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 .
Information Disclosure Statement
The information disclosure statement(s) submitted on 10/07/2023, 01/26/2024, 06/24/2024 is/are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
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.
Claim(s) 1 – 21 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1
According to the first part of the analysis, in the instant case, claims 1 – 21 are directed to a process, machine, manufacture, or composition of matter. As each of the claims falls within one of the four statutory categories (i.e., process, machine, manufacture, or composition of matter).
Independent Claim 1:
Step 2A Prong 1:
A federated learning method, comprising:
and performing, (This step for generating a posterior distribution is understood to represent a mathematical concept. This is supported by the specification of the application, see paragraph 220 in which a mathematical equation is used to calculate the posterior distribution of a parameter in the local model.)
If a claim limitation under its broadest reasonable interpretation, covers performance of the limitation that can be performed in the human mind and/or using pen and paper as a physical aid, then it falls within the “Mental Processes” grouping of abstract ideas. If a claim limitation under its broadest reasonable interpretation, covers performance of the limitation that can be defined by a mathematical relationship, formula, equation, or calculations. Then it falls within the “Mathematical Concept” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
Step 2A Prong 2:
The judicial exception is not integrated into a practical application.
receiving, by a first node from a second node, a prior distribution of a parameter in a federated model, wherein the federated model is a machine learning model whose parameter obeys a distribution (Receiving data is rejected under insignificant extra-solution activity. See MPEP 2106.05(g). Using a node to perform the operations is understood to be an mere instructions to apply an exception. See MPEP 2106.05(f).)
by the first node, training based on the prior distribution of the parameter in the federated model and local training data of the first node (Training a model based on data is understood to be an mere instructions to apply an exception. Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f))
Step 2B:
receiving, by a first node from a second node, a prior distribution of a parameter in a federated model, wherein the federated model is a machine learning model whose parameter obeys a distribution (Receiving data is well-understood, routine, and conventional activities as supported under Berkheimer Evidence “Receiving or transmitting data over a network”. See MPEP 2106.05(d)(ll). Using a node to perform the operations is understood to be an mere instructions to apply an exception. See MPEP 2106.05(f).)
by the first node, training based on the prior distribution of the parameter in the federated model and local training data of the first node (Training a model based on data is understood to be an mere instructions to apply an exception. Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f))
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because when considered separately and in combination, they do not add significantly more (also known as an “inventive concept”) to the exception.
Dependent Claims 2 – 12 are also ineligible for the same reasons given with respect to claim 1. The dependent claims describe further mental processes and/or do not include additional active functional limitations/steps:
Claim 2:
Step 2A, Prong 1:
The method according to claim 1, wherein the method further comprises:
determining, by the first node, an uncertainty degree of the local model based on the posterior distribution of the parameter in the local model; (This step for determining an uncertainty degree is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. Calculating an uncertainty degree, as currently recited, could just be looking at the distribution and creating a uncertainty value based on it by comparing the distribution to another.)
Step 2A Prong 2 & Step 2B:
and sending, by the first node, the posterior distribution of the parameter in the local model to the second node when the uncertainty degree of the local model meets a first preset condition. (Under Step 2A Prong 2, sending data is rejected under insignificant extra-solution activity. See MPEP 2106.05(g). Under Step 2B, sending data is well-understood, routine, and conventional activities as supported under Berkheimer Evidence “Receiving or transmitting data over a network”. See MPEP 2106.05(d)(ll). This step regarding sending data when it meets a condition also falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case setting a restriction that the local model sends data when it meets a condition does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h)). )
Claim 3:
Step 2A, Prong 1:
Step 2A Prong 2 & Step 2B:
The method according to claim 2, wherein the uncertainty degree of the local model is measured based on at least one of the following:
a variance of the posterior distribution of the parameter in the local model, a convergence speed of the posterior distribution of the parameter in the local model, or inferential accuracy of the posterior distribution of the parameter in the local model. (The calculation of the degree based on variance, convergence speed, or inferential accuracy falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application – See MPEP 2106.05(h))
Claim 4:
Step 2A, Prong 1:
The method according to claim 1, wherein the method further comprises:
determining, by the first node, an uncertainty degree of a first parameter in the local model based on a posterior distribution of the first parameter, wherein the local model comprises at least one parameter, and the first parameter is any one of the at least one parameter; (This step for determining an uncertainty degree is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. Calculating an uncertainty degree, as currently recited, could just be looking at the distribution and creating a uncertainty value based on it by comparing the distribution to another.)
