an alternative data processing device which, through the observation of situations, makes it possible to analyze data, and to predict other situations, without being dependent on a model or structure of data. physical or logical implementation.
• situation is meant here a more or less complex and more or less vague information describing a particular action or state.
• the situation arises from an observation by a human being or a machine and can be as real as imaginary,
• the present invention proposes a data processing device, its architecture being able to be physical or logical and forming a situational analysis processor.
• an inference engine implementing, in parallel or in series, inference rules grouped together in libraries, said rules being programmed to regenerate a sequence comprising at least one segment of the memory, by creating, deleting or modifying them at their implementing in at least one memory space at least one datum and / or at least one attribute value and / or at least one connectivity link,
• resource allocation means activating or deactivating the inference rules according to a priority rule
• data extraction means identified by programming and / or by at least one chosen attribute value and / or at least one link of connectedness chosen; and / or located in one or more selected segments of the memory, to one or more outgoing information flows.
• the memory is organized into at least three dimensions, including at least one temporal dimension associating with the data one or more attributes making it possible to date the storage of the data, an idiosyncratic dimension associating with the data one or more attributes making it possible to determine the relative specificity of the data , a conceptual dimension associating with the data one or more attributes allowing to hierarchize the data according to levels of abstraction;
• the first space of the memory containing attributes of the data is constituted by even levels identified as levels of situations, and the second space of the memory containing relationships between data is constituted by odd levels identified as levels of connectivity;
• At least one inference rule implemented by the inference engine induces a level of 2k + 1 connectivity abstraction from a situation 2k abstraction level and / or a 2k + abstraction level. 2 of hypersituation from a level of abstraction 2k + 1 of connectivity;
• the priority rule implemented by the resource allocation means is a function of the hierarchy of the segments of memory to be regenerated, of predefined parameters of the inference rules and of the interval separating the activation occurrences of the rules of inference;
• the memory is connected to a mass storage space in which segments can be saved
• the data extraction means choose singular situations, situations that have varied, and plausible forecast situations.
• This innovative device opens up great prospects in many economic fields using tools of analysis, prediction and simulation. It is able to receive one or more situational flows as input, to extract situations from them, to distinguish the important elements and to continuously apply treatments to it, to detect phenomena and to foresee evolutions of solutions, in particular by induction.
• this device is not constrained by a model or an architecture of impiementation, and therefore adapts permanently. It is able, like the human brain, to focus on the essentials by managing its resources. Finally, its possibilities appear much more universal than those of any current expert system, partitioned in a particular field.
• the device according to the invention is asynchronous, non-deterministic, and proves neither connectionist nor computational since its power is both in its structure and its logic. It accepts sequential operation as parallel.
• the present invention proposes a physical implementation of a situational analysis processor, namely an information processing system comprising at least one processing unit, at least one memory unit, a first interface with at least one stream. of incoming information and a second interface with at least one outgoing information flow, characterized in that it implements the architecture of the processing device according to the first aspect of the invention.
• said processing unit comprises at least one core-core processor
• said processing unit impales a neural network structure.
• the present invention provides a computer program product comprising program code instructions which, when executed, implement the architecture of the device according to the first aspect of the invention.
• FIG. 1 is a diagram of an embodiment of a device according to the invention.
• FIG. 2 is a diagram of an advantageous embodiment of a situational memory used by the invention.
• FIG. 3 is a diagram showing how the device according to the invention can integrate into a computing environment. DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT
• the basic unit in situational analysis is the situation. This can be more or less true (for "past” situations), more or less plausible (for "future” situations).
• IDcolor (mouse, 001, black); black COLOR attribute
• OBJ Object
• HUM Human
• Situations may not be coded in natural language to escape the ambiguities inherent in common vocabularies.
