JP2018523175A5 - - Google Patents

Description

Spatial calculations on computer equipment

The present invention relates to an operating system and method of a computing environment in a computer device, and specifically to a system and method constituting a computer device that performs non-Turing-limited calculations.

Calculations that can be understood as a string of events that are executed in hardware or are not specified separately are basically irreversible in each step, like time. Sequential computations, by their nature, consume information, have concurrency, and systematically execute multiple copies of one or more sequential processes. These facts are based on the mathematical theory of Alan Turing and Alaude Shannon.

The present invention presents a new kind of calculation method, spatial calculation, which is the opposite concept to temporal calculation. Spatial calculations are reversible (like waves) and produce information. Spatial calculations rely on new mathematics that is fundamentally different from existing sequential (parallel) calculations.

As a result, there is no concept of "time" in spatial calculations, even with many changes. Such changes can be expressed as the progression of complex waveforms (indicating the range of action of the calculation) and the propagation behavior of discrete parallel bit flips (see Perseval's Identity (1799)). ..

According to Parseval's identity, the Fourier decomposition of F is the projection of the function F into the n-dimensional orthogonal space. This is a generalization of the Pythagorean theory to n dimensions. In the n-dimensional coordinate system, the current value of F'applies to the hyper-hypotenuse in the n-dimensional hypercube, and when projected, several hyper-hypotenuse according to each dimension that is a constituent factor. It is separated into pieces (pieces).

If you try to make an n-dimensional cube, it starts with a right triangle whose sides are a and b. After reflecting this triangle on the hypotenuse to make a rectangle with both sides a, b, area ab, and diagonal line √ (a ² + b ² ), lift the rectangle vertically (length c) to create a three-dimensional (3D) cube abc. create. The diagonal line of this is d = √ (a ² + b ² + c ² ), and the sum of such squares is continued to form a 4D cube, a 5D cube, and the like.

At the same time, returning to the first right-angled triangle, it can be expressed by a = d cos θ and b = d sin θ, where θ is the angle between a and the oblique side. With such a sine-cosine value, a, b, c ,. .. .. Is replaced with a dimension for the n-dimensional hypotenuse (which is the current value of the function F starting from the projected value), resulting in the Fourier world, which is the sum of the sine and the cosine.

Therefore, the world of waveforms and the world of Cartesian coordinate systems are the same world, and the world connected to calculations is the latter. Such a connection causes each dimension to correspond to the state value of some processes, at which time all such processes a, b, c ,. .. .. , Ab, ac ,. .. .. , Abc ,. .. .. Is conceptually independent (orthogonal), but the interactions are free and simultaneous. In addition, a, b, c ,., Which are each of such processes. .. .. , Ab, ac ,. .. .. , Abc ,. .. .. Is in one of two possible states at any particular time. In the geometric (Clifford) algebra over Z3 = {0, 1, -1} used here, the 1-vector is processed with ± 1 which is one-state bit, and the m-vector has m-state bits. Write a + b for the parallel processes a and b, and write ab when a and b act. ab is a process with two state bits and a ± 1 appearance. Analysis of process-state progression processes in such a coordinate system shows that some or all of the processes change state at some frequencies, and interacting processes (m-vectors) at lower frequencies than the constituent processes. change. Such a process is considered to be separated into frequency bands according to the number of state bits (m), but the high frequency band is a more frequently changing individual one-dimensional process that presents short-term high resolution data for the system. a, b, c ,. .. The low frequency band, including, corresponds to the long-term symmetry and global evolution represented by the dimension increase (ab, abc, abcd) process.

According to Parseval's identity, the frequency band corresponds to the algebraic equation, which corresponds to the set of interactions. It can be said that the Fourier decomposition of this coordinate system is to execute the cross sum of such frequency bands. The relationship between such cross-sum Fourier bands to the frequency of individual process changes in the coordinate system is the relationship between sensory feeling or experience and perception, eg, the sum of red experiences detected by an individual's retinal cells. The relationship between and the optical frequency can be analyzed.

The spatial calculation S has the following six characteristics.

1. 1. Spatial computations are decentralized, which is a synchronous and independent system consisting of one or more processes, but with a large number of interacting entities, with little / no central control. It means that it shows no consistent global behavior. It has been proven that accurate technical standards for the method of designing and constructing a distributed system are difficult to specify, and the present invention solves the above problems.

2. 2. Spatial computations are self-organizing, which is a component of spatial computations in relation to several methods and required inputs given by the surrounding environment (called "peripheries"), including spatial computations. It means assembling such an input into a consistent entity of individual units (ie, other entities, individual processes or combinations thereof) that interact with the surroundings in a stable manner over the course of. The method of designing self-organizing calculations is a research area of current interest directed by the present invention.

3. 3. Spatial calculations are hierarchical, which involves the creation of new individual units, where the self-organizing ones indicate a combination of initial units-a combination of such combinations-and such a new unit is a self-organizing method. Means to belong to. Hierarchies are technical tools that have been proven to be versatile, control conceptual complexity, and have been found to be used in the field from programming languages to databases of communication protocols.

4. Spatial computations are not Turing-restrictive, which means that a universal Turing machine restricted to sequential computations (including parallel) cannot simulate actual spatial computations. The coin demonstration shows this in an easy-to-understand manner. Turing theory argues against this here because it shows that all calculations are basically sequential.

5. Spatial calculations are meaningless because they cannot grow without being connected to the periphery and interacting with the periphery. This world is perfect, in contrast to the sequential calculation of the π value, which is probably the calculation of the true meaning.

6. Spatial calculations utilize broadcast / listen communication discipline. That is, the spatial calculation broadcasts its own state and listens to (reacts to) other broadcast states. This is because otherwise the request / response discipline is basically functional to the character, and y = f (x), because f acts on x and its This is because it is a request to respond to the result. However, since y of one person is x of another person, z = g (y) is also possible. However, at this time, g (y) = g (f (x)), and if z is to be obtained, g must be obtained after finding f first, which is basically sequential. Targeted and time-based. Today's computer systems are organized this way. On the other hand, for spatial calculations, the steps of a sequential process are assembled each time, which will be described later. Most ignore the broadcast / listen protocol because modern technology does not fit the y = f (x) (ie, sequential) organizational paradigm.

The following coin demonstration shows the basic principle.

Act I. A man is standing in front of you with his hands behind him. This man shows you the hand holding the coin and then hides it behind him. A little later, this guy shows you the same hand holding what is found on the same coin, then hides that hand again and asks, "How many coins do I have?"

The best answer at this point is "one or more", which is state 1 = "one coin" and state 2 = one piece of information and two possible but mutually exclusive states. Imply that you are looking for "one or more coins".

What should be decided is that if it is one coin, X or Y is possible, and the answer can be obtained with exactly one piece of information. In other words, when such a decision can be made, we have received one piece of information for this fact.

Act II. Now, this guy is out and holding two of the same coins.

Assuming that the two coins are the same as the coins shown earlier in all respects, we now know that there are two coins, that is, receive one piece of information and resolve the ambiguity. And now it has reached a dramatic peak.

Act III. The man asks, "Where did this information come from?"

This information is due to the simultaneous existence (simultaneous occurrence) of two coins, the observed fact that when two processes, each state of which is each coin, are in said state, do not exclude each other's existence. To encrypt. Moreover, since such states are simultaneous and independent, they cannot be distinguished from the viewpoint of time.

Therefore, there is information about the environment that cannot be acquired sequentially, and therefore the Turing machine cannot simulate true simultaneity. Can the given states of process a co-exist with the given states of process b, or can they exclude each other's existence? This is a fundamental division. More officially, of course, a + a'= 0 and can be written as b + b'= 0 ('= not = minus), which means that (process state) a excludes (process state) a', and ( It means that process state) b excludes (process state) b'. Simultaneous existence can capture these two expressions together, and the first of the two ways to show the result is:
(A + b') + (a'+ b) = 0.