Step 2A Prong 2 & Step 2B:
and sending, by the first node, the posterior distribution of the first parameter to the second node when the uncertainty degree of the first parameter meets a second preset condition. (Under Step 2A Prong 2, sending data is rejected under insignificant extra-solution activity. See MPEP 2106.05(g). Under Step 2B, sending data is well-understood, routine, and conventional activities as supported under Berkheimer Evidence “Receiving or transmitting data over a network”. See MPEP 2106.05(d)(ll). This step regarding sending data when it meets a condition also falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case performing an operation once a condition has been met does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h)). )
Claim 5:
Step 2A, Prong 1:
Step 2A Prong 2 & Step 2B:
The method according to claim 4, wherein the uncertainty degree of the first parameter is based on a variance of the posterior distribution of the first parameter. (The calculation of the degree based on the variance falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application – See MPEP 2106.05(h))
Claim 6:
Step 2A, Prong 1:
The method according to claim 1, wherein the method further comprises:
determining, by the first node, an uncertainty degree of the local model based on the posterior distribution of the parameter in the local model; (This step for determining an uncertainty degree is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. Calculating an uncertainty degree, as currently recited, could just be looking at the distribution and creating a uncertainty value based on it by comparing the distribution to another.)
when the uncertainty degree of the local model meets a first preset condition, determining, by the first node, an uncertainty degree of a first parameter in the local model based on a posterior distribution of the first parameter, wherein the local model comprises at least one parameter, and the first parameter is any of the at least one parameter; (This step for determining an uncertainty degree between the distributions is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. Calculating an uncertainty degree, as currently recited, could just be looking at the distribution and creating a uncertainty value based on it by comparing the distribution to another.)
Step 2A Prong 2 & Step 2B:
and sending, by the first node, the posterior distribution of the first parameter to the second node when the uncertainty degree of the first parameter meets a second preset condition.(Under Step 2A Prong 2, sending data is rejected under insignificant extra-solution activity. See MPEP 2106.05(g). Under Step 2B, sending data is well-understood, routine, and conventional activities as supported under Berkheimer Evidence “Receiving or transmitting data over a network”. See MPEP 2106.05(d)(ll). This step regarding sending data when it meets a condition also falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case performing an operation if a parameter meets a condition does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h)). )
Claim 7:
Step 2A, Prong 1:
The method according to claim 1, wherein the prior distribution of the parameter in the federated model comprises a plurality of local prior distributions, and the plurality of local prior distributions are in a one-to-one correspondence with a plurality of Bayesian models; (This step for defining the distributions contain a one-to-one correspondence with a plurality of Bayesian models is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. As currently recited, the limitation can be understood as just defining Bayesian models related to the distributions, which can be done on a piece of paper.)
determining, by the first node, a prior distribution of the parameter in the local model of the first node based on degrees of matching between the local training data and the plurality of local prior distributions; (This step for determining an uncertainty degree is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process.)
