DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Status of Claims
This action is responsive to Applicant’s claims filed 03/19/2026.
Claims 1-2 and 5-20 are currently pending and have been examined here.
Claims 1-2 and 5-20 have been amended.
Claims 3-4 have been canceled.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-2 and 5-20 are rejected under 35 U.S.C. § 101. The claims are drawn to ineligible patent subject matter, because the claims are directed to a recited judicial exception to patentability (an abstract idea), without claiming something significantly more than the judicial exception itself.
Claims are ineligible for patent protection if they are drawn to subject matter which is not within one of the four statutory categories, or, if the subject matter claimed does fall into one of the four statutory categories, the claims are ineligible if they recite a judicial exception, are directed to that judicial exception, and do not recite additional elements which amount to significantly more than the judicial exception itself. Alice Corp. v. CLS Bank Int'l, 375 U.S. ___ (2014). Accordingly, claims are first analyzed to determine whether they fall into one of the four statutory categories of patent eligible subject matter. Then, if the claims fall within one of the four statutory categories, it must be determined whether the claims are directed to a judicial exception to patentability (i.e., a law of nature, a natural phenomenon, or an abstract idea). In determining whether a claim is directed to a judicial exception, the claim is first analyzed to determine whether the claim recites a judicial exception. If the claim does not recite one of these exceptions, the claim is directed to patent eligible subject matter under 35 U.S.C. 101. If the claim recites one of these exceptions, the claim is then analyzed to determine whether the claim recites additional elements that integrate the exception into a practical application of that exception. Claims which integrate the exception into a practical application of that exception are directed to patent eligible subject matter under 35 U.S.C. 101. If the claim fails to integrate the exception into a practical application of that exception, the claim is directed to an abstract idea. Finally, if the claims are directed to a judicial exception to patentability, the claims are then analyzed determine whether the claims are directed to patent eligible subject matter by reciting meaningful limitations which transform the judicial exception into something significantly more than the judicial exception itself. If they do not, the claims are not directed towards eligible subject matter under 35 U.S.C. § 101.
Regarding independent claims 1, 19, and 20 the claims are directed to one of the four statutory categories (a machine, a process, and an article of manufacture, respectively.) The claimed invention of independent claims 1, 19, and 20 is directed to a judicial exception to patentability, an abstract idea. The claims include limitations which recite elements which can be properly characterized under at least one of the following groupings of subject matter recognized as abstract ideas by MPEP 2106.04(a):
Mathematical Concepts: mathematical relationships, mathematical formulas or equations, and mathematical calculations;
Certain methods of organizing human activity: fundamental economic principles or practices (including hedging, insurance, mitigating risk); commercial or legal interactions (including agreements in the form of contracts; legal obligations; advertising, marketing or sales activities or behaviors; business relations); managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions); and
Mental processes: concepts performed in the human mind (including an observation, evaluation, judgment, opinion)
Claims 1, 19, and 20, as a whole, recite the following limitations:
obtaining blocks of operators corresponding to the model by determining operators constituting the model by parsing the model information, and (claims 1, 19, 20; the broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could obtain blocks of operators by determining operators constituting a model and parsing information)
configuring the model in units of a block of operators by grouping the determined operators into one or more blocks based on the model information; (claims 1, 19, 20; the broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could configure a model in unites of blocks of operators by grouping the operators into blocks based on model information)
and changing a structure of the model to correspond to the target device, by replacing at least one of the blocks of operators corresponding to the model with at least one candidate block, based on a comparison between the at least one candidate block determined by using the target device information as filtering criteria among blocks in a block pool and the blocks of operators corresponding to the model. (claims 1, 19, 20; the broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could change a structure of a model to correspond to a device by replacing blocks of operators with candidate blocks based on comparisons between a candidate block and filtering criteria comprising device information)
The above elements, as a whole, recite mental processes since, but for the requirement to implement the above steps on a set of generic computer components, the entirety of the above set of steps could be performed by a human using their mind, pen and paper, and simple observation, evaluation, and judgment.
Moving forward, the above recited abstract idea is not integrated into a practical application.