This means that it is neither A nor B as the general law of excluded middle (excluded middle). Such a principle can be applied to the present invention, and the existing 1/0 or 1/1'division can be encrypted. The second is
(A + b) + (a'+ b') = 0.

This is either or neither of the two superpositions. That is, technically, there is no superposition of a / a'and b / b'as a one-dimensional process, and rather these are exclusion categories. The first superposition appears at the second level of the process (ie, ab) through segmentation exclusion vs. co-occurrence.

Coin demonstrations show that two simultaneous occurrences, such as a + b, have one piece of information due to the existence itself. Simultaneous relationships are structural or spatial in nature. Such spatial information (as opposed to temporal information) eventually forms the structure and content of the Fourier band (like all 2-vectors). As described above, {xy}, {xyz}, {wxyz}, which are a set of m-vectors. .. .. Is a continuous system boundary x + y + z +. .. .. In, as a subordinate meaning of simultaneous flux, it forms a simultaneous structural and basic decomposition component of flux into a stable and metastable process hierarchy.

Act IV. This man puts his hands forward and one is free, but the other has a coin. Hold your hand and hide it behind, then pull the back of your hand forward again, showing that the hand with the coin has changed and asking, "What happened?"

Here, if simultaneous exclusion inference is applied to the two simultaneous exclusion processes a and b, and a eliminates a'and b eliminates b', (a + b') eliminates (a'+ b). Then (or a + b excludes a'+ b'). The conclusion of simultaneous exclusion reasoning is that ab exists. The reason is that we can logically replace two 1-bit state processes a and b with one 2-bit state process ab, which counts in the process not by state magnitude but by sequentiality. , Birth order (here, the form of alternation between two complementary states) is to be excluded. That is, the existence of two simultaneous exclusions (a + b') │ (a'+ b) and (a + b) │ (a'+ b') provides sufficient information to encrypt the ab and is therefore logical and computational. It can be explained by giving a legitimate example of ab.

Let δ (a + b') = ab = -δ (a'+ b), δ (a + b) = ab = -δ (a'+ b'), and δ is a boundary operator (similar to the integral of calculus). It is called a boundary operator). On the contrary, the differential method is ∂ (ab) = a + b. The fully realized ab consists of two conjugate simultaneous exclusion values, forming a sine / cosine form. Higher grade operators abc, abcd ,. .. .. Is similarly configured. δ (ab + c) = abc, δ (ab + cd) = abcd, etc. This is the core of the self-organizing method of the present invention.

Now, to the question "What happened?", Answer "When the hand with the coin changes, the system state rotates 180 degrees and ab (a + b') ba = a'+ b". be able to. It can be seen that one bit of information (“what happened”) came from the alternation of two mutually exclusive states. The transition δ (a + b) = ab indicates that the basic perceptual behavior calls for the first perception, followed by the induction of meta-perception.

The systems and methods described above implement spatial computations in computer equipment and have the above implications because the computational process is based on simultaneous exclusion.

The present invention provides a system and a method for constructing a computer environment including a computer device in order to realize spatial calculation.

The computer apparatus of the present invention realizes spatial computation using a plurality of simultaneous threads that are each activated based on the occurrence, simultaneous occurrence, or exclusion of states belonging to the rest of the concurrent threads. This device is a computer device that uses multiple concurrent threads to perform spatial computations and is activated based on the occurrence, concurrency, or exclusion of states in which each concurrency thread belongs to the remaining concurrency threads. Is. This device is a plurality of two-state sensors that detect peripheral phenomena, and a tuple space, which is a record space of a file structure, is included in the vicinity of such two-state sensors, and the detected phenomenon exists in the tuple space. Multiple two-state sensors that produce a binary signal indicating whether or not to do so, one that affects the surroundings, and at least one of the multiple sensors for each effector detects the effect of the effector on the surroundings. The above effectors, and a plurality of actions each having an external state, a level, and a grade, and each of the plurality of actions includes a plurality of actions that execute simultaneous exclusion of two simultaneous threads. Such an external state is at least partially determined by the state of the sensor, the level of action indicates the number of simultaneous exclusions that the action can perform in the hierarchy, and the grade of action is to determine the external state of the action. Indicates the number of sensors used, multiple actions are arranged in a hierarchy, and each of the plurality of actions is a node in the hierarchy.

The computer device is configured to detect simultaneous exclusion between two actions out of multiple actions at one or more levels of the hierarchy and return such simultaneous exclusion to the Corm that created the event window instance. It further includes one or more computers configured to instantiate at least one event window. In addition, such a Corm receives the simultaneous exclusion of the first action and the second action of a plurality of actions from at least one of the event windows, and the node of the first action and the second action. The action of concatenating the new action as a child node of the node of the new action, the action of setting the external state of the new action according to the external state of the first and second actions, and the action of setting the grade of the new action to the first and second actions. The action of grading a new action based on the grade of may be configured to further include one or more Corms that instantiate the new action in multiple actions. Each Corm may be configured to instantiate a new action only if the first action does not share any sensor with the second action. A Corp that creates an instance of a new action emits a tuple indicating that the new action is one or more of the top nodes in the hierarchy, the top node being a node that is not a child node of the other node. , You may remove all tuples that indicate that one of the child nodes of the new node is one of the top nodes. The Corp that instantiates the new action in this way sets the grade of the new action based on the grade of the first action and the second action as follows, and the first and second actions combine other combinations. Wherever there is no need to instantiate a new action , when the first action has a 0th grade and the second action has a 1st grade, the Corp will instantiate a new 1st grade action. When the first action has 1st grade and the 2nd action has 1st grade, the Corp instantiates a new action with 2nd grade, the 1st action has 1st grade and the 2nd action has 2nd grade. When it has a grade, the Corp instantiates a new action with a 3rd grade, and when the first action has a 2nd grade and the second action has a 2nd grade, the Corp instantiates a new action with a 1st grade. Then, when the first action has 2nd grade and the 2nd action has 3rd grade, the Corp instantiates a new action with 1st grade, and the 1st action has 3rd grade and the 2nd action. When has a 3rd grade, the Corp instantiates a new 2nd grade action.

The bottom level of the hierarchy may include one or more of one or more sensors and one or more effectors. Multiple actions are arranged hierarchically to perform one or more goals, and the goals modulate the spatial calculation from the current state to the target state based on one or more impulses around it, and the multiple actions Throughout the hierarchy, each impulse propagates upwards, each target propagates downwards, and one or more of the actions may cause one or more of the effectors to have the above-mentioned effect on the periphery. Good. Computer devices may automatically self-organize into one or more submission architectures that react multiple actions to their surroundings using simultaneous exclusion between sensor pairs and action pairs.

Where each impulse is introduced into the spatial calculation from the periphery as an inversion of the state of the receiving sensor among multiple sensors, and multiple actions can update the external state as follows when introducing one or more impulses: Executes a bubble up process that transmits one or more states of multiple sensors upward from the bottom level of the hierarchy to at least one level above the bottom level, such a bubble up process of each impulse. Creating traces, such traces include actions that update the external state while transmitting the state of multiple sensors upwards, reverse tracking the trace of the bubble-up process, one from a level higher than the bottom level. Execute a trickle down process that transmits one or more sub-targets downward until the above actions, and the effect that one or more effectors are confirmed by the corresponding sub-targets by such one or more actions. Then, one or more of the above-mentioned sub-targets are gathered to form a target.