• the syntax used to describe situations is arbitrary but must be adapted to computer tools.
• the language chosen is a metaianguage, in particular the XML language (extensible markup language or Extensible Markup Language in English terminology).
• the processing device 1 according to the invention is capable of working on the situations and situatrons described in this form.
• the memory 10 is a situational memory, structured to retain the data necessary for the situational analysis. This memory 10 is organized in two spaces.
• the first space contains the attributes of the stored data, that is to say directly situations, situatrons, or groups of situations or situatrons.
• the second space contains links of connectivity between the stored data, that is to say the relations of connectivity between the situations and situatrons of the first space.
• Connectivity links will be established when situatrons are present in common situations, or share common indices. Thus, in the previous example, being connected to a computer creates a connection link between different mice.
• This second space is a dual space of the first, in fact it is a space of induced functions.
• the situatrons, the attributes of a situatron and the links of connectedness are thus induced by observation and thus created as information is acquired or when unobserved situations are induced. may be deleted when they become obsolete, and may be changed if they evolve or become inaccurate. Their types are not predefined, and can adapt to any circumstance.
• the memory 10 has a specific organization, an example of which is shown in FIG. 2. This organization follows one or more dimensions, each associated with attributes, and divided into one or more hierarchical segments. Each segment is characterized by one or more attribute values or ranges of particular attribute values. Data is populated by creating, deleting or modifying the attribute values characterizing these segments, which defines them a place in the memory 10.
• these dimensions are three in the preferred embodiment. This results in an organization in the form of an assembly of cubes 101, each cube being characterized by a unique combination of one segment by dimension.
• the first dimension is a temporal dimension, and makes it possible to date the storage of the data.
• the date of the memorization of the data remains distinct from the date of the facts carried by the situation, the latter being able to be the object of one or several distinct attributes, but not related to those of the temporal dimension .
• the segments of the temporal dimension correspond to periods. It extends on the X axis in Figure 2.
• information about the battle of Marignan can be stored in memory as follows:
• IDC country, 1000, "Italy”
• Segment 5 - Faraway rest Then, on May 7, 2010, the situational scene on the Battle of Marignan, observed on April 30, 2010, will be in segment 4.
• IDC name, army, "Swiss”
• This new information is housed in Segment 2 and brings the event situation 1001 fully back to Segment 2.
• the second dimension is an idiosyncratic dimension.
• idiosyncratic we mean “who constitutes the specificity, the singularity of someone or something”.
• This dimension is indeed related to the level of incidence, which generally encompasses any criterion making a remarkable data, and in particular the singularity or the relative specificity of the data.
• an "orange mouse” will be a situation with a strong singularity in a universe where we usually observe white mice. This dimension extends on the Y axis in Figure 2.
• the third dimension is a conceptual dimension. It associates with the data one or more attributes allowing to hierarchize the data according to levels of abstraction. Each level of abstraction defines a segment of the conceptual dimension, which is called an abstraction plane 100. These planes 100 extend over the Z axis and represent increasingly evolved forms of situational knowledge as they progress. as one gains the upper layers.
• the duality of the memory is implemented in the form of an alternation of situations pics 100a and connectivity frames 100b: the set of even levels 100a is a partition of the situation space, and the set of odd levels 100b is a partition of the connectivity space,
• Level 0 contains situations, situatrons and groups of situations and situatrons.
• Level 1 contains relations of connectivity between situations and situatrons of the first level. Continuing the previous example, it will be for the first series of situations and situatrons, links of connectivity between: SourisOOl and ordiOOl; OrdiOOl and userOOl; UserOOl and mouseOOl (this last link being induced);
• Level 2 contains hypersituations, hypersituations and hypersituations and hypersituations induced by an analysis of second level connectivity. From the connectivity link of the lower level, it is possible to induce a hypersituatron MOUSE or COMPUTER or
• Level 3 hyperconnectivity which corresponds to the links of connectedness established between hypersituatrons and hypersituations of level 2;
• the memory 10 is connected to a mass storage space 16 in which all or part of this memory can be saved.
• a mass storage space 16 in which all or part of this memory can be saved.
• memory cubes will be saved.