The added limitations do not represent an integration of the abstract idea into a practical application because:
the claims represent mere instructions to implement an abstract idea on a computer, and merely use a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).
the claims merely add insignificant extra-solution activity to the judicial exception (activity which can be characterized as incidental to the primary purpose or product that is merely a nominal or tangential addition to the claim). See MPEP 2106.05(g) and/or
the claims represent mere general linking of the use of the judicial exception to a particular technological environment or field of use. See MPEP 2016.05(h)
Beyond those limitations which recite the abstract idea, the following limitations are added:
A method for optimizing an artificial intelligence based model, which is performed by a computing device, the method comprising: (claim 1; the broadest reasonable interpretation of this limitation represents mere instructions to implement the abstract idea on a generic computer used as a tool in its ordinary capacity; alternatively, the broadest reasonable interpretation of this limitation represents mere general linking of the abstract idea to a particular computer environment or field of use)
A non-transitory computer-readable medium comprising a computer program, wherein when the computer program is executed by at least one processor, the computer program allows at least one processor to perform operations for optimizing an artificial intelligence based model, and the operations comprise: (claim 19; the broadest reasonable interpretation of this limitation represents mere instructions to implement the abstract idea on a generic computer used as a tool in its ordinary capacity; alternatively, the broadest reasonable interpretation of this limitation represents mere general linking of the abstract idea to a particular computer environment or field of use)
A computing device comprising: (claim 20; the broadest reasonable interpretation of this limitation represents mere instructions to implement the abstract idea on a generic computer used as a tool in its ordinary capacity; alternatively, the broadest reasonable interpretation of this limitation represents mere general linking of the abstract idea to a particular computer environment or field of use)
at least one processor; (claim 20; the broadest reasonable interpretation of this limitation represents mere instructions to implement the abstract idea on a generic computer used as a tool in its ordinary capacity; alternatively, the broadest reasonable interpretation of this limitation represents mere general linking of the abstract idea to a particular computer environment or field of use)
and a memory, wherein the at least one processor is configured to perform: (claim 20; the broadest reasonable interpretation of this limitation represents mere instructions to implement the abstract idea on a generic computer used as a tool in its ordinary capacity; alternatively, the broadest reasonable interpretation of this limitation represents mere general linking of the abstract idea to a particular computer environment or field of use)
The claims, as a whole, are directed to the abstract idea(s) which they recite. The claim limitations do not present improvements to another technological field, nor do they improve the functioning of a computer or another technology. Nor do the claim limitations apply the judicial exception with, or by use of a particular machine. The claims do not effect a transformation or reduction of a particular article to a different state or thing. See MPEP 2106.05(c). None of the hardware in the claims "offers a meaningful limitation beyond generally linking 'the use of the [method] to a particular technological environment' that is, implementation via computers” such that the claim as a whole is more than a drafting effort designed to monopolize the exception. See MPEP 2106.05(e); Alice Corp. v. CLS Bank Int’l (citing Bilski v. Kappos, 561 U.S. 610, 611 (U.S. 2010)). Therefore, because the claims recite a judicial exception (an abstract idea) and do not integrate the judicial exception into a practical application, the claims, as a whole, are directed to the judicial exception.
Turning to the final prong of the test (Step 2B), independent claims 1, 19, and 20 do not include additional elements that are sufficient to amount to significantly more than the judicial exception, because there are no meaningful limitations which transform the exception into a patent eligible application.
As outlined above, the claim limitations do not present improvements to another technological field, nor do they improve the functioning of a computer or another technology. Nor do the claim limitations apply the judicial exception with, or by use of a particular machine. The claims do not effect a transformation or reduction of a particular article to a different state or thing. See MPEP 2106.05(c). None of the hardware in the claims "offers a meaningful limitation beyond generally linking 'the use of the [method] to a particular technological environment' that is, implementation via computers” such that the claim as a whole is more than a drafting effort designed to monopolize the exception. See MPEP 2106.05(e); Alice Corp. v. CLS Bank Int’l (citing Bilski v. Kappos, 561 U.S. 610, 611 (U.S. 2010)).