Also, the nodes in the hierarchy include a parent node, a child node for the parent node, and a top node that is a parent node but not a child node, and a plurality of threads associated with each action include the first to third sets. It may be arranged in a set. The first set detects changes that occur in one or more originaling actions or one or more sensors in a plurality of actions, and updates the external state of the action according to the changes. If the node of the action is a child node, it includes a thread that transmits such changes upward to the action of the parent node of the node to execute the bubble-up process, and the originating action is at a lower hierarchy level than the action. Since it is the first action among a plurality of actions, the reversal of the state of the receiving sensor may be detected. The second set contains threads that run the distributed backchaining process, where in a distributed backchaining process, if the action does not go to ground, then the originaling action goes from the original action to the ground. Add a tree of potential causal links to actions that have the same level as, the tree contains 0, 1 or more links to other actions out of multiple actions, and the other actions are ground. It may have the same level as the originalizing action without becoming. When the action becomes ground, the third set detects the first lower target among the lower targets that identify the action from one of the hierarchical levels higher than the action, and the first lower level. Divide the goal into second and third sub-goals, send the second sub-goal to one of the child nodes and the third sub-goal to the remaining child nodes and link to the action. However, it may include a thread that initiates the execution of all chains waiting for the execution of the action and executes the trickle down process. When one or more of the simultaneous exclusions are related to the surrounding element and the impulse is related to the element, the originating action involves a thread belonging to the simultaneous exclusion of the element, at which time the originating action is bubble The up process may be started.

Even if the target state of the spatial calculation is linked to the current state of the spatial calculation at the intersection of the vertical and horizontal features, the vertical features include the bubble-up process and the trickle-down process, and the horizontal features are determined by the backchaining process. Good. Horizontal features may correspond to one of the levels of the hierarchy.

In the distributed backchaining process, it includes the first sensor of multiple sensors and the first impulse receiving sensor, while the first simultaneous exclusion having the first impulse receiving sensor in the first child node. , The first action that does not go to ground issues the first new impulse using the second thread set, the first new impulse identifies the first sensor as the receiving sensor of the first new impulse, and one The second action of the actions containing the second simultaneous exclusion of the simultaneous exclusions having the same level as the second action and including the first sensor and the second sensor detects the first impulse, and the first Link the second action and add it to the tree, issue a second new impulse with the second set of threads if the second action is not grounded, the second new impulse is the second Identifies the sensor to the receiving sensor of the second new sensor, and the distributed backchaining process uses the remaining action of multiple actions until the most recent new impulse is detected by the grounded action. You may repeat the steps of the action. The distributed backchaining process produces multiple reverse chains of causal links between two of the actions, multiple reverse chains form a tree, and one or more traditional chains required for the reverse chain. The execution of each of the reverse chains may be stopped until the lower target state has a higher level than the action of the reverse chain and is issued by the action that is grounded. When the node of the first action is not the top node, the first action runs the bubble-up process by issuing an impulse that triggers a distributed backchaining process for the next higher level action in the hierarchy, and the top node Each action with a non-node may convert one or more impulses into one or more targets and then transmit one or more targets to the child nodes of the action to perform a trickle down process. See the attached TLinda code for this.

Spatial calculations are facts, and when the computer device further performs a recall operation in response to a request for a fact from the user, and when such a recall operation is performed, the computer device creates an instance of a dedicated event window, described above. The dedicated event window contains the fact, the computer device generates a recall impulse as one of the impulses, and the recall impulse is one or more where the bubble-up process grounds one or more of the first actions. The target state may be set to generate the associated indicator of. Spatial computations may seek a new process that uses a dedicated event window to insert one or more new subgoals associated with fact recalls.

Spatial calculations realize an avatar structure that represents a human user in a virtual world, and in such spatial calculations, multiple actions provide the avatar structure while multiple systems are supplied to the spatial calculations by the system designer. A supply action is included, and sensors and effectors included in multiple system supply actions include an interface to a virtual world in which the avatar structure operates, one or more impulses are sent by the user to a spatial calculation, and the avatar structure is the user's. It may be fixed to a public / individual identity.

Computer devices detect patterns of spatial computation that represent a series of processes that are locked to deadlocks and livelocks in a closed cycle, identify and participate in actions that participate in such cycles. The action to be performed may be configured to temporarily eliminate the sub-targets and impulses in which it operates to relax the cycle and reset the removed sub-targets and impulses after a predetermined period of time.

Each of the actions may have a name known to the spatial calculation but not to the surroundings. Each action may further have two child nodes, which are also actions, and the name of each action is derived from the name of the child node of the action so that the name of the child node is not derived from the name of the action. May be done. Each tuple in the tuple space may contain one or more security bits, which may be set so that the tuple or tuple field does not match an unbound field.

It is a block diagram which shows the basic element of a distributed computer system (a, b) and a parallel system (c). It is a figure which shows the bubble up and back chaining (A) of an impulse, and trickle down (B) of a target through the hierarchy of actions.

The system and method of the present invention, a reversible self-organized hierarchical spatial computation for generating information computer using computer readable code that can be performed reversibly. This code can be executed from a non-transitory computer-readable storage medium that specifies distributed computation. The environment set by this system and method is realized by one computer device, but its potential can be maximized when distributed in the computer device network.

TLinda Coordination Language
Spatial calculations are easiest to define using a collaborative language. A collaborative language, called a thread, is a procedural or control-flowing language that tunes the interactions of operating processes. As a result, collaborative languages do not have arithmetic computing capabilities, and maintain common control flow logic such as conditional structures (eg, conditional expressions) and loop structures (eg, iterative loops) to tune threads. Control flow arithmetic can be included.

Other collaborative languages and models can be used, but collaborative languages using TL Linda, which is a language derived from the existing Linda language, are used, and this language assumes the global tuple space TS, and this space is out of threads. Tuple T = [field1, field2 ,. .. .. ] Do 4 tasks. The four standard Linda operations are Out (T), Rd (T), In (T), and Ever (T).

-Out (T) makes T that exists in TS.

-Rd (T) first matches the morphology of T to the tuple of TS, but if the morphology of T matches the morphology of other taples present in TS, then Rd (T) matches the variable of T. (Ie, the value of the corresponding field is received in the taple where the variable of T matches), and the issuing thread of T can proceed, otherwise the taple whose Rd (T) matches T Block the issuing thread until it appears in the TS.

-In (T) is the same as Rd (T), but T may be eliminated from TS under mutual exclusion. Therefore, a synchronous token can be created when necessary, and at this time, it is prevented that a copy of T (that is, a token) is erroneously created by using In (T). On the other hand, each thread manages its own tuples that remain allocated, and the presence counters for all assigned tuples (not visible to the user, incremented by In (T), by Out (T)). Decreded, but not less than 0) indicates its usefulness (ie, if the presence counter is greater than 0, there is a tuple in the TS).

-Eval (T) creates a new, independent thread specified in the tuple T that points to the TLinda code. Eval (T) does not interrupt the issuing thread and does not promise to return the value to the issuing thread (or all other threads).

In such an existing Linda operation, TL Linda is Co (U, ..., V), NotCo (U, ..., V), AntiRd (U, ..., V) and AntiNotCo (U, ..., V). ., V) is added. Such operations are performed according to Table 1 below. .. .. , Test for simultaneous occurrence of V and its shortage and shut off (ie, wait for one or more threads) (in the table, "ALL" means all tuples specified by the operation). Each of these operations has versions under "one-shot", Cop, NotCop, AntiRdp, AntiCop, AntiNotCop, which perform the operation (preferably without thread blocking) and display True / False. return.

Therefore, the overall calculation style is derived from the association matching, which is the perfect simultaneity between the tuples (representing the current thread state) within the perimeter including the TS, and the inherent synchronization characteristics of the tuple operation itself (ie, thread blocking). Combined with.

Spatial Computational Components The systems and methods of the present invention use TLinda's coordinating principles to construct TS and various objects that appear around it. Such objects, which may be components of spatial computation, include tuples, simultaneous exclusions, threads, impulses, targets, sensors, metasensors, effectors, meta-effectors, actions, event windows and Colms. It is not limited. Such an object is defined as follows (see the attached TLinda code Appendix).