• This interaction with a mass storage space is advantageously implemented in the form of a permutation ("swapping" in English terminology) by the skilled person. It allows you to work on several contexts from one to the other, or to work with situational memories larger than the addressability of the processor.
• the device 1 comprises means 1 1 for integrating one or more incoming information flows.
• These integration means 1 1 (or situational integrator) recognize in the information flow or flows of situations approaching the situations already present in the memory 10. New situations and new situations are produced from the flow data. These data are structured and their attributes defined in accordance with the organization of the memory 10.
• the situational integrator could be a module that reads lists of computer mouse sales presented as XML feeds on the input ports. He interprets the flows and separates the information for:
• the integrator advantageously loads the situations and situatrons them in:
• the memory 10 contains data.
• An inference engine 14 implements processes that will apply to all or part of the memory 10. These processes are loaded from a library 15 and are called inference rules 141. At each rule launch, a sequence comprising at least one segment of memory will be scanned, and the associated memory regenerated. By regenerate means here create, delete or modify at least one data item and / or at least one attribute value and / or at least one connectivity link in at least one space of the memory.
• a treatment that can be performed by the inference engine 100 on an abstraction level 100 can be applied to levels of the same parity.
• the rules of inference include all the treatments that may be interesting to apply.
• the choice of these rules 141 depends on the desired effects, and is to the appreciation of the person skilled in the art according to the field of application for which the architecture is implemented.
• Temporalization consists of dragging the situations and the situatrons concerned from one temporal segment to another.
• the "reduction” consists of removing situational indices as they slide into segments of the past. This operation is based on the observed use of the indices.
• - “Aggregation” consists of bringing together several situatrons that no longer need to be discerned into a single aggregate inheriting the main properties of the situatrons and replacing them with the aggregate whenever possible.
• Prediction consists in applying situational phenomena to observed situations in order to predict future situations.
• the "plausabiiisation” is to prefer in the situations envisaged those related to past situations that have been realized.
• a rule is used to induce a level of 2k + 1 connectivity abstraction from a situation abstraction level 2k. and a rule making it possible to induce a level of 2k + 2 hypersituation abstraction from a level of 2k + 1 connectivity abstraction, that is, to generate a higher level from the previous one.
• the new connectivity links can be deduced according to the rules of connectivity declared in the library, induced by the observation and generalization of past connections, statistics according to the connectivities distributed in situational memory, probabilistic on Bayesian subsets, genetic by mutation of previous connectivity circles, etc.
• the second of these two rules is more simply a generalization of a group of situations and situatrons from a number of cases of connectivity greater or less depending on the level of importance.
• Means 13 of resource allocation manage the distribution of processing times between different treatments. For this, they have a law of priority according to which they activate or deactivate the rules of inference. There must never be too many active rules at once, otherwise the capacity of the system will be exceeded.
• the priority law allows both a spatial hierarchy of the memory cubes 101 to be processed, and a temporal hierarchy of the rules 141 to be launched.
• this distribution is carried out according to a so-called "essentiality" criterion.
• essentiality we calculate the criterion of essentiality for a rule and we launch the rule or rules for which it is maximum if they are not yet activated.
• the criterion of essentiality on the memory cubes 101 is then calculated, the sequence is generated by sorting the segments according to a decreasing criticality, and the cubes are processed according to this rule. sequence, which stops when the rule is deactivated.
• the essentiality of the memory cubes depends on predefined parameters inherent to the current rule. For example the larger groups of situations will be preferred for rules to establish connectivity links. In addition, advantageously, the most abstract, the most recent, and the most singular memory cubes will have the maximum essentiality, also by analogy with human reflection.
• the essentiality of the rules also depends on predefined parameters ranking the rules, with equal context. But above all, the essentiality of a rule of inference depends on the interval separating its activation occurrences. The longer the rule has been inactive, the more its essentiality increases. This allows you to perform all kinds of treatment regularly.