Furthermore, no specific limitations are added which represent something other than what is well-understood, routine, and conventional activity in the field. See MPEP 2106.05(d). Besides performing the abstract idea itself, the generic computer components only serve to perform the court-recognized well-understood computer functions of receiving or transmitting data over a network, performing repetitive calculations, electronic record keeping, and storing and retrieving information in memory. See MPEP 2106.05(d). Looking at the limitations as an ordered combination adds nothing that is not already present when looking at the elements taken individually. Their collective functions merely provide conventional computer implementation. The specification details any combination of a generic computer system program to perform the method. Generically recited computer elements do not add a meaningful limitation to the abstract idea because they would be routine in any computer implementation and because the Alice decision noted that generic structures that merely apply the abstract ideas are not significantly more than the abstract ideas. Therefore, independent claims 1, 19, and 20 are rejected under 35 U.S.C. §101 as being directed to ineligible subject matter.
Claims 2 and 5-18, recite the same abstract idea as their respective independent claims.
The following additional features are added in the dependent claims:
Claim 2:
wherein each of the blocks of operators includes operators requisitely used for achieving functions of the model, and each of the blocks of operators has a structure including a node indicating an operator and an edge indicating a connection between operators.
The broadest reasonable interpretation of this limitation merely alters the types of blocks used in the abstract idea above and therefore further recites one or more abstract ideas for the reasons outlined above.
Claim 5:
determining the at least one candidate block to be compared with the blocks of operators among the blocks in the block pool by using the filtering criteria, wherein the filtering criteria are determined based on a combination of the model information and the target device information.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could determine candidate blocks using filtering criteria, and determine the filtering criteria in this manner.
Claim 6:
wherein the filtering criteria include: a first filtering criterion for determining a block having an input size or an output size corresponding to an input size or an output size of each of the blocks of operators constituting the model among a plurality of prestored blocks in the block pool, as the candidate block.
The broadest reasonable interpretation of this limitation merely alters the types of filtering criteria used in the abstract idea above and therefore further recites one or more abstract ideas for the reasons outlined above.
Claim 7:
wherein the filtering criteria include: a second filtering criterion for determining a block constituted only by operators supported in the target device among a plurality of prestored blocks in the block pool, as the candidate block.
The broadest reasonable interpretation of this limitation merely alters the types of filtering criteria used in the abstract idea above and therefore further recites one or more abstract ideas for the reasons outlined above.
Claim 8:
wherein the obtaining the blocks of operators corresponding to the model includes: determining blocks of operators for covering the operators corresponding to the model information among blocks in the block pool including a reference block corresponding to a set of predefined operators and a custom block corresponding to a set of operators according to user definition, and wherein one block among the blocks of operators corresponds to a subset of the operators constituting the model.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could determine blocks of operators in this manner.
Claim 9:
wherein the comparison includes a comparison between first performance of the candidate block in the target device and second performance of the blocks of operators in the target device.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could compare first and second performance of blocks in a target device.
Claim 10:
wherein the first performance and the second performance include: at least one of block latency measured or expected in the target device, block accuracy measured or expected in the target device, or block power consumption measured or expected in the target device.
The broadest reasonable interpretation of this limitation merely alters the types of performance metrics used in the abstract idea above and therefore further recites one or more abstract ideas for the reasons outlined above.
Claim 11:
wherein the changing the structure of the model includes: quantizing the model by using a scheme of changing a type of the model to at least one of a floating point type or an integer type, and wherein the quantizing includes: determining whether each of the blocks of operators to be optimized supports mixed precision;
and maintaining an operator in which a quantization error is a first value to have a floating point type and performing quantization to an integer type for an operator in which the quantization error is a second value, in the block of operators supporting the mixed precision, and wherein the first value is larger than the second value.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could determine whether blocks to be optimized support mixed precision, and maintain operators in this fashion based on these conditions. Furthermore, the broadest reasonable interpretation of this limitation recites mathematical concepts since quantizing a model sets forth and describes a mathematical operation.
Claim 12:
converting the changed model to have a runtime supportable by the target device;
obtaining a benchmark result generated by executing the converted model in the target device;
determining whether the benchmark result satisfies at least one constraint of a first constraint corresponding to the target device or a second constraint set by a user;
and determining reperforming of the changing the structure of the model to correspond to the target device when it is determined that the benchmark result does not satisfy the at least one constraint.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could perform each of the convert, obtain, determine, and determine steps.