Step 2A Prong 2 & Step 2B:
and the performing, by the first node, training based on the prior distribution of the parameter in the federated model and local training data of the first node, to obtain a posterior distribution of a parameter in a local model of the first node comprises: (Training a model based on data is understood to be an mere instructions to apply an exception. Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f))
and performing, by the first node, training based on the prior distribution of the parameter in the local model and the local training data, to obtain the posterior distribution of the parameter in the local model. (Training a model based on data is understood to be an mere instructions to apply an exception. Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f))
Claim 8:
Step 2A, Prong 1:
The method according to claim 7, wherein
and the determining, by the first node, a prior distribution of the parameter in the local model of the first node based on degrees of matching between the local training data and the plurality of local prior distributions comprises: (These steps for determining a degree of matching between the distributions is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. As currently recited, a person can choose to look at both distributions and think of a value that represents the degree that they are similar.)
determining, by the first node, the prior distribution of the parameter in the local model of the first node based on differences between a historical posterior distribution and the plurality of local prior distributions, wherein the historical posterior distribution is a posterior distribution that is of the parameter in the local model and that is obtained by the first node before the current round of iteration. (These steps for comparing two distributions and creating a new distribution is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. Currently, generating a distribution based on differences between two distributions is recited so generically that it understood as just drawing up a distribution after viewing the two distributions.)
Step 2A Prong 2 & Step 2B:
federated learning comprises a plurality of rounds of iterations, and the posterior distribution of the parameter in the local model is a posterior distribution that is of the parameter in the local model and that is obtained through a current round of iteration; (This step falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case performing federated learning based on performing a plurality of rounds and the posterior distribution being obtained based on the rounds of iteration does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Claim 9:
Step 2A, Prong 1:
The method according to claim 8, wherein the prior distribution of the parameter in the local model is a prior distribution in the plurality of local prior distributions that has a smallest difference from the historical posterior distribution; or the prior distribution of the parameter in the local model is a weighted sum of the plurality of local prior distributions, and weights respectively occupied by the plurality of local prior distributions in the weighted sum are determined by the differences between the historical posterior distribution and the plurality of local prior distributions. (These steps for generating a distribution based on the combination of other distributions and the weights are determined based on “differences” is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. )
Step 2A Prong 2 & Step 2B:
Under Step 2A Prong 2, This judicial exception is not integrated into a practical application.
Under Step 2B, The claim does not include additional elements that are sufficient to amount to
significantly more than the judicial exception.
Claim 10:
Step 2A, Prong 1:
Step 2A Prong 2 & Step 2B:
The method according to claim 1, wherein the method further comprises: sending, by the first node, the posterior distribution of the parameter in the local model to the second node. (Under Step 2A Prong 2, sending data is rejected under insignificant extra-solution activity. See MPEP 2106.05(g). Under Step 2B, sending data is well-understood, routine, and conventional activities as supported under Berkheimer Evidence “Receiving or transmitting data over a network”. See MPEP 2106.05(d)(ll).
Claim 11:
Step 2A, Prong 1:
Step 2A Prong 2 & Step 2B:
The method according to claim 1, wherein the prior distribution of the parameter in the federated model is a probability distribution of the parameter in the federated model, or a probability distribution of the probability distribution of the parameter in the federated model. (This step falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case the distribution being a probability distribution does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Claim 12:
Step 2A, Prong 1:
Step 2A Prong 2 & Step 2B:
The method according to claim 1, wherein the first node and the second node are respectively a client and a server in a network. (This step falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case defining the nodes as a client and a server in a network does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Independent Claim 13:
Step 2A Prong 1:
A federated learning method, comprising:
and
updating, (This step for updating a distribution is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. )
If a claim limitation under its broadest reasonable interpretation, covers performance of the limitation that can be performed in the human mind and/or using pen and paper as a physical aid, then it falls within the “Mental Processes” grouping of abstract ideas. If a claim limitation under its broadest reasonable interpretation, covers performance of the limitation that can be defined by a mathematical relationship, formula, equation, or calculations. Then it falls within the “Mathematical Concept” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
Step 2A Prong 2:
The judicial exception is not integrated into a practical application.
receiving, by a second node, a posterior distribution of a parameter in a local model of at least one first node; (Receiving data is rejected under insignificant extra-solution activity. See MPEP 2106.05(g).)
by a second node (Using a node to perform the operations is understood to be a mere instructions to apply an exception. See MPEP 2106.05(f).)