A thread is a "lightweight" process, a small sequential calculation whose state is minimal. Threads in TLinda are similar to oriented objects and other non-value-returning subroutines in high-level programming, and consist primarily of operations involving tuple space. The activity of multiple related threads (eg, in actions described below) is tuned through blocking properties such as primitives RD, In. Therefore, from a reductionist point of view, the hierarchy consists of threads.

The sensor is a single-threaded TLinda object that detects or does not detect a specific phenomenon in the surrounding environment and mediates the result to the tuple space. The sensor has two states, and the detection result can be translated into a binary signal. In the present specification, a 2-state example is assumed. As an example, +1 indicates that what is detected by the sensor is currently present in the TS, and -1 indicates that what is detected is not currently present in the TS. The 0 value of the sensor indicates an error or exception. Such a Z ₃ = {0, 1, 2} = {0, 1, -1} number system can be generalized to _Zn and real numbers R, but at the expense of mathematical ease of handling. .. On the other hand, the sensor can also have two or more states (eg, a spinor that detects an electromagnetic field).

Effectors allow calculations to affect the environment: (1) Output commands that cause effector effects (eg, function calls that cause peripheral device operation with or without passed variables in the program code. ) And (2) are defined from the viewpoint of a sensor (that is, an effector sensor) that detects the effect of the effector on the periphery, and the effector sensor switches its own state in response to the effect. More complex effectors can also be easily defined using such samples.

In spatial calculations, there may be far more sensors than effectors, because it is the number of sensors n that determines the amount of information that can be learned, and (in principle) effectors consume information.

An action is an object that defines, controls, and represents the simultaneous exclusion of two or more different entities (eg, processes, sensors, other actions), and all actions have names, external states, levels, grades, and so on. Category information is attached. The external state is a value, and in the case of the 2-state, it is one of two "spin" (ie, scalar) values 1 and -1. The level of action is hierarchically based on the number of simultaneous exclusions below it. The action grade is one of {1, 2, 3} as an integer indicating the number of sensors that determine the external state of the action. For example, 1-vector a is 1 grade and 2-vector ab is 2. Grade etc. Scalar numbers such as 1 and -1 are grade 0, and grades higher than 3 do not exist in the examples presented. When the conditions that an action fires are currently met in the surroundings, the action goes to ground.

The categorical information can identify an action as the number of equivalence classes given to the action. Examples of such equivalents include "all Actions with grade 2 mod 4" and "all Actions with [level, grade] = [12, 3]". Such equivalence classes correspond to several frequency bands according to Parseval's Identity, as described above. All such equivalence classes of the nature of the components external state (spin) is, for example, can be represented by a single action that is determined by the nature, such as the sum of the spins in the Z ₃ algorithm.

The action has two main parts, a metasensor part and a mete effector part, each of which has several threads (see Attached MetaSensor, MetaEffector threads). Threads in the metasensor part of an action can encrypt the action's name, external state, level, grade, etc. into one or more tuples. The remaining threads in the metasensor portion are also called metasensors because each monitors the status of other action sensors or metasensors. Thus, a metasensor represents a list of sensors for an action, i.e., a list of all sensors that are monitored by an action directly through one or more metasensors of another action.

While such metasensors raise information from bottom-level sensors into the hierarchy described below (“bubble up”), such metasensors are monitored by other metasensors of higher actions in the hierarchy, and thus the child entities of the action It changes its external state, and such changes are propagated through this action to the parent entity of the action. For this reason the TLinda code is very simple (see attachment).

Similarly, high-level action information is propagated down the hierarchy by meta-effects to the lowest primitive effector level (“trickle down”), but the associated effectors have a direct impact on the environment.

The event window (EW) is an object for effectively searching for simultaneous exclusion in a tuple that changes to an action (m-vector, m ≧ 2). The operating principle of WE is introduced in US Pat. No. 5,864,785. The EW contains one or more matching criteria such as actions that certain m-vectors match, but one or more of such m-vectors can be "contained" in the EW. Recognizing that the m-vector can itself be the focus subject of the EW, it provides a self-organizing component of spatial computation. The EW has a fixed sampling interval ΔT, but the EW can also have a minimum sampling interval, a maximum sampling interval, or both. Therefore, the EW can be configured to register events that occur only in the minimum and / or maximum time intervals.

Spatial calculations are the most hierarchical, such hierarchies consist of nodes arranged at discrete levels, with at least one node at each level concatenated to higher level nodes (except the top level). At least one level of each level is linked to a lower level node. The lowest level (level 1) node is occupied by sensors (threads) that detect the surroundings, and is mathematically represented by a 1-vector. The remaining levels of the hierarchy are formed from actions discovered by being instantiated through the event window mechanism (see Corms). Therefore, all subsequent levels (level 2 and above) represent m-vectors because they are constituents (m> 1). The action is a combination of 1-vectors. The level of action is its own level within the hierarchy, but the level of the hierarchy does not necessarily correspond to m, nor does it need to correspond to the number of 1-vectors represented by the action at that level. Thus, for example, a level 2 action may be a combination of two or three 1-vectors. The new action formed from the actions of the two elements moves one level up from the constituent action, and the node of the new action is the parent node of the child node that is the constituent action.

Corm is an object that creates an instance of EW , and in the future, as described above, it searches for simultaneous exclusion and returns it to Corm to take over the growth of calculation. The receiving Corm creates a new action (creates an instance ) from the characteristics of the entities in the simultaneous exclusion. Corm is bound to a single or multiple level hierarchy to instantiate an action. A simple Corm can instantiate only one EW and, for example, generate all actions at that level. More complex Corms can instantiate multiple EWs , including dedicated EWs that match specific actions and tuples described below. CORM is configured to generate a new instance of CORM, itself also a hierarchy of actions to generate the re instance can be readily constructed. In particular, the current level of Corm can generate such a new level of new Corm, given that the formation of new actions is what gives rise to a new level. The Corp is responsible for calculating the grade of the new action and has the advantage of being able to utilize the Clifford-optimization calculation described below in connection with Table 2.

Actions exist to carry out a goal, which when successful triggers the successful activation of one or more effectors, which propagate the goal's intent to the surrounding environment. The target originates from the impulse, which originates from the periphery (for current purposes) and includes peripheral phenomena that are configured to be detected by the sensor. Impulses are spatially calculated tuples at some level of the hierarchy, which can be called the "bottom" level for current purposes because of the self-similarity of the hierarchy. Such "induction" requires a simple inversion of the value of sensor X from the current state to the opposite state. The sensor X that receives the impulse, that is, the sensor X that detects the presence of the impulse-tuple is the receiving sensor. Algebraic measurement form (± 1 + X), │X│ = 1 or TLinda tuple encoding for impulses [↑, X, NotX] and [!] For goals to represent impulses and goals. , X, NotX], but in each case X is a sensor or metasensor.

As will be described later, reversing the state of the receiving sensor changes the external state of the action with the receiving sensor in the sensor list. The action of directly monitoring the receiving sensor (that is, including this sensor and at the lowest level of the receiving sensor upper layer) is the first action to detect the reversal of the state of the receiving sensor, and the state change of this sensor is detected. Responsible for propagating to the grounded action. The first such receiving action is referred to as an origining action below.

Primitive Functions (Primitive Functionalities)
Unlike temporal calculations, spatial calculations are almost / not arithmetic at all, but some arithmetic calculations are essential. It can be extended to include TLinda or simulated by spatial calculations. For example, spatial calculations can map binary numbers 10011 to the binary hierarchy + abcde-abcd-abc + ab + a through 1 → + and 0 → -mapping. You can also define special purpose actions in response to requests for various arithmetic calculations.