Claim 13:
wherein the blocks of operators or the candidate block are selectable from the block pool including a reference block corresponding to a set of predefined operators or a custom block corresponding to a set of operators according to user definition, and each of the blocks in the block pool is defined as metadata indicating a feature of a block or features of operators in the block.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could select blocks from a pool in this manner, wherein blocks are defined as metadata.
Claim 14:
wherein the metadata includes an order of operators in the block, a type of each of operators, input data information of each of operators, output data information of each of operators, and a constraint for each of operators, and the changing the structure of the model includes: changing the structure of the model to correspond to the target device by a scheme of varying a quantitative level of the change of the model or a scheme of the change of the model based on the metadata.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could use this metadata and change the structure of the model by varying a quantitative level of change or scheme of change.
Claim 15:
wherein the changing of the structure of the model includes: performing at least one of compression, weight-lightening, and quantization for a block or an operator in the block in units of a block of operators.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could perform compression, lightening, or quantization of a model. Furthermore, the broadest reasonable interpretation of this limitation recites mathematical concepts since quantizing, weight lightening, or compressing a model sets forth and describes a mathematical operation.
Claim 16:
wherein the changing the structure of the model includes: changing the structure of the model to correspond to the target device by replacing at least one of the blocks of operators corresponding to the model with the at least one candidate block constituted by replacement operators executable in the target device.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could change the structure of the model in this fashion.
Claim 17:
wherein the replacement operators are determined among operators supportable in the target device based on a mathematical operation result of the operator, an input size of the operator, an output size of the operator, an input value of the operator, an output value of the operator, and a relationship between previous or post operators of the operator.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could determine replacement operators based on these factors.
Claim 18:
wherein an operator which is not the same as an operator included in the model but identically performs a function performed by the operator included in the model, or an operator having a shape of data input or output by the operator included in the model and data of a shape quantitatively replaceable, among operators supportable in the target device, is determined as the replacement operators.
The broadest reasonable interpretation of this limitation recites mental processes since a human using their mind, pen and paper, and simple observation, evaluation, and judgment could include this type of operator in a model based on these conditions.
The above limitations do not represent a practical application of the recited abstract idea. The claim limitations do not present improvements to another technological field, nor do they improve the functioning of a computer or another technology. Nor do the claim limitations apply the judicial exception with, or by use of a particular machine. The claims do not effect a transformation or reduction of a particular article to a different state or thing. See MPEP 2106.05(c). None of the hardware in the claims "offers a meaningful limitation beyond generally linking 'the use of the [method] to a particular technological environment' that is, implementation via computers” such that the claim as a whole is more than a drafting effort designed to monopolize the exception. See MPEP 2106.05(e); Alice Corp. v. CLS Bank Int’l (citing Bilski v. Kappos, 561 U.S. 610, 611 (U.S. 2010)). Therefore, because the claims recite a judicial exception (an abstract idea) and do not integrate the judicial exception into a practical application, the claims are also directed to the judicial exception.
Furthermore, the added limitations do not direct the claim to significantly more than the abstract idea. No specific limitations are added which represent something other than what is well-understood, routine, and conventional activity in the field. See MPEP 2106.05(d). Accordingly, none of the dependent claims 2 and 5-18, individually, or as an ordered combination, are directed to patent eligible subject matter under 35 U.S.C. 101.
Please see MPEP §2106.05(d)(II) for a discussion of elements that the Courts have recognized as well-understood, routine, conventional, activity in particular fields.
Please see MPEP §2106 for examination guidelines regarding patent subject matter eligibility.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-2, 5-10, 12-13, and 16-20 are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al. (WIPO Document No. WO2022083624A1; hereinafter "Wang") in view of Kirshenboim et al. (U.S. PG Pub. No. 20240144051; hereinafter "Kirshenboim").