• the concentrator can be designed as a treatment stack in which the most essential treatments are at the top of the stack, the least essential at the bottom. The last treatment arrived is at the end of the stack with the essentiality of the "present" segment.
• the concentrator scans the entire stack at a fixed frequency and recalculates each cycle the level of essentiality of the treatments according to different parameters. For example, the following parameters can be retained: - D: their waiting time. At each passage of the concentrator, D is incremented by a fixed amount, for example 0.25;
• T the temporality of the associated memory cube. T takes a value representing the order of the temporal segment (from 1 to N, 1 characterizing the oldest segment)
• I the incidence of the associated memory cube. I takes a value representing the order of the incidence segment (from 1 to N, 1 characterizing the segment of lesser incidence);
• A the level of abstraction of the associated memory cube.
• A takes a value representing the order of the level of abstraction (from 1 to N, 1 characterizing the lowest level).
• each treatment can for example be defined as unique or permanent: a single treatment will come out of the stack when it is realized while a permanent treatment is placed at the tail of the stack and will be renewed. It can also be set as active or deferred: active processing is supported by the hub while delayed processing is waiting in the stack for its time, the hub ignoring delayed processing that has not completed. Permanent deferred processing can also be defined as to be launched on a fixed date or at a fixed frequency.
• Developer Means 12 for extracting data make it possible to deliver an information flow by means of which the results of the numerous processing cycles carried out by the inference engine 14 can be collected on the memory 10.
• the means 12 identify data within the memory 10 according to different non-exclusive principles of each other.
• the targeted data can be defined according to attribute values, or their belonging to a related group.
• the extraction means listen to one or more memory cubes.
• the extraction criteria can also be predefined arbitrarily by programming.
• the information flow composed by the data extraction means 12 is composed of:
• the extraction means are able to select singular situations (new, abnormal) or situations that have varied (situational phenomena);
• the invention proposes an information processing system.
• This is a physical implementation of the architecture of the processing device 1 according to the first aspect of the invention.
• This information processing system is for example in the form of a co-processor.
• the processing system comprises a processing unit, dedicated to the execution of the rules of inference.
• a processing unit dedicated to the execution of the rules of inference.
• it is a microprocessor microprocessor. Indeed, as we saw the situation analysis allows the parallel processing of data, with the ability to simultaneously activate multiple rules.
• this processing unit can implement a neural network around several "cells" elementary treatment. Such a network is particularly suitable for learning.
• the processing system also comprises a memory unit, for example a random access memory of the RAM type. This is where the memory 10 of the situational architecture will be housed.
• the system also comprises a mass memory 16, such as a hard drive, to perform the data swapping described previously.
• Two interfaces are also necessary, one for the incoming information flow or flows to the integration means 11, and the other for the outgoing information flow (s) provided by the data extraction means (12). data.
• a situational analysis co-processor can be purely logical, and emulated as a program of a workstation.
• the invention relates to a computer program product comprising program code instructions which, when executed, implement the previously described situational analysis processor architecture.
• the integration means 1 1 can be particularly adapted to the internet through which the data to be recovered are abundant. For example, data can be generated by browsing one or more customers in a store. It may be interesting to understand the situations in which they place themselves during a purchase, or precisely a non-purchase.
• the results of the situational analysis obtained via the data extraction means 12 can be transmitted to the client himself, for example to help him to make his choice: if he looks for a computer mouse, the analysis of the situations of the previous buyers can allow him to determine a suitable product.
• the data can of course also be transmitted and exploited by experts, especially alerts on abnormal situations.

Applications Claiming Priority (2)

Publications (1)

Family

ID=43532608

Family Applications (1)

Country Status (5)

Families Citing this family (9)

Family Cites Families (19)

• 2010
  • 2010-07-13 FR FR1055703A patent/FR2962823B1/fr not_active Expired - Fee Related
• 2011
  • 2011-07-13 EP EP11730694.4A patent/EP2593909A1/de not_active Withdrawn
  • 2011-07-13 CN CN201180044029.4A patent/CN103262103B/zh not_active Expired - Fee Related
  • 2011-07-13 WO PCT/EP2011/061910 patent/WO2012007489A1/fr active Application Filing
  • 2011-07-13 US US13/809,564 patent/US9349097B2/en active Active

Non-Patent Citations (2)

Also Published As

Similar Documents

Legal Events