Simultaneous exclusions and actions can be recorded as they occur, and such records serve as a means of recovering or replicating a given spatial calculation. However, according to quantum mechanics replication impossibility, the restored / replicated calculation must be different from the original calculation because it must preserve a specific internal state at the time the recent item was recorded. Must be. Therefore, such a "dump" calculation must be done outside of the spatial calculation being replicated. The storage process consists of two stages: continuous recording and dumping, such as memory or storage of computer equipment, when desired. Restoration work uses such dumps to reconfigure the structure and state of interrupted calculations.

Optimizing the event window in the Corm According to our previous work, in principle all tuples in the TS can be simultaneously eliminated to form an action, such an action being done by the event window. To be effective (ie, δ discovers a simultaneous exclusion for the creation of an instance of a new action and returns it to the Corm). Use efficient hardware and software in large systems to optimize such versatility to determine which tuples are best for concentration.

According to a simple but important optimization that can be performed by the inverse operation of TLinda, the two inverse spin values of the external state of the newly formed action are not eliminated at the same time by themselves, which is not possible. This is because it is necessary. Also, when creating an instance of an action (that is, when realizing δ), the actions of the child nodes B1 and B2 of the action do not share the sensor s _x at all. That is, B1 ∩ B2 = 0. This records the respective configuration sensors s _i actions in the sensor list associated with the action generation time instances are efficiently attained. Therefore, at any level, the sensor list for an action contains the names of all the sensors that actually make up this action. Sensors at the bottom level have no child nodes. Since the actions of the child nodes B1 and B2 own the sensor list, their intersections can be easily calculated. If B1∩B2 ≠ 0, no action is instantiated . This is not an error, but a semantically unacceptable combination, as it incorrectly states the effective level and grade of the action.

When instantiating an action , also release a tuple that indicates that this action occupies the top node and has no parents. As a supplement, remove all tuples that represent the child node of the action as the top node.

Also, when creating an instance of an action , the creation of a hierarchy in which the existing node is linked with the node of the new action accelerates by toggle the external state of the new action back and forth several times between the two spin values. Can be transformed into. With such toggling, the new action is registered with another event window (other than the event window being created) that has a matching criterion that matches it. Along with this, the cascade of event windows rises through the hierarchy, completing the hierarchy even faster than if the environment provided the same consistency.

An action of magnitude ⁿ represents 2 ^n-1 individual simultaneous exclusion. This means that in principle there can be ^2n-1 assigned and executed instances of a given action. Such instances are involved in the mutual exclusion protocol so that only one instance acts in a given context. The first result is wasteful and the second result is complex. Another optimal way to avoid all such consequences is to configure the action to specify / control all ^2n-1 alternatives to one thread, and such thread sequentiality is desired. It has a pleasant side effect that can automatically issue mutual exclusion. The system, in some cases, in one th event window hit, or event window only when clearly hit a prompt, it is possible to create an instance of all 2 ^n-1 instance.

In other optimization methods that guarantee both stable arithmetic and manageable mathematical semantics (providing formal function), the Corm modulates the grade of the new action, which is essentially simultaneous. The sum of the grades of action excluded. In the Corm-based embodiment, the new grade can be down-modulated into the set {0, 1, 2, 3} described above, at which time an event window is adopted to detect simultaneous exclusion as follows: Return (Note; EW (p, q), where p, q> 0, the two names and spins of this action are simultaneously eliminated so that they belong to the match criteria of the event window and form a new action. Vector grade).

The EW (1, 1) simultaneously eliminates entities having a grade of 1 mod4 and creates an action having a grade of 1 + 1 = 2 mod4. For example, δ (a, b) → ab, = grade 2.

EW (1, 2) simultaneously eliminates entities having grades 1 mod 4 and 2 mod 4. For example, δ (a + bc) → abc, = grade 3.

EW (2, 2) simultaneously eliminates entities having a magnitude of 2 mod 4. For example, δ (ab + cd) → abcd, = grade 4. Map to grade 1, which is an abcd sign / orientation (conceptually new) 1-vector. Such a 1-vector substitutes for abcd.

EW (2, 3) simultaneously eliminates entities having grades 2 mod 4 and 3 mod 4. For example, δ (ab + cde) → abcde, = grade 5. The sign / orientation of the abcde is mapped to a (conceptually new) 1-vector, grade 5mod4 = 1. Such a 1-vector substitutes for abcde.

EW (3, 3) simultaneously eliminates entities having a grade of 3 mod 4. For example, δ (abc + def) → abcdef, = grade 6 mod4 = grade 2, and abcdef is also replaced by 2-vector.

EW (1, 3) is unacceptable because its behavior is inherently confusing and creates a system that cannot be handled mathematically. Other unlisted grade combinations are similarly unacceptable.

All of these features can be accomplished by the same module (Corm), which can be activated when the event window for this module discovers a new simultaneous exclusion. Table 2 summarizes the simultaneous exclusion for action level pairs.

Here, the highest grade is 6 and the lowest grade is 1, but such results can be transformed into a set {1, 2, 3}, and such a mod4 transformation of the new grade is Corm-. It is for the generation of instances of control macroscopic growth patterns.

The top "level" of the action hierarchy is a dynamically changing set of actions, located on the top node without a "parent" node. When two of these actions are eliminated from each other, the node becomes a child node of the new node in the final action, and this node is now the top node.

In other optimization strategies, the system can represent one or more event windows that participate in one or more of the top actions (actions on the top node). Such an event window provides users and administrators of the system with a high-level view of computational growth and evolution, as well as the opportunity to coordinate them.

Therefore, a new action is added by Corm via the event window. The action can also be deleted, at which time all upward (parent) actions that depend on this action are also deleted (ie, the nodes of the second action and the second action of the node of the deleted action's node. (Until you reach the top node by iteratively climbing the hierarchy, such as the third action of the parent node of). The thread that deletes the action releases one or more tuples that identify the actions in all child nodes of the deleted action as if they reside on the top node, and such actions have a parent action as a result of the deletion. If you don't. Deleting a sensor means deleting all actions associated with the sensor.

To combine the two spatial calculations P, Q, the system combines P's sensor / metasensor list with Q and vice versa as needed.

Each time goal (sensor, effector, actions relating etc.) According to the above description, the spatial calculations when receiving an impulse (e.g., "change x to x ^1") collects changes set the environment, the Such changes are executed in order to obtain the desired result. One example is resetting the state of a satellite with many sensors and effectors. It is not so easy to find such sets of changes and execute them in the proper order.

Let's assume you can perform 5 other actions. Each step in the final plan must or has been performed, and five other possibilities have been constructed from these. It means that the number of possible plans (plans) of n steps is 5'' and the growth is exponential. Moreover, in all real systems such as satellites, the "5" is too small. 50 effectors and n = 500 steps show 50 ⁵⁰⁰ possibilities. If you look for a plan that works in space, such a size is a very difficult task. The following optimization strategies offer a new and highly efficient solution to such problems.

The first optimization is to assemble the plan on a case-by-case basis rather than pre-preparing it, and the latter (pre-preparing) is not always possible in a surprising environment (Stanford AI's Shakey). The current practice of obedience architecture specifies only a very simple plan for a very simple situation, stores it in the system, and then uses it automatically. Everything more complex is treated "upper" by higher level actions, which can be a collection of simple plans to solve more complex problems. Such a highly successful submission architecture focuses on simple failure prevention bottom-level actions (eg, "look for light / darkness", "go along a wall", etc.). Everything that is not such a simple action will be high-level modular work, most of which will be designed and assembled by human programmers. The self-organizing ability of spatial computation replaces all such specifications and human inputs to automate the construction and assembly of complex subordination architectures.