As per claim 1, Wang teaches:
A method for optimizing an artificial intelligence based model, which is performed by a computing device, the method comprising:
Wang teaches a system and method for determining modifications to a first model based on constraints (filtering criteria) which include hardware capabilities of a target device. (Wang: abstract, paragraph [0010, 36])
obtaining model information corresponding to the artificial intelligence based model, and obtaining target device information corresponding to a target device in which the model is to be executed;
Wang teaches that the system may receive constraint conditions including hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 36])
obtaining blocks of operators corresponding to the model by determining operators constituting the model by parsing the model information, and
Wang teaches an initial model set based on structural units and connections and relationships within them, wherein the search space comprising blocks of operators may be parsed in order to generate and configure an initial model. (Wang: paragraph [0022, 101-102, 105])
configuring the model in units of a block of operators by grouping the determined operators into one or more blocks based on the model information;
Wang teaches an initial model set based on structural units and connections and relationships within them, wherein the search space comprising blocks of operators may be parsed in order to generate and configure an initial model. (Wang: paragraph [0022, 101-102, 105])
With respect to the following limitation:
and changing a structure of the model to correspond to the target device, by replacing at least one of the blocks of operators corresponding to the model with at least one candidate block, based on a comparison between the at least one candidate block determined by using the target device information as filtering criteria among blocks in a block pool and the blocks of operators corresponding to the model.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112])
To be thorough, and to the extent that Wang does not explicitly teach choosing blocks from the original model to maintain into the derived model, Kirshenboim teaches this element. Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim teaches combining the above elements with the teachings of Wang for the benefit of providing a system and method which can automatically generate models that are far more efficient than those generated using traditional techniques that do not consider the availability of hardware-supported inference operations during model generation while still providing comparable accuracy to models generated using traditional techniques. (Kirshenboim: paragraph [0019]) Therefore, before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the teachings of Kirshenboim with the teachings of Wang to achieve the aforementioned benefits.
As per claim 2, Wang in view of Kirshenboim teaches all of the limitations of claim 1, as outlined above, and further teaches:
wherein each of the blocks of operators includes operators requisitely used for achieving functions of the model, and each of the blocks of operators has a structure including a node indicating an operator and an edge indicating a connection between operators.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) The motivation to combine Kirshenboim persists.
As per claim 5, Wang in view of Kirshenboim teaches all of the limitations of claim 1, as outlined above, and further teaches:
determining the at least one candidate block to be compared with the blocks of operators among the blocks in the block pool by using the filtering criteria, wherein the filtering criteria are determined based on a combination of the model information and the target device information.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the consideration of input and output sizes in determining which blocks to preserve in the model based on the filtering criteria comprising the device hardware. (Kirshenboim: paragraph [0018, 29-30]) The motivation to combine Kirshenboim persists.
As per claim 6, Wang in view of Kirshenboim teaches all of the limitations of claim 5, as outlined above, and further teaches:
wherein the filtering criteria include: a first filtering criterion for determining a block having an input size or an output size corresponding to an input size or an output size of each of the blocks of operators constituting the model among a plurality of prestored blocks in the block pool, as the candidate block.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the consideration of input and output sizes in determining which blocks to preserve in the model based on the filtering criteria comprising the device hardware. (Kirshenboim: paragraph [0018, 29-30]) The motivation to combine Kirshenboim persists.
As per claim 7, Wang in view of Kirshenboim teaches all of the limitations of claim 5, as outlined above, and further teaches:
wherein the filtering criteria include: a second filtering criterion for determining a block constituted only by operators supported in the target device among a plurality of prestored blocks in the block pool, as the candidate block.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the consideration of input and output sizes compatible with the device in determining which blocks to preserve in the model based on the filtering criteria comprising the device hardware. (Kirshenboim: paragraph [0018, 29-30]) The motivation to combine Kirshenboim persists.
As per claim 8, Wang in view of Kirshenboim teaches all of the limitations of claim 1, as outlined above, and further teaches:
wherein the obtaining the blocks of operators corresponding to the model includes: determining blocks of operators for covering the operators corresponding to the model information among blocks in the block pool including a reference block corresponding to a set of predefined operators and a custom block corresponding to a set of operators according to user definition, and wherein one block among the blocks of operators corresponds to a subset of the operators constituting the model.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches that the user may specify which parameters are to be incorporated into a model. (Kirshenboim: paragraph [0070], Fig 8) The motivation to combine Kirshenboim persists.
