EP3523730A1 - Machine stochastique modulaire et procédé associé - Google Patents
Machine stochastique modulaire et procédé associéInfo
- Publication number
- EP3523730A1 EP3523730A1 EP17788170.3A EP17788170A EP3523730A1 EP 3523730 A1 EP3523730 A1 EP 3523730A1 EP 17788170 A EP17788170 A EP 17788170A EP 3523730 A1 EP3523730 A1 EP 3523730A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- stochastic
- module
- variable
- representation
- random
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
- G06F7/58—Random or pseudo-random number generators
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
- G06F7/58—Random or pseudo-random number generators
- G06F7/588—Random number generators, i.e. based on natural stochastic processes
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
Definitions
- the present invention relates to a modular stochastic machine.
- the invention also relates to a method for calculating corresponding probabilities.
- a modular stochastic machine finds applications in all fields involving many probability calculations. These include, for example, areas related to financial markets, economic modeling or weather forecasting.
- a modular stochastic machine solves problems generally solved by the Markov Chain Monte Carlo method.
- asset prices are represented as the solution of a stochastic differential equation.
- the valuation of derivatives on such assets is, in some cases, based on methods of approximating a solution of the stochastic differential equation by a discrete Markov chain combined with the Monte Carlo method.
- software instructions are stored in a separate memory of the microprocessor. It is able to execute, by reading in the memory, software instructions based on commands received.
- conventional microprocessor is meant a microprocessor processing information presented in a deterministic form, for example by means of bits corresponding to predefined digital words, in opposition to a representation in probabilistic form.
- the machine comprises at least one stochastic distribution module corresponding to a plurality of random variables, the stochastic distribution module being adapted to receive as input random values of variables specified among the plurality of random variables, and to output a representation of the probability distribution of at least one unspecified random variable conditioned by the values of specified random variables received as input.
- the machine also includes at least two stochastic variable modules, each stochastic variable module corresponding to a single random variable.
- Each stochastic variable module includes a stochastic multiplier and a stochastic proportional normalizer.
- a stochastic multiplier is adapted to receive as input representations of a first probability distribution and a second probability distribution of the same random variable, and to return a representation of the product probability distribution of said random variable .
- a stochastic proportional normalizer is adapted to receive as input a bit representation of a probability distribution, and to output a proportionally normalized representation of the received probability distribution, the proportionally normalized representation comprising more than one bit than the representation received as input.
- the machine comprises one or more of the following characteristics, taken separately or in any technically possible combination: the stochastic distribution module and the stochastic variable modules comprise stochastic bit generators, each stochastic bit generator being able to generate a stochastic bit stream whose probability of occurrence of bits at 1 is proportional to a value stored in the stochastic bit generator.
- the stochastic distribution module comprises as many stochastic bit generators as possible combinations of the different values of the plurality of random variables.
- the representation of the probability distribution returned at the output of the stochastic distribution module is generated from joint or conditional probability values stored in the stochastic distribution module.
- a representation of a probability distribution on a random variable is a time series of stochastic bit vectors called sample-vectors, each sample-vector comprising as many coordinates as the cardinal of the random variable, the number of bits at 1 in the time series for the coordinates of the same rank being proportional to the probability value corresponding to said rank.
- the stochastic bit vectors are value samples, a sample-value comprising only a single bit at 1.
- each stochastic variable module comprises a stochastic sampler capable of being activated and deactivated, a stochastic sampler being adapted to take input samples representing a probability distribution, and to return the value samples representing the same distribution of probabilities; probabilities.
- each stochastic variable module is adapted to operate according to a regime, called a deterministic regime, in which the stochastic variable module is capable of delivering a representation of a fixed value of the random variable.
- each stochastic variable module is connected to the stochastic distribution module by a data bus, each data bus comprising as many wires as the cardinal of the random variable corresponding to the stochastic variable module.