The second optimization is to counter the explosive explosion of possibilities by reducing algebra by hierarchy. The hierarchy of actions, such as the 1st level action directly above the effector of the system, and the 2nd level action above it, including the 1st level action, are all inferred from the observed behavior of the environment and are very important. The event window mechanism builds such a hierarchy exactly, but the basic mechanism is non-selective and cannot be adapted to large systems. Optimizations regarding the use of event windows by Corm solve this problem.

The third optimization is a specific way of using action hierarchies to create many complex levels of plans each time. The plan is represented by one or more goal states of the spatial calculation S that is to be achieved by performing various possible actions. In general, the target state is a set of sensor states, but it is also a more abstract state that corresponds to a slightly higher level.

There are two phases that perform actions to achieve the target state, but the impulse from the boundary of S to the upper level of the hierarchy (that is, the state change of the receiving sensor) and the bubble up of other sensor information, and the final goal. That is the trickle down from the upper level of. Both of these flows are responsible for the action. Therefore, the several threads that make up the action are divided into three main sets.

The first set of threads is responsible for downward peering within the mutable hierarchy, and when such changes occur, it is also responsible for updating the external state of the action and the upward bubble to the parent action of such changes. .. This set has two features: impulse bubble-up (which induces backchaining as described below) and bubble-up of other sensor information.

The second set is in charge of back chaining, which will be described later.

The third set is responsible for upward peering in the hierarchy that tricks down the target by the name of the action, and also has two characteristics: once the action is on the ground, when the target appears, the target Is divided into the targets (sub-targets) of the child nodes of the action, causing a "firing" toward all chains linked to this action.

The TL Linda code for all such threads is included in the attachment, but only the minimal functionality is given as an example.

Example problem The computer environment described above by the example problem can be explained better. Consider block world surround as the next sensor (see attached TL Linda code).

Prior to this, the calculation inferred the following actions (“│” means xor, which takes precedence over those smaller than +).

In the bubble-up phase, for example, the change to x is picked up by its simultaneous exclusion xy, xz and sent to the upper layer. This means both sensors and metasensors.

The example for invoking a goal generation algorithm to create an instance aims to move Block from Place A to Place C. That is, an impulse [↑, c', c] for satisfying Place C is issued in surround. In addition to the above actions, the spatial calculation S has learned the actions ab'⇔ a'b and bc'⇔ b'c, but the action ac'⇔ a'c has not been learned yet, so [! , C', c], first assume that the block must be moved from a to b and then from b to c.

Therefore, the impulse is the current target whose target is c [! , C', c] simply reflects, there is no clear connection from c to a (which currently has a block), so there is no basis for issuing a subgoal to empty a. Therefore, even with all the necessary machines, this goal cannot be achieved. The algorithm solves the very common problem and is very efficient.

Impulse and Target Propagation If you want to propagate the initial impulse and convert it into several targets, you must set up a missing connection between the current state and the target state. It has all vertical and horizontal properties (see Figure 2). The vertical type is an upward propagation in the hierarchy of environmental impulses [↑, X, X'] for change, and such bubble-up is an important feature of the efficiency of spatial calculations, in this case the search space. This is to reduce logistically. Such a bubble-up impulse is eventually transformed into a goal that traces the impulse as it spreads and rises as water flows.

The horizontal feature is a distributed version of backchaining. Without considering the target state for the current state, the algorithm uses a (causal) path that connects the two states after approaching in the opposite direction from the target state to the current state. Of course, the possibilities expand exponentially in any direction, so it is important to ensure that such an explosion does not affect efficiency. This algorithm achieves it using the intersection of vertical and horizontal features.

All levels of the action hierarchy fall into such a "horizontal" plane. Also, as you go up in the hierarchy, the effective grade of the action of the level increases step by step (up to double), and the cause of this calculation also increases globally in terms of scope and consideration.

When the action Me-XY represents an impulse [↑, X, X'], X is one of the two child nodes of Me, and if Me is not grounded, the impulse [↑, Y , Y'], which is an action that flips Y (which is a partner of X) as part of the simultaneous exclusion (at this level shown by the ellipses around several action nodes). Is there a question? " When such an action is discovered, Me can flip X on its own. The impulse emitted by Me is a back-linking impulse. Every action has a thread that looks for such a backlinking impulse and adds itself to the chain in the same way. This is backchaining (indicated by the dotted arrow from the emitting node to the receiving node in FIG. 2). It continues until the chain reaches the grounded action (in the form of simultaneous distribution), which means that the action in question can eventually initiate an effector activation chain that reaches the inciting action Me, which is the current state of the environment. This is because the state (probably) matches the state required to succeed. This algorithm, which is a coordination algorithm, stores such state information as threaded states that are not in memory state (along with surround), avoiding unnecessary duplication of things, for example in the case of overlapping or outdated chains. It is the same as the existing calculation. The algorithm determines the maximum length of the chain and limits the current state that can change when the target state is reached.

There are many link chains that connect the actions of each level, all of which can be sent from B to A (three such chains at each of the first and second levels of the hierarchy in Figure 2) and origining. Causal to action Form a reverse linking tree. Therefore, according to the backchaining process, the chain does not start executing, which is often transmitted from B to A, and once such a chain is executed, much effort is wasted. Or even worse. On the contrary, it stays stopped until the upper level or top level node where the reverse chain at all levels is grounded issues the desired target. These goals then trickled down into a tree-out, becoming all grounds that met the goal (ie, grounded against sub-goals with previous sub-goals that had just finished through trickle-down. ) Only actions can activate child meta-effectors. Hierarchical stacking of such chains releases sub-goals in the proper order, or else through learning, eventually solves them.

When the action Me does not reach the ground, the impulse that reaches the Me or the reverse relinking chain causes the chain to propagate in the reverse direction and bubble up the impulse (see the solid line in Fig. 2 (A)), and the Me itself The new impulse form is MeIpulse = ['∧', XY, -XY]. This causes backchaining at the next level. The impulse that reaches the top node is converted to the desired target and returned to the bottom of the hierarchy (see the solid line in FIG. 2B). Such transformations, as most do, can be extended to slightly slightly different decisions, including humans.

Actions that occur as a backchain-the pair moves forward. Tuples ["D;", AtoB] indicate that two actions A, B as child nodes of the same parent node must be performed strictly in the order A and B. In some cases, the counter is incremented each time such a pair is used, and the pair with the higher counter value may have priority over the alternative plan. If this happens, a long chain of "good" plans and their hierarchies will eventually be generated. Unfortunately, this solves the problem of non-existence, because all given pairs are always selected in the same situation and the selection algorithm is very efficient. Therefore, nothing can be obtained. Also, the flexibility of spatial stratification, which is a collection of all the relevant features of a situation, may be occupied by short-sighted, inflexible, and outdated temporal solutions.

There can be impulses, goals, actions, etc. for goals, as well as search capabilities (to ground the actions).

Geographically distant action nodes and their parent-child nodes automatically propagate a given state change between such nodes, whether they are sensors, impulses, effectors or control tuples. It can be linked by a subscription mechanism. You can also add an action list of connected nodes along with the sensors to keep them subscribed. Such additions are propagated to the first node of the concatenated nodes and vary with the sensors of the nodes below the second concatenated node. It can also include node actions on nodes with concatenated subscriptions. Alternatively, the join mechanism may be a system function such as a function call, or may be initiated by a user or administrator identifying the nodes connected by the join.

Although the concept of "recalling something from memory" itself is part of a spatial calculation operation, the system performs a recall operation to perform all specific calculations and facts F known to S (ie, actions and sensors). Specific state) can be recalled. As an example, the system creates a dedicated event window with matching criteria that match the facts. Therefore, the dedicated event window "contains" the fact. The dedicated event window can also contain additional actions, such as the actions of child nodes in fact mode. It may also be an impulse in which the facts bubble up to the actions contained in the dedicated event window, rather than matching the dedicated event window itself. If an attempt is made to recall a fact from the dedicated event window, an impulse for this fact F is issued. As a result, the bubble-up F-related impulses can be seen in the buffer of the dedicated event window.