As per claim 9, Wang in view of Kirshenboim teaches all of the limitations of claim 1, as outlined above, and further teaches:
wherein the comparison includes a comparison between first performance of the candidate block in the target device and second performance of the blocks of operators in the target device.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the consideration of accuracy of the candidate blocks in performing the task in determining which blocks to include. (Kirshenboim: paragraph [0036-38]) The motivation to combine Kirshenboim persists.
As per claim 10, Wang in view of Kirshenboim teaches all of the limitations of claim 9, as outlined above, and further teaches:
wherein the first performance and the second performance include: at least one of block latency measured or expected in the target device, block accuracy measured or expected in the target device, or block power consumption measured or expected in the target device.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the consideration of accuracy of the candidate blocks in performing the task in determining which blocks to include. (Kirshenboim: paragraph [0036-38]) The motivation to combine Kirshenboim persists.
As per claim 12, Wang in view of Kirshenboim teaches all of the limitations of claim 1, as outlined above, and further teaches:
converting the changed model to have a runtime supportable by the target device;
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the determination as to whether the model meets an execution time metric in determining whether the model is to be used. (Kirshenboim: paragraph [0036-38]) The motivation to combine Kirshenboim persists.
obtaining a benchmark result generated by executing the converted model in the target device;
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the determination as to whether the model meets an execution time metric in determining whether the model is to be used. (Kirshenboim: paragraph [0035-38, 44]) The motivation to combine Kirshenboim persists.
determining whether the benchmark result satisfies at least one constraint of a first constraint corresponding to the target device or a second constraint set by a user;
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the determination as to whether the model meets an execution time metric in determining whether the model is to be used. (Kirshenboim: paragraph [0036-38]) The motivation to combine Kirshenboim persists.
and determining reperforming of the changing the structure of the model to correspond to the target device when it is determined that the benchmark result does not satisfy the at least one constraint.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the determination as to whether the model meets an execution time metric in determining whether the model is to be used. (Kirshenboim: paragraph [0036-38]) The motivation to combine Kirshenboim persists.
As per claim 13, Wang in view of Kirshenboim teaches all of the limitations of claim 1, as outlined above, and further teaches:
wherein the blocks of operators or the candidate block are selectable from the block pool including a reference block corresponding to a set of predefined operators or a custom block corresponding to a set of operators according to user definition, and each of the blocks in the block pool is defined as metadata indicating a feature of a block or features of operators in the block.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches that the user may specify which parameters and which operations (metadata in the form of data describing data of the blocks) are to be incorporated into a model. (Kirshenboim: paragraph [0070], Fig 8) The motivation to combine Kirshenboim persists.
As per claim 16, Wang in view of Kirshenboim teaches all of the limitations of claim 1, as outlined above, and further teaches:
wherein the changing the structure of the model includes: changing the structure of the model to correspond to the target device by replacing at least one of the blocks of operators corresponding to the model with the at least one candidate block constituted by replacement operators executable in the target device.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the replacement of operators in the first model with others in the second, wherein the replacement operators may be compatible with the target device. (Kirshenboim: paragraph [0026, 31-33]) The motivation to combine Kirshenboim persists.
As per claim 17, Wang in view of Kirshenboim teaches all of the limitations of claim 16, as outlined above, and further teaches:
wherein the replacement operators are determined among operators supportable in the target device based on a mathematical operation result of the operator, an input size of the operator, an output size of the operator, an input value of the operator, an output value of the operator, and a relationship between previous or post operators of the operator.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches the consideration of input and output sizes, input and output values, weights for edges, and accuracy in determining which blocks to preserve in the model based on the filtering criteria comprising the device hardware. (Kirshenboim: paragraph [0018, 21, 29-30, 27, 369-38, 98]) The motivation to combine Kirshenboim persists.
As per claim 18, Wang in view of Kirshenboim teaches all of the limitations of claim 16, as outlined above, and further teaches:
wherein an operator which is not the same as an operator included in the model but identically performs a function performed by the operator included in the model, or an operator having a shape of data input or output by the operator included in the model and data of a shape quantitatively replaceable, among operators supportable in the target device, is determined as the replacement operators.