- the method comprises the steps of:
- the stochastic distribution module being adapted to receive, as input, values of random variables specified among the plurality of random variables, and to output a representing the probability distribution of at least one unspecified random variable conditioned by the specified random variable input values, and
- each stochastic variable module corresponding to a single random variable and comprising:
- a stochastic multiplier adapted to receive as input representations of a first probability distribution and a second probability distribution of the same random variable, and to output a representation of the product probability distribution of said variable random
- a stochastic proportional normalizer adapted to receive as input a bit representation of a probability distribution, and to output a proportionally normalized representation of the received probability distribution, the proportionally normalized representation comprising more than one bit of 1 that the representation received as input,
- a representation of a probability distribution on a random variable being a time series of stochastic bit vectors called sample-vectors, each vector-sample comprising as many coordinates as the cardinal of the random variable, the number of bits at 1 in the time series for the coordinates of the same rank being proportional to the probability value corresponding to said rank,
- the stochastic bit vectors being value samples, a sample-value comprising only a single bit at 1,
- each stochastic variable module comprising a stochastic sampler adapted to be activated and deactivated, a stochastic sampler being adapted to input vector samples representing a distribution of probabilities, and to return the value samples representing the same distribution of probabilities ,
- each stochastic variable module being adapted to operate according to a regime, said deterministic regime, in which the stochastic variable module is adapted to deliver a representation of a fixed value of the random variable, and
- the machine comprising as many stochastic variable modules as random variables on which the calculation is performed; activation or deactivation of a stochastic sampling of at least one stochastic variable module according to the calculation of probabilities to be performed;
- the method of calculation comprises one or more of the following characteristics, taken separately or in any technically possible combination:
- the step of providing a modular stochastic machine comprises a substep of determining a number of stochastic distribution modules.
- the step of providing a modular stochastic machine comprises a substep of determining a number of stochastic variable modules.
- the step of providing a modular stochastic machine comprises a substep of determining a number of data buses.
- the step of providing a modular stochastic machine comprises the implementation of an optimization algorithm for the number of stochastic distribution modules, the number of stochastic variable modules, the number of data buses and the number of stochastic distribution modules; arrangement of stochastic distribution modules, stochastic variable modules and data buses.
- FIG. 2 a schematic representation of an exemplary stochastic distribution module
- FIG. 3 a schematic representation of an example of a stochastic variable module
- FIG. 4 a schematic representation of an example of a stochastic multiplier
- FIG. 5 a schematic representation of an example of a stochastic proportional normalizer
- FIG. 6 a schematic representation of an example of a modular stochastic machine
- FIG. 7 an example of a configuration of the modular stochastic machine of FIG. 6 for performing a calculation of probabilities.
- a modular stochastic machine is an arrangement of components of different types.
- the choice of the type of component, the number of components as well as the arrangement of the components is a function of a calculation of probabilities to be achieved.
- FIG. 1 A set of components for making a modular stochastic machine is shown in FIG. 1
- the set of components comprises three types of components, namely a stochastic distribution module, a random variable module, and a data bus.
- a stochastic distribution module is designated by the acronym
- SD "Stochastic Distribution” a stochastic distribution module is designated by the acronym SV of the English “Stochastic Variable” and a data bus is designated by the acronym BD.
- ⁇ X denotes a discrete and finite random variable.
- nX denotes the cardinal of the random variable X, that is to say the number of possible values taken by the random variable X.
- ⁇ ⁇ , ⁇ 2 , ... ⁇ ⁇ denotes the values taken by the random variable X .
- P (X) denotes a probability distribution on the random variable X, that is to say the probability values as a function of the possible values of the random variable X.
- P (X 1 , X 2 , - X n ) denotes the random variable joint probability distribution X 1 , X 2 , - X n .
- information is conventionally represented in the form of bytes or, more generally, in the form of digital words formed of a fixed number of bits.
- a “stochastic bit stream” is a set of bits, each bit taking by definition the value 1 or the value 0.
- a bit taking the value 1 is designated by the expression "bit at 1".
- a stochastic bit stream is generated by a stochastic bit generator, hereinafter referred to as the Stochastic Bit Generator (SBG).
- SBG Stochastic Bit Generator
- An SBG is a physical device that includes a memory for storing a value, the SBG being able to generate a bit stream whose probability of being 1 is proportional to the stored value.
- a "bit ratio of 1" of a stochastic bit stream designating the number of bits at 1 divided by the total number of bits of the stream, the ratio of bits to 1 is proportional to the stored value.
- An SBG is, for example, realized using the stochastic properties of magnetic tunnel junctions.
- a magnetic tunnel junction is, in its simplest form, a thin insulating barrier, generally not exceeding 1 to 2 nanometers, between two conductive electrodes.
- the passage of an electric current is made by tunnel effect through this barrier, the tunnel effect designating the property that possesses a quantum object to cross a potential barrier even if the energy of the quantum object is less than the minimum energy required to cross this barrier.
- a super-paramagnetic tunnel junction oscillates stochastically between a parallel state and an antiparallel state. This results in random changes in the resistance of the nano-component that can be used to generate stochastic bits.
- a probability distribution P (X) on a random variable X is represented by a time series of stochastic bit vectors called vector samples.
- a vector comprises a set of coordinates each corresponding to a rank. Each coordinate is a bit.