You can extend the recall operation described above to look for new processes, which are basically action chains that trigger effectors to achieve goals that you didn't know how to achieve S before. is there. The contents of the event window dedicated to the recall operation can be used to emit subgoals that the spatial computation performs simultaneously, and such subgoals simulate (ie, ground) the actions associated with the recall operation. .. Therefore, new processes are defined in terms of new actions formed by the simultaneous elimination of existing actions and / or existing simulated actions, and the missing concepts / nodes are combined together. Such a problem can also be pursued from the aspect of Fourier with the goal of generating the specific resonance that corresponds to the recalled fact. You can configure your system to start discovering new processes in this way with something like function calls or other program code, but during periods of low activity during computation or when goals are unsatisfactory ( For example, when the top-level action that receives the bubble-up impulse does not go to the ground at all), the utilization of system resources is low.

Computational Security Because the spatial computation S is due to the environment and reacts to the environment, the external process can approach S only in terms of S itself, otherwise it will not react at all. This is the basis of the security of spatial computation for the environment.

Individual communications (called individual tuples within the TS) can be encrypted as needed inside the spatial computation. Regardless of encryption, such communication has almost no value and usually has an a ± 1 value and an anonymous tag because the range is very microscopic and content is atomic. It is created and managed by spatial computations that are practically meaningful only in dynamic event structures.

The basic principle is that information about S can only be obtained through interaction. Question process P? Consider the surrounding S including. Is S P? Generate (internal) information through interaction with. Is such information P? Bubbles up inside S before trickle down in the form of a reaction (in a shared environment) until is recognized. Such recognition is inevitably P about S? Is the only source of information. Therefore, P? Is S P? You cannot receive more information from S than you receive from. P? Information exchange is equivalent only when is a spatial calculation such as S.

In particular, the name of the action (hierarchical structure constitutes S) is P? This is because only the behavior of S (“motion”) is visible from the outside. Due to such a lack of names, P? Cannot penetrate S. Also, because the actions remain anonymous to each other in S, the action names NA are created from the child nodes B1 and B2 as NA = f (B1, B2), where f loses information, the so-called trap-door. It is a (trap-door) function, and B1 and B2 cannot be derived from NA. If f = xor, it is preferable for general and efficient hardware calculation. It is possible to select f as the cipher, but f is reversible in nature and has a weakness in security.

In this way, only the action itself knows its child node, no other action, which means that the names B1 and B2 are known only within the action, which itself is already in the same way. Because it was created. Same for all actions. You can receive a new name each time Bi is eliminated at the same time. By giving a random name to the sensors that make up the boundary of S with the periphery, the periphery can be made defensive. With such a code-naming strategy, it is possible to build a stable security system with strong and fundamental architectural support for spatial computation.

This is especially useful where all such security occurs automatically, especially when considering that spatial computations are self-organizing, as all actions follow the same interaction protocol. Actions that do not follow such a protocol are very difficult to interact with all other actions because they do not know the names of all other actions.

The actions of the child nodes B1 and B2 can be further protected by encrypting the sensor list of such actions because the sensor address is not disclosed.

As another countermeasure against the transitional TLinda code, each tuple may have one or more security bits that deny tuple matching for unbound fields. That is, the standard minimal Linda tuple match has been extended and "wildcard" matches are not allowed for the specified field and / or tuple. Such a simple measure simplifies the "fishing exposure".

P? Issuing an impulse [↑, s → s'] requesting that the state of s (environmentally shared) be changed to s'to react S. Such an impulse is bubbled up to satisfy one or more actions that satisfy the impulse (ie, this action has a variable that matches the variable of the impulse and does not necessarily have to be ground), resulting in , The target is issued by each action. Each goal trickles down, and eventually the effector goal [! , S → s']. Is the transition from s to s'P? Is the only one to look for. Exceptionally, S may announce that higher level actions (ie, NA vs. sensor boundaries {s}) are available for external impulses. However, since the action NA is the goal achiever, the impulse does not propagate further up the hierarchy, so P? To the extent that S allows. Restrict access to.

When an action goes to ground, it means that it can be executed but can only be potentially. The groundness (ie, relevance) of an action indicates that a partial array of actions in the hierarchy below the grounded action is synchronized to perform this action. Despite this useful information applicable to human administrators showing grounded actions, the fact that they are grounded does not accurately indicate when, where, and how this action is performed. .. The lack of information can be used to enhance the privacy of local information and have the ability to legally monitor possible outcomes at a higher level. That is, the legitimate monitor finds that the lower level actions are tuned to an acceptable configuration, otherwise the actions are not grounded. However, the monitor is still locally private because it does not know the details of the lower level actions. Because the actions are arranged in a concatenated hierarchy with appropriate tolerance (such as a search warrant in a criminal situation), the monitor can descend along the node to a level where the action behaves inappropriately. it can.

Other Features An avatar is a software structure that represents an entity (usually a human) for the rest of a computer system. Generally, avatars are used to represent people in videos and online games, but the concept is for users (with great flexibility in choosing their own alternatives) and avatars (safe, durable, versatile and comprehensive). Very common to both system designers (designing interfaces). Such an avatar structure can also be realized in spatial computation through system-supplied actions in which sensors and effectors are in a virtual world. Alternatively, the sensor and effector itself may be the avatar's interface to the virtual world. That is, the spatial calculation effectively instantiates a virtual world avatar with its own sensor and effector through the sensor and effector. The user can define sub-avatars of the comprehensive avatar and decorate them freely. Avatar is operated entirely or partially by software, but there is someone responsible for the action because it must always be tied to the public identity of the individual human being.

The target induction process may be deadlock or livelock, both of which lock a set of processes into a permanent closure cycle. Through the Parseval's Identity, we can see the operations of spatial computation in the wave (ie, Fourier) domain. "Dead spots", "lines" and other characteristic patterns in this calculated spectrum indicate the presence of deadlocks or live rocks and may even identify actions that participate in such locks. The lock can temporarily remove or silence the associated target or impulse, and then after a period of time (eg, according to the basic principles of current operational system design), remove the removed target or impulse. Can be reset. Most of these locks are unlucky and are unlikely to reappear after such an interruption.

The mental mechanism focused on the most modern events is called STM (short term memory), which can be generalized by being included in the event window mechanism that can operate over all appropriate periods including the STM-domain period.

Claims (16)