Step 2B:
receiving, by a second node, a posterior distribution of a parameter in a local model of at least one first node; (Receiving data is well-understood, routine, and conventional activities as supported under Berkheimer Evidence “Receiving or transmitting data over a network”. See MPEP 2106.05(d)(ll).
by a second node (Using a node to perform the operations is understood to be an mere instructions to apply an exception. See MPEP 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because when considered separately and in combination, they do not add significantly more (also known as an “inventive concept”) to the exception.
Dependent Claims 14 – 19 are also ineligible for the same reasons given with respect to claim 13. The dependent claims describe further mental processes and/or do not include additional active functional limitations/steps:
Claim 14:
Step 2A, Prong 1:
The method according to claim 13, wherein before the receiving, by a second node, a posterior distribution of a parameter in a local model of at least one first node, the method further comprises:
selecting, by the second node, the at least one first node from a candidate node, (This step for selecting a node is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process.)
Step 2A Prong 2 & Step 2B:
wherein the second node enacts federated learning using a plurality of rounds of iterations, the at least one first node is a node participating in a current round of iteration, and the candidate node is a node participating in the federated learning before the current round of iteration; (The second node performing federating learning using rounds of iterations and the candidate node fall under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying the use of a node for federated learning does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
and sending, by the second node, the prior distribution of the parameter in the federated model to the at least one first node. (Under Step 2A Prong 2, sending data is rejected under insignificant extra-solution activity. See MPEP 2106.05(g). Under Step 2B, sending data is well-understood, routine, and conventional activities as supported under Berkheimer Evidence “Receiving or transmitting data over a network”. See MPEP 2106.05(d)(ll).)
Claim 15:
Step 2A, Prong 1:
The method according to claim 14, wherein the selecting, by the second node, the at least one first node from a candidate node comprises:
selecting, by the second node, the at least one first node from the candidate node based on evaluation information sent by the candidate node to the second node, wherein the evaluation information indicates a degree of matching between the prior distribution of the parameter in the federated model and local training data of the candidate node,
or the evaluation information indicates a degree of matching between the local training data of the candidate node and a posterior distribution obtained by the candidate node through training based on the prior distribution of the parameter in the federated model,
or the evaluation information indicates a degree of matching between the prior distribution of the parameter in the federated model and the posterior distribution obtained by the candidate node through (These steps for determining a degree of matching between the distributions is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. As currently recited, a person can choose to look at both distributions and think of a value that represents the degree that they are similar.)
Step 2A Prong 2 & Step 2B:
training based on the prior distribution of the parameter in the federated model (Training a model based on data is understood to be an mere instructions to apply an exception. Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f))
Claim 16:
Step 2A, Prong 1:
The method according to claim 14, wherein the selecting, by the second node, the at least one first node from a candidate node comprises:
selecting, the parameter in the federated model, wherein the historical posterior distribution is a posterior distribution that is of the parameter in the local model and that is obtained by the candidate node before the current round of iteration. (This step for selecting a node based on a value (difference) is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. The selection of the first node may be performed manually by a user observing the distribution and using judgement to select the node.)
Step 2A Prong 2 & Step 2B:
by the second node (Using a node to perform the operations is understood to be an mere instructions to apply an exception. See MPEP 2106.05(f).)
Claim 17:
Step 2A, Prong 1:
Step 2A Prong 2 & Step 2B:
The method according to claim 13, wherein the local model comprises no parameter whose uncertainty degree does not meet a preset condition. (This step falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case setting a restriction that the local model comprises no parameter whose uncertainty does not match a condition does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Claim 18:
Step 2A, Prong 1:
The method according to claim 13,
and the updating, (This step for updating the distributions that are related to the nodes based on an observed distribution is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process.)
if a difference between the posterior distributions of the first parameters of the plurality of first nodes is greater than a preset threshold, updating, by the second node, the prior distribution of the parameter in the federated model to split the first parameters into a plurality of parameters. (This step for updating the distributions that are related to the nodes based on an observed distribution is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process.)