- the coordinate of rank i is indifferently designated by the i-th coordinate and denotes indifferently the i-th bit or bit of rank i.
- Each vector sample has as many coordinates as possible values for the random variable.
- the vector samples comprise n coordinates and the coordinate of rank i corresponds to the situation in which the random variable X is susceptible to take the value x t .
- the coordinate of rank i is a bit at 1 when the random variable X takes the value x t .
- the ratio of bits to 1 in the time series for the coordinates of the same rank relative to the total number of bits is proportional to the probability of the value corresponding to said rank.
- the stochastic bit vectors of a time series representing a probability distribution P (X) are value samples.
- a sample-value includes only one bit at 1.
- bit ratios to 1 are respectively equal to the values of the represented probability distribution.
- a deterministic value that is to say the particular case where the random variable is known, is represented by a time series of identical sample-values where, for each sample-value, the coordinate corresponding to the value of the random variable is a bit at 1, the other coordinates being bits at 0.
- Table 3 Alternatively, the value samples are replaced by bits encoding a given piece of information, as in classical deterministic computing.
- An SD module corresponds to a plurality of random variables.
- plurality of random variables is meant at least two random variables.
- An SD module corresponds, for example to two random variables.
- An SD module corresponds, for example at least twenty random variables.
- An SD module is adapted to receive as input random variable values specified among the plurality of random variables, and to output a representation of the probability distribution of at least one unspecified random variable conditioned by the values of random variables. specified received as input.
- a module SD corresponding to the random variables X 1 , X 2 , ... X n is adapted to return to the output, for example, a representation of the following probability distribution:
- an SD module is, for example, adapted to receive as input n - 1 sample-values respectively representing the values
- an SD module includes a plurality of SBGs.
- the SD module comprises as many SBGs as possible combinations of the different values of the plurality of random variables.
- a SD module corresponding to two random variables X, Y each taking two values ( ⁇ , ⁇ ') and (y,') comprises four SBGs since there are four possible combinations that are (x, y), ( ⁇ , ⁇ '), (x', y) and (x ', y').
- the representation of the probability distribution returned at the output of the SD module is generated from joint or conditional probability values stored in the SD module.
- each SBG is capable of storing in the SBG memory a joint probability value on several random variables to generate a stochastic bit stream representative of the stored value.
- An SD module also includes a control system capable of activating SBG outputs corresponding to specified random variable values.
- the control system is able to select the outputs of the SBGs corresponding to probability values knowing in particular that the random variable Y ; is set to the value y t .
- the SBG outputs corresponding to the probability values knowing that the random variable Y is set to different values y j ⁇ i are deactivated and the corresponding SBGs then generate no stochastic bit stream.
- SBG when SBG is deactivated, it is able to generate only bits at 1. This has the advantage of generating a stream of "transparent" stochastic bits, especially when such a stochastic bit stream is found at the input of an AND logic gate.
- the SD module of FIG. 2 corresponds to two random variables X and Y and is capable of storing the probability distribution P (X, Y).
- the SD module includes nX x nY SBG. Since nX is 3 and nY is 2, the SD module includes six SBGs.
- the SBGs are denoted SBGi j , where i is an integer varying from 1 to 3, j being an integer varying from 1 to 2.
- the SBGi , 2 is specific to generate, once the output of the SBGi , 2 has been activated by the SBGi control system , 2 , a stochastic bit stream representative of the value 0.2.
- the SD module receives as input the following sample-value representative of the information
- the SBGu SBG 2, i and SBG 3 1 are, in this case, adapted to generate a time series of stochastic bit vectors whose rank 1 coordinates are formed by the flow of stochastic bits generated by the SBG ⁇ , whose row coordinates 2 are formed by the flow of bits generated by the stochastic SBG 2, i and whose row coordinate 3 are formed by the flow of bits generated by the stochastic SBG 3 1.
- An SV module corresponds to a single random variable.
- each SV module corresponds to a random variable chosen from among the plurality of random variables corresponding to an SD module.
- Each SV module is adapted to be connected to at least one SD module by a data bus adapted to ensure the transmission of stochastic bits. SD data buses are described in more detail later.
- an SV module is connected only to the SD module (s) corresponding to a plurality of random variables, the plurality of random variables containing the random variable corresponding to the SV module.
- An SV module includes a stochastic multiplier, hereafter referred to as the Stochastic Product Operator (SPO), and a stochastic proportional normalizer, hereafter referred to as the Stochastic Proportional Normalisator (SPN).
- SPO Stochastic Product Operator
- SPN Stochastic Proportional Normalisator
- an SV module also includes a stochastic sampler, hereinafter referred to as the Stochastic Proportional Sampler (SPS).