Wang further teaches that the first model may be altered to perform a new task, wherein the altering may comprise selecting new sets of blocks to comprise the model, and wherein the new sets of blocks may be chosen based on the filtering constraint comprising the hardware deployment conditions and chip memory size. (Wang: paragraph [0013, 22, 36, 101-102, 105, 112]) Kirshenboim teaches the determination of a seed model 202, from which one or more child models may be determined based on processing capability of a device to implement the child model within, wherein the system may choose blocks (inference operations and connections therebetween) to incorporate into the child model based on this filter criteria. (Kirshenboim: paragraphs [0029-35, 39], Fig. 1, 2A) Kirshenboim further teaches that child models may be selected such that the shape of data output matches that of the parent models. (Kirshenboim: paragraph [0047-54], Figs. 4-6) The motivation to combine Kirshenboim persists.
As per claim 19, Wang in view of Kirshenboim the limitations of this claim which are substantially identical to those of claim 1, as outlined above, and further teaches:
A non-transitory computer-readable medium comprising a computer program, wherein when the computer program is executed by at least one processor, the computer program allows at least one processor to perform operations for optimizing an artificial intelligence based model, and the operations comprise:
Wang teaches a system and method for determining modifications to a first model based on constraints (filtering criteria) which include hardware capabilities of a target device. (Wang: abstract, paragraph [0010, 36]) Wang further teaches the implementation of the system and method using a computer device and therefore teaches a non-transitory computer readable storage medium storing code which, when executed by a processor, performs the functions of the system. (Wang: paragraphs [0040-44])
As per claim 20, Wang in view of Kirshenboim the limitations of this claim which are substantially identical to those of claim 1, as outlined above, and further teaches:
A computing device comprising:
Wang teaches a system and method for determining modifications to a first model based on constraints (filtering criteria) which include hardware capabilities of a target device. (Wang: abstract, paragraph [0010, 36]) Wang further teaches the implementation of the system and method using a computer device and therefore teaches a non-transitory computer readable storage medium storing code which, when executed by a processor, performs the functions of the system. (Wang: paragraphs [0040-44])
at least one processor;
Wang teaches a system and method for determining modifications to a first model based on constraints (filtering criteria) which include hardware capabilities of a target device. (Wang: abstract, paragraph [0010, 36]) Wang further teaches the implementation of the system and method using a computer device and therefore teaches a non-transitory computer readable storage medium storing code which, when executed by a processor, performs the functions of the system. (Wang: paragraphs [0040-44])
and a memory, wherein the at least one processor is configured to perform:
Wang teaches a system and method for determining modifications to a first model based on constraints (filtering criteria) which include hardware capabilities of a target device. (Wang: abstract, paragraph [0010, 36]) Wang further teaches the implementation of the system and method using a computer device and therefore teaches a non-transitory computer readable storage medium storing code which, when executed by a processor, performs the functions of the system. (Wang: paragraphs [0040-44])
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Wang in view of Kirshenboim further in view of Alakuijala et al. (U.S. PG Pub. No. 20210027195; hereinafter "Alakuijala").
As per claim 15, Wang in view of Kirshenboim teaches all of the limitations of claim 1, as outlined above, but does not appear to explicitly teach:
wherein the changing of the structure of the model includes: performing at least one of compression, weight-lightening, and quantization for a block or an operator in the block in units of a block of operators.
Alakuijala, however, teaches that a model may be derived from another model through compression and quantization of the first model. (Alakuijala: paragraphs [0062-68], Fig. 2) Alakuijala teaches combining the above elements with the teachings of Wang in view of Kirshenboim for the benefit of reductions in bandwidth use which is a significant cost to mobile computing, and allowing for increased use of machine learned models in bandwidth-limited networks. (Alakuijala: paragraph [0033]) Therefore, before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the teachings of Alakuijala with the teachings of Wang in view of Kirshenboim to achieve the aforementioned benefits.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to EMMETT K WALSH whose telephone number is (571)272-2624. The examiner can normally be reached Mon.-Fri. 6 a.m. - 4:45 p.m..
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Jessica Lemieux can be reached at 571-270-3445. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/EMMETT K. WALSH/Primary Examiner, Art Unit 3626
Prosecution Insights