A computer device that implements spatial computation using multiple concurrent threads and is activated based on the occurrence, concurrency, or exclusion of states in which each concurrency thread belongs to the remaining concurrency threads.
A plurality of two-state sensors that detect peripheral phenomena, and a tuple space, which is a record space of a file structure, is included in the vicinity of the two-state sensors, and whether or not the detected phenomenon exists in the tuple space is determined. Multiple two-state sensors, which generate the indicated binary signal,
One or more effectors that affect the surroundings and at least one of the plurality of sensors for each effector detects the effect of the effector on the surroundings, and
Includes multiple actions, each of which has an external state, level, and grade, each of which performs simultaneous elimination of two simultaneous threads.
The external state is at least partially determined by the state of the sensor.
The level of the action indicates the number of simultaneous exclusions that the action can perform in the hierarchy.
The action grade indicates the number of sensors used to determine the external state of the action.
A computer device in which the plurality of actions are arranged in a hierarchy, and each of the plurality of actions is a node in the hierarchy. At least one configured to detect simultaneous exclusion between two actions out of multiple actions at one or more levels of the hierarchy and return such simultaneous exclusion to the Corm that created the instance of the event window. Action to instantiate an event window,
The first action and the second action of the plurality of actions are simultaneously eliminated from one of the at least one event window, and the node of the first action and the second action is the node of the new action. Operation to connect as a child node,
The operation of setting the external state of the new action according to the external state of the first and second actions, and
By the action of setting the new action grade based on the first and second action grades,
Including one or more Corms that instantiate new actions in multiple actions,
Claim 1 is characterized in that each of the one or more Computers is configured to instantiate a new action only if the first action does not share any sensor with the second action. The computer device described. The Corm that creates an instance of a new action
It emits a tapple indicating that the new action is one of one or more top nodes in the hierarchy, and the top node is a node that is not a child node of the other nodes.
The computer apparatus according to claim 2, wherein all tuples indicating that one of the child nodes of the new action is one of the top nodes are removed. The Corm that instantiates the new action sets the new action grade as follows, based on the grades of the first and second actions:
When the first action has a grade of 0 and the second action has a grade of 1, the Corm instantiates the new action with a grade of 1.
When the first action has grade 1 and the second action has grade 1, the Corm instantiates a new action of grade 2.
When the first action has a grade of 1 and the second action has a grade of 2, the Corm instantiates a new action of grade 3.
When the first action has a second grade and the second action has a second grade, the Corm instantiates a new action with a first grade.
When the first action has a second grade and the second action has a third grade, the Corm instantiates a new action of the first grade.
When the first action has grade 3 and the second action has grade 3, the Corm instantiates a new action with grade 2.
The computer device according to claim 2, wherein the first and second actions do not instantiate a new action if they have other combinations. The bottom level of the hierarchy includes one or more sensors and one or both of the effectors.
The plurality of actions are arranged in the hierarchy so as to execute one or more goals, and the goals modulate the spatial calculation from the current state to the target state based on one or more impulses in the periphery. A plurality of actions propagate each impulse upward and each target downward through the hierarchy, and one or more of the plurality of actions is such that one or more of the effectors are related to the periphery. To influence
A claim, wherein the computer device automatically self-organizes the plurality of actions into one or more belonging architectures that react to the periphery by using simultaneous exclusion between a sensor pair and an action pair. The computer device according to 1. Each impulse is introduced into the spatial calculation from the periphery as a reversal of the state of the receiving sensor among the plurality of sensors, and the plurality of actions update the external state as follows when introducing one or more impulses.
Performs a bubble-up process that transmits one or more states of multiple sensors upward from the bottom level of the hierarchy to at least one level above the bottom level, such a bubble-up process creating traces of each impulse. Such traces include actions that update the external state while transmitting the state of multiple sensors upwards.
Perform a trickle-down process that reversely tracks the traces of the bubble-up process and transmits one or more sub-goals downward from a level higher than the bottom level to one or more actions, such one or more actions. One or more effectors produce the effect confirmed by the corresponding sub-goal, and the one or more sub-goals gather to form a target.
The computer device according to claim 2, wherein the computer device is characterized by the above. The nodes in the hierarchy include a parent node, a child node for the parent node, and a top node that is a parent node but not a child node, and the threads associated with each action are in multiple sets including the first to third sets. Arranged
The first set detects an originalizing action of one or more of the plurality of actions and a change occurring in one or more sensors, and updates the external state of the action according to the change. If the node of the action is a child node, the originating action is at a lower hierarchy level than the action, including a thread that carries the change upwards to the action of the parent node of the node to execute the bubble-up process. Since it is the first action among multiple actions, it detects the reversal of the state of the receiving sensor and detects it.
The second set contains threads that perform a distributed backchaining process, where in a distributed backchaining process, if the action is not grounded, it goes from the originalizing action to the ground and at the same level as the originalizing action. Add a tree of potential causal links to the actions you have, the tree contains 0, 1 or more links to other actions out of multiple actions, and the other actions are not grounded and are called the origining action. Have the same level
When the action becomes ground, the third set detects the first lower target among the lower targets that identify the action from one of the hierarchical levels higher than the action, and the first one. The sub-target of is divided into the second and third sub-targets, the second sub-target is sent to one of the child nodes, the third sub-target is sent to the remaining nodes of the child nodes, and the action is performed. while link, only contains a thread that all of the execution of the chain to start waiting for the execution of the action to execute the trickle-down process,
Claim 6 is characterized in that the condition that the action emits is currently satisfied in the periphery, the action becomes the ground, and the condition that the action emits is not currently satisfied in the periphery, the action does not become the ground. The computer device described. When one or more of the simultaneous exclusions are associated with a peripheral element and the impulse is associated with the element, the originating action includes a thread belonging to the simultaneous exclusion of the elements, at which time the originating action. 7. The computer device according to claim 7, wherein the computer device initiates a bubble-up process. The target state of the spatial calculation is linked to the current state of the spatial calculation at the intersection of the vertical and horizontal features, the vertical feature includes the bubble-up process and the trickle-down process, and the horizontal feature is back. The computer device according to claim 7, characterized in that it is determined by a chaining process. In the distributed back chaining process
The first sensor that does not go to the ground, including the first sensor and the first impulse receiving sensor among multiple sensors, and the first simultaneous exclusion having the first impulse receiving sensor in the first child node. The action uses the second set of threads to issue the first new impulse, the first new impulse identifies the first sensor to the receiving sensor of the first new impulse,
The second action of the actions containing the second simultaneous exclusion of the simultaneous exclusions having the same level as the first action and including the first sensor and the second sensor detects the first impulse and said Link the first action and add it to the tree, if the second action is not grounded, use the second set of threads to issue a second new impulse, the second new impulse is the second Identifies the sensor as the receiving sensor of the second new sensor,
9. The distributed backchaining process is characterized in that the step of the second action is repeated with the remaining actions of the plurality of actions until the newest impulse is detected by the grounded action. The computer device described in. The distributed backchaining process produces a plurality of reverse chains of causal links between two of the plurality of actions, the plurality of reverse chains forming the tree, and one required for the reverse chain. 10. A aspect of claim 10, wherein one or more conventional subtarget states have a higher level than the action of the reverse chain, and the execution of each reverse chain is stopped until it is issued by the action that is grounded. The computer device described in. When the node of the first action is not the top node, the first action runs the bubble-up process by issuing an impulse that triggers a distributed backchaining process for the next higher level action in the hierarchy.
Each action with a node that is not the top node is characterized by converting one or more impulses into one or more goals and then transmitting one or more goals to the child nodes of the action to perform a trickle-down process. The computer device according to claim 11. When the spatial calculation is a fact, the computer device further performs a recall operation in response to a request for the fact from the user, and the recall operation is performed, the computer device creates an instance of a dedicated event window , and the above. The dedicated event window contains the fact, the computer device generates a recall impulse as one of the impulses, and the recall impulse is one or more where the bubble-up process grounds one or more of the first actions. setting a target state to produce such relevant indicator,
Claim 6 is characterized in that the condition that the action emits is currently satisfied in the periphery, the action becomes the ground, and the condition that the action emits is not currently satisfied in the periphery, the action does not become the ground. The computer device described. 13. The computer device of claim 13, wherein the spatial calculation finds a new process that uses the dedicated event window to insert one or more new subgoals associated with fact recall. The spatial calculation realizes an avatar structure that represents a human user in a virtual world.
In the spatial calculation, the plurality of actions realize the avatar structure, and the plurality of system supply actions supplied to the spatial calculation by the system designer are included, and the sensors and effectors included in the plurality of system supply actions are avatars. 6. A claim comprising an interface to a virtual world in which the structure operates, wherein one or more impulses are sent by the user to the spatial calculation and the avatar structure is fixed to the user's public / personal identity. The computer equipment described in. The first aspect of the present invention is that each tuple in the tuple space contains one or more security bits, and the security bits are set so that the fields of the tuples and tuples do not match the unjoined fields. The computer device described.