- SPS Stochastic Proportional Sampler
- a module SV corresponding to the random variable X taking the values x 1 , x 2 ,..., X n is adapted to receive as input representative samples-vectors of a plurality of distributions. of different probabilities P 1 (3 ⁇ 4P 2 (X), ..., P k (X) of the random variable X to output a representation in the form of a proportionally normalized time series of vector samples of a probability distribution P (X) of said random variable X.
- an SV module comprises an activated SPS
- the SV module is able to output a representation in the form of a proportionally normalized time series of sample-values of a probability distribution P (X) of said random variable X.
- P (X) probability distribution
- the representation of the probability distribution generated at the output of the SV module is the proportionally normalized representation of the probability distribution produced from the input probability distributions.
- the SV module is adapted to operate according to a regime, said deterministic regime, wherein the SV module delivers stochastic bit streams representative of a fixed value for a random variable.
- the random variable is said to be deterministic and the module SV is configured to operate in deterministic mode.
- the SV module is firstly able to receive a first and a second probability distribution on the random variable X which are respectively denoted P t (X) and P 2 ( X), and secondly to generate the distribution of product probabilities denoted P (X) probability distributions / ⁇ () and P 2 (X).
- the SV module receives more than two probability distributions.
- the SV module is, for example, a cascading arrangement of two SV modules as described below.
- the SPO is adapted to receive, as input, representations in the form of vector samples respectively of the first distribution of probabilities Pi () and of the second distribution of probabilities P 2 X) to return a representation of the distribution of product probabilities. of the same random variable X.
- the SPO comprises AND logic gates each corresponding to a possible value of the random variable X.
- An AND logic gate is a component implementing the AND function.
- the "AND" function is a logical operator of the Boolean algebra which takes as input two operands taking the TRUE value or the FALSE value and which returns the TRUE value if and only if both operands are TRUE.
- a logic gate is, for example, an electronic component comprising as input two terminals powered by either a zero voltage or a voltage of 5 V and which comprises an output having a voltage of zero or equal to 5 V.
- the voltage value 5 V is the TRUE value, that is, a 1 bit
- the zero voltage value is FALSE, that is, one bit at 0.
- the voltage output is 5 V if and only if the input voltages are both equal to 5 V.
- the SPO comprises three AND gates denoted AND 1; AND 2 , and AND 3 . I being an integer varying from 1 to 3, the AND gate, corresponding to the value xi of the random variable X.
- the SPO is able to generate a time series of sample-vectors which is represented in the form:
- An SPN is able to receive as input a bit representation of a probability distribution P (X), and to output a proportionally normalized representation of the received probability distribution.
- the proportionally normalized representation comprising more than one bit than the representation received at the input.
- the addition of bits to 1 is done proportionally, that is, the ratios between bit ratios at 1 for different ranks of the vector samples are unchanged.
- An SPN is particularly adapted to overcome the problem of time dilution.
- the output of the SPO is connected to the input of the SPN and the vector samples received at the input of the SPN are the vector samples returned at the output of the SPO.
- the represented SPN is able to generate representative vector samples of the probability distribution P (X) whose coordinates corresponding to the maximum probability of the probability distribution P (X) are all bits to 1.
- FIG. 5 illustrates the particular case of a random variable X taking three values x 1 , x 2 , x 3 and a probability distribution P (X) on the random variable X defined as follows:
- Each SBG corresponding to a value of the random variable X receives a stream of stochastic bits corresponding to the coordinates, in the received vector samples, corresponding to this same value of the random variable X.
- each SBGi, SBG 2 and SBG 3 comprises a counter denoted respectively Ci, C 2 and C 3 .
- Counters Ci, C 2 and C 3 all have the same capacity, denoted N max .
- Each counter d, C 2 and C 3 is able to increment upon receipt of a bit at 1.
- N3 ⁇ 4 the number of increments of a counter C is denoted by N3 ⁇ 4 and is called the content of the counter C 1.
- the SPN receiving representative vector samples of the probability distribution P (X) the counters d, C 2 and C 3 are incremented until a counter is filled.
- the SPN is able to stop incrementing the other counters.
- each SBG generates stochastic bit streams with the probability of having a bit equal to the content N3 ⁇ 4 of the counter C, divided by the capacity N max of the counters.
- the capacity N max of the counters is 200.
- bit ratios at 1 of the first, second and third coordinates are respectively:
- the output-vector samples SPN representative of the probability distribution P (X) comprise more bits than 1 SPN input vector samples representative of the same probability distribution P (X).
- the addition of bits to 1 thus has the advantage of reducing the time dilution.
- the SPN also comprises counters which, unlike the example described above, are able to decrement when their capacity is reached and the bit to be generated is 1.
- An SPS is suitable for receiving vector samples forming a proportionally normalized representation of a probability distribution to return the output of the sample-values representing the same distribution of probabilities.
- the SPS is input connected to the output of the SPN to receive vector samples forming a proportionally normalized representation of the probability distribution P (X) and to return to output representative sample values. the same probability distribution P (X).
- the SPS is able to be activated and deactivated.
- the SPS When the SPS is disabled, the SPS behaves like a passive component.
- the output of the SV module corresponds to the output of the SPN and the SV module returns proportionally normalized vector samples representative of the probability distribution produced representations of the probability distributions received at the input of the SV module.
- the SPS When the SPS is activated, the SPS "converts" the input vector samples to return sample values. In such a case, the SPS delivers sample values representative of sampled values of the random variable corresponding to the SV module.
- the SV module when the SPS is activated, the SV module returns to the output of the value samples comprising a single bit at 1 and for which the probability that the coefficients of rank i are bits at 1 is proportional to the probability that the random variable corresponding to the SV module takes the value x t .
- an SV module is the arrangement of an SPO, an SPN and an SPS
- the operation of an SV module results from the respective operations of an SPO, an SPN and an SPS described previously.
- the SV module does not operate in a deterministic regime.
- the SV module receives as input a set of vector samples representative of a plurality of different probability distributions of the same random variable.
- the input of the SV module being confused with the input of an SPO, the SPO returns at the output of the representative vector samples of the probability distribution product of the different probability distributions received at the input.
- the output of the SPO being connected to the input of an SPN, the SPN receives as input the vector samples generated by the SPO and which are representative of the probability distribution produced by the different probability distributions received at the input of the SV module.
- the SPO then performs a normalization of the representation received in the form of vector samples, the normalization of adding bits to 1 while keeping the ratios between the bit ratios at 1 identical to the ratios between the bit ratios at 1 of the representation provided at the input of the SV module.
- the action of the SPN has the effect of reducing the temporal dilution induced by the SPO.
- the SPS receives as input the proportionally normalized vector samples representative of the probability distribution produced from the different probability distributions received at the input of the SV module.
- the SPS is deactivated and the SPS behaves like a passive component.
- the output of the SV module corresponds to the output of the SPN.
- the SV module returns as output proportionally normalized vector samples representative of the probability distribution produced from the different probability distributions received at the input of the SV module.
- the SPS is activated and the SPS then converts the input vector samples into value samples.
- the output of the SV module corresponds to the output of the SPS and the SV module returns to the output sample-values representative of sampled values of the random variable corresponding to the SV module.
- the SV module operates in deterministic mode. In such a case, a value for the random variable corresponding to the SV module is set and the SV module returns output samples representative of the value of the random variable. 5 - The data buses
- BD data buses are able to connect SD modules and SV modules together.
- each data bus BD has as many wires as the cardinal of the random variable corresponding to the SV module.
- an SV module corresponding to the random variable X is connected to an SD module corresponding to the random variables X, Y by a data bus comprising nX son.
- the data buses BD are all identical and in fact each comprise the same number of wires.
- the data buses are suitable for being connected or disconnected. In other words, each data bus is able to be activated or inhibited so that no stochastic bit is transmitted.
- a connection or a disconnection of a data bus results, for example, from the activation of a function integrated in an SD module or an SV module.
- the data buses are formed by any type of physical medium capable of conveying binary information.
- the data buses are wires, prints on a circuit, or devices adapted to exchange electromagnetic waves or optical signals.
- a modular stochastic machine is capable of performing probability calculations on stochastic bit streams from at least two random variables.
- the modular stochastic machine comprises SBGs, at least one SD module, at least two SV modules, and BD data buses capable of connecting SD modules and SV modules together.
- the stochastic machine is described as modular because the stochastic machine is an assembly of SD modules, SV modules and BD data buses.
- the assembly of a modular stochastic machine depends on the calculations of probabilities to be carried out.
- a given assembly corresponds to a set of probability calculations and a particular configuration of the assembly is capable of performing a particular probability calculation among the set of probability calculations.
- FIG. 10 An example of a modular stochastic machine 10 is shown in FIG.
- the machine 10 comprises three SV modules and two SD modules.
- a first SV module is denoted SV 1; a second SV module is noted SV 2 and a third SV module is noted SV 3 .
- the first module SVi corresponds to a random variable O
- the second module SV 2 corresponds to a random variable D
- the third module SV 3 corresponds to a random variable Z.
- a first SD module is denoted SD ⁇ and a second SD module is denoted SD 2 .
- the first module SD ⁇ corresponds to the random variables 0, D and the second module SD 2 corresponds to the random variables 0, D, Z.
- the machine 10 comprises a first data bus BD 1; a second data bus BD 2 , a third data bus BD 3 , a fourth data bus BD 4 and a fifth data bus BD 5 .
- the first BD ⁇ data bus connects the first SV Module SD ⁇ ⁇ the first module.
- the second data bus BD 2 connects the first module SV ⁇ to the second module SD 2 .
- the third data bus BD 3 connects the second module SV 2 to the first module SD ⁇ .
- the fourth data bus BD 4 connects the second module SV 2 to the second module SD 2 .
- the fifth data bus BD 5 connects the third module SV 3 to the second module SD 2 .
- Fig. 7 corresponds to Fig. 6, in which the data buses BD 1; BD 2 ,
- BD 3 , BD 4 and BD 5 are shown differently to illustrate an example of calculation from machine 10 of FIG.
- an arrow represented in dashed line represents the transmission of samples-values representative of a sampling by the random variable corresponding to the SV module from which the arrow originates;
- an arrow represented in thick lines represents the transmission of sample-values, all identical, representative of the value fixed for a random variable
- an arrow represented in fine line represents the transmission of samples-vectors representing a distribution of probabilities.
- the random variables 0, D, Z describe an automatic driving assistance system for a motor vehicle.
- the space at the front of the vehicle is represented by a probabilistic occupation grid formed of different cells in which, for each cell, sensors are installed to calculate probabilities.
- Random variable 0 is binary and is 1 if the cell is busy.
- the random variable D is binary and is equal to 1 if there has been detection by the sensor.
- the random variable Z is a numerical random variable that corresponds to the value provided by a sensor.
- the problem to be solved is: what is the probability that a cell is occupied knowing the measurement of the sensor?
- the probability distribution P (0, D, Z) describes the operation of the sensor in cases of good detections, false detections and missed targets.
- the calculation of probabilities above is carried out using a modular stochastic machine according to the method of calculating probabilities.
- the method comprises a step of providing a modular stochastic machine.
- the provisioning step includes sub-steps.
- the number of SV modules is determined.
- the calculation to be performed involving three random variables 0, D, Z, three SV modules respectively corresponding to the three random variables 0, D, Z are provided.
- the number of SD modules is determined.
- the probability distributions P (0, D), P (0, D, Z) being known, two SD modules respectively corresponding to the random variables 0, D and 0, D, Z are provided.
- the number of data buses BD is determined.
- the third substep is implemented according to the rule that an SV module is connected only to SD module (s) corresponding to a plurality of random variables, the plurality of random variables containing the random variable corresponding to the SV module. In this case, five BD data buses are needed.
- the modular stochastic machine 10 of FIG. 6 is obtained.
- the provisioning step comprises the implementation of an algorithm for optimizing the number of SD modules, the number of SV modules, the number of data buses and the arrangement of the SD modules. , SV modules and data buses.
- the provisioning step is performed automatically by a software method consisting in traversing the structure of the decomposition of the desired probability distribution to produce an optimal arrangement of SD modules, SV modules and BD data bus.
- the first set of steps comprises substeps for activating or deactivating an SPS of at least one SV module according to the calculation of probabilities to be performed.
- the SPS of the second module SV 2 is activated.
- the second module SV 2 delivers in this case representative sample values of sampled values of the random variable D which are transmitted to the first module SD-i on the one hand and the second module SD 2 on the other hand, since the first and the second modules SD- ⁇ and SD 2 each correspond to the random variable D.
- the third and fourth data bus BD 3 , BD 4 connecting the second module SV 2 to the first module SD ⁇ on the one hand and the second module SD 2 on the other hand are then represented by dotted arrows.
- the first set of steps includes substeps of operating SV modules according to the deterministic regime.
- the third SV module 3 is configured to operate according to the deterministic regime.
- the third module SV 3 delivers in this case all identical sample values representative of the value z fixed for the random variable Z which are transmitted to the second module SD 2 since only the second module SD 2 corresponds to the random variable Z.
- the fifth data bus BD 5 connecting the third module SV 3 to the second module SD 2 is then represented by a thick line arrow.
- the first module SD- ⁇ receives from the second module SV 2 sampled values of the random variable D which is specified.
- the module SD ⁇ then generates a representation in the form of vector samples of a first distribution of probabilities Pi (O) which is expressed as follows:
- the representative vector samples of the first probability distribution Pi (O) are transmitted to the first module SVi by the first data bus BD 1; represented by an arrow in fine line.
- the second module SD 2 receives from the third module SV 3 identical value samples representative of the fixed value z for the random variable Z and also receives from the second module SV 2 representative sample values of sampled values of the random variable D which is specified.
- This operation is possible because of the reception of representative sample values of sampled values of the random variable D and of sample-values representative of the value z fixed for the random variable Z.
- the vector samples of the second probability distribution P 2 (0) are transmitted to the first module SN ⁇ by the second data bus BD 2 , represented by a thin line arrow.
- the first module SN ⁇ then performs the following operation:
- the first module SN ⁇ returns a proportionally normalized representation of the product of the probability distributions P t. 0) and P 2 (0) and provides the result to the problem.
- a modular stochastic machine thus offers the possibility of performing a wide variety of probability calculations.
- the computation speed is considerably increased compared to conventional microprocessors.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Algebra (AREA)
- Probability & Statistics with Applications (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Operations Research (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Complex Calculations (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| FR1601463A FR3057372B1 (fr) | 2016-10-10 | 2016-10-10 | Machine stochastique modulaire et procede associe |
| PCT/EP2017/075857 WO2018069349A1 (fr) | 2016-10-10 | 2017-10-10 | Machine stochastique modulaire et procédé associé |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| EP3523730A1 true EP3523730A1 (fr) | 2019-08-14 |
Family
ID=58314283
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| EP17788170.3A Withdrawn EP3523730A1 (fr) | 2016-10-10 | 2017-10-10 | Machine stochastique modulaire et procédé associé |
Country Status (7)
| Country | Link |
|---|---|
| US (1) | US20200050957A1 (fr) |
| EP (1) | EP3523730A1 (fr) |
| JP (1) | JP7048622B2 (fr) |
| KR (1) | KR102493657B1 (fr) |
| CN (1) | CN110235124B (fr) |
| FR (1) | FR3057372B1 (fr) |
| WO (1) | WO2018069349A1 (fr) |
Family Cites Families (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS5026069B1 (fr) * | 1970-02-28 | 1975-08-28 | ||
| SU940175A1 (ru) * | 1980-12-18 | 1982-06-30 | Московский Ордена Ленина И Ордена Октябрьской Революции Авиационный Институт Им.Серго Орджоникидзе | Веро тностное устройство дл анализа сетей |
| CA2166247A1 (fr) * | 1995-12-28 | 1997-06-29 | Ravi Shankar Ananth | Circuit de surveillance |
| WO2009111559A2 (fr) * | 2008-03-04 | 2009-09-11 | Massachusetts Institute Of Technology | Logique stochastique combinatoire |
| US8458114B2 (en) * | 2009-03-02 | 2013-06-04 | Analog Devices, Inc. | Analog computation using numerical representations with uncertainty |
| US20110010140A1 (en) * | 2009-07-13 | 2011-01-13 | Northrop Grumman Corporation | Probability Distribution Function Mapping Method |
| US8161329B2 (en) * | 2009-11-11 | 2012-04-17 | International Business Machines Corporation | Generating random sequences based on stochastic generative model having multiple random variates |
| US8645286B2 (en) * | 2010-02-23 | 2014-02-04 | Prior Knowledge, Inc. | Configurable circuitry for solving stochastic problems |
| EP3125161A4 (fr) * | 2014-03-25 | 2017-12-06 | Hitachi, Ltd. | Système d'inférence probabiliste |
| CN105913118B (zh) * | 2015-12-09 | 2019-06-04 | 上海大学 | 一种基于概率计算的人工神经网络硬件实现装置 |
| US10520975B2 (en) * | 2016-03-03 | 2019-12-31 | Regents Of The University Of Minnesota | Polysynchronous stochastic circuits |
-
2016
- 2016-10-10 FR FR1601463A patent/FR3057372B1/fr active Active
-
2017
- 2017-10-10 KR KR1020197013606A patent/KR102493657B1/ko active Active
- 2017-10-10 EP EP17788170.3A patent/EP3523730A1/fr not_active Withdrawn
- 2017-10-10 WO PCT/EP2017/075857 patent/WO2018069349A1/fr not_active Ceased
- 2017-10-10 JP JP2019540700A patent/JP7048622B2/ja not_active Expired - Fee Related
- 2017-10-10 US US16/340,636 patent/US20200050957A1/en not_active Abandoned
- 2017-10-10 CN CN201780062504.8A patent/CN110235124B/zh not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| KR20190121288A (ko) | 2019-10-25 |
| FR3057372B1 (fr) | 2022-05-20 |
| WO2018069349A1 (fr) | 2018-04-19 |
| CN110235124A (zh) | 2019-09-13 |
| JP2019537173A (ja) | 2019-12-19 |
| FR3057372A1 (fr) | 2018-04-13 |
| KR102493657B1 (ko) | 2023-01-31 |
| CN110235124B (zh) | 2023-09-26 |
| JP7048622B2 (ja) | 2022-04-05 |
| US20200050957A1 (en) | 2020-02-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP4137999B1 (fr) | Circuit neuromorphique à base de cellules rram 2t2r | |
| FR2644264A1 (fr) | Reseau neuronal analogique programmable | |
| WO2010066563A1 (fr) | Dispositif a architecture pipeline de flot de donnees pour la reconnaissance et la localisation d'objets dans une image par balayage de fenetres de detection | |
| WO2011121218A1 (fr) | Générateur de séquences chaotiques et système de génération correspondant | |
| WO2017001212A1 (fr) | Microprocesseur parallèle stochastique | |
| FR2935059A1 (fr) | Procede de detection d'anomalies dans un circuit de cryptographie protege par logique differentielle et circuit mettant en oeuvre un tel procede | |
| BE898544R (fr) | Calculateur associatif permettant une multiplication rapide. | |
| EP3523730A1 (fr) | Machine stochastique modulaire et procédé associé | |
| FR3066871A1 (fr) | Dispositif logique de detection de fautes | |
| EP4576094B1 (fr) | Circuit électronique à base de cellules rram 2t2r avec précision améliorée | |
| EP3343427B1 (fr) | Procédé et dispositif de gestion de la consommation en courant d'un module intégré | |
| EP2315388B1 (fr) | Procédé sécurisé de calcul cryptographique et composant électronique correspondant. | |
| EP0283393A1 (fr) | Dispositif de calcul binaire | |
| FR2499788A1 (fr) | Circuit generateur d'impulsions utilisant une source de courant et ensemble de memorisation utilisant un tel circuit | |
| EP3511870B1 (fr) | Procédés d'apprentissage de paramètres d'un réseau de neurones à convolution, et de classification d'une donnée d'entrée | |
| WO2016050595A1 (fr) | Procede de codage d'un signal reel en un signal quantifie | |
| FR3059805A1 (fr) | Dispositif et procede de modulation, synapse artificielle comportant ledit dispositif de modulation, procede de plasticite a court terme dans un reseau de neurones artificiels comportant ladite synapse artificielle | |
| FR3073645A1 (fr) | Procede de modification aleatoire du profil de consommation d'un circuit logique, et dispositif associe | |
| EP4557166A1 (fr) | Procédé et dispositif de détermination d'appartenance aux données d'apprentissage utilisées pour l'entraînement d'un modèle de classification de données par apprentissage machine | |
| FR2928769A1 (fr) | Dispositif permettant l'utilisation d'un composant programmable dans un environnement radiatif naturel | |
| FR2773652A1 (fr) | Circuit de generation d'un signal d'activation commande | |
| EP2329596B1 (fr) | Module reconfigurable et procede de mise en oeuvre de ce module reconfigurable pour realiser des operations morphologiques | |
| EP3622445A1 (fr) | Procede, mise en oeuvre par ordinateur, de recherche de regles d'association dans une base de donnees | |
| EP4436101A1 (fr) | Verification d'une donnee | |
| EP3516769A1 (fr) | Dispositif de génération d'un signal porteur continu à partir d'un signal impulsionnel de référence |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
| PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
| 17P | Request for examination filed |
Effective date: 20190430 |
|
| AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
| AX | Request for extension of the european patent |
Extension state: BA ME |
|
| DAV | Request for validation of the european patent (deleted) | ||
| DAX | Request for extension of the european patent (deleted) | ||
| RAP1 | Party data changed (applicant data changed or rights of an application transferred) |
Owner name: SORBONNE UNIVERSITE Owner name: CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE Owner name: THALES Owner name: PROBAYES Owner name: UNIVERSITE GRENOBLE ALPES Owner name: UNIVERSITE PARIS-SACLAY |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: EXAMINATION IS IN PROGRESS |
|
| 17Q | First examination report despatched |
Effective date: 20211111 |
|
| RAP3 | Party data changed (applicant data changed or rights of an application transferred) |
Owner name: UNIVERSITE GRENOBLE ALPES Owner name: UNIVERSITE PARIS-SACLAY Owner name: SORBONNE UNIVERSITE Owner name: THALES Owner name: PROBAYES Owner name: CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN |
|
| 18D | Application deemed to be withdrawn |
Effective date: 20250712 |