Step 2A Prong 2 & Step 2B:
wherein the at least one first node comprises a plurality of first nodes, and posterior distributions of parameters in local models of the plurality of first nodes each comprise a posterior distribution of a first parameter; (This step falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case setting each node to comprise a plurality of nodes and each node using a local distribution does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
by the second node (Using a node to perform the operations is understood to be an mere instructions to apply an exception. See MPEP 2106.05(f).)
Claim 19:
Step 2A, Prong 1:
Step 2A Prong 2 & Step 2B:
The method according to claim 13, wherein the prior distribution of the parameter in the federated model comprises a plurality of local prior distributions, and the plurality of local prior distributions are in a one-to-one correspondence with a plurality of Bayesian models. (This step falls under Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case creating a distribution of the parameter that is comprised by multiple distributions and creating a correspondence to the Bayesian models does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Independent Claim 20:
Step 2A Prong 1:
A federated learning apparatus, wherein the federated learning apparatus may
performing, (This step for generating a posterior distribution is understood to represent a mathematical concept. This is supported by the specification of the application, see paragraph 220 in which a mathematical equation is used to calculate the posterior distribution of a parameter in the local model.)
If a claim limitation under its broadest reasonable interpretation, covers performance of the limitation that can be performed in the human mind and/or using pen and paper as a physical aid, then it falls within the “Mental Processes” grouping of abstract ideas. If a claim limitation under its broadest reasonable interpretation, covers performance of the limitation that can be defined by a mathematical relationship, formula, equation, or calculations. Then it falls within the “Mathematical Concept” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
Step 2A Prong 2:
The judicial exception is not integrated into a practical application.
comprise a memory, the memory stores instructions, a processor is configured to execute the instructions stored in the memory, and when the instructions are executed, the processor is configured to perform (The memory and processor are understood to be generic computer equipment. See MPEP 2106.05(f).)
receiving, by a first node from a second node, a prior distribution of a parameter in a federated model, wherein the federated model is a machine learning model whose parameter obeys a distribution (Receiving data is rejected under insignificant extra-solution activity. See MPEP 2106.05(g). Using a node to perform the operations is understood to be an mere instructions to apply an exception. See MPEP 2106.05(f).)
by the first node, training based on the prior distribution of the parameter in the federated model and local training data of the first node (Training a model based on data is understood to be an mere instructions to apply an exception. Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f))
Step 2B:
comprise a memory, the memory stores instructions, a processor is configured to execute the instructions stored in the memory, and when the instructions are executed, the processor is configured to perform (The memory and processor are understood to be generic computer equipment. See MPEP 2106.05(f).)
receiving, by a first node from a second node, a prior distribution of a parameter in a federated model, wherein the federated model is a machine learning model whose parameter obeys a distribution (Receiving data is well-understood, routine, and conventional activities as supported under Berkheimer Evidence “Receiving or transmitting data over a network”. See MPEP 2106.05(d)(ll). Using a node to perform the operations is understood to be an mere instructions to apply an exception. See MPEP 2106.05(f).)
by the first node, training based on the prior distribution of the parameter in the federated model and local training data of the first node (Training a model based on data is understood to be an mere instructions to apply an exception. Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f))
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because when considered separately and in combination, they do not add significantly more (also known as an “inventive concept”) to the exception.
Independent Claim 21:
Step 2A Prong 1:
A federated learning apparatus, wherein the federated learning apparatus may
updating, (This step for updating a distribution is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. )
If a claim limitation under its broadest reasonable interpretation, covers performance of the limitation that can be performed in the human mind and/or using pen and paper as a physical aid, then it falls within the “Mental Processes” grouping of abstract ideas. If a claim limitation under its broadest reasonable interpretation, covers performance of the limitation that can be defined by a mathematical relationship, formula, equation, or calculations. Then it falls within the “Mathematical Concept” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
Step 2A Prong 2:
The judicial exception is not integrated into a practical application.
comprise a memory, the memory stores instructions, a processor is configured to execute the instructions stored in the memory, and when the instructions are executed, the processor is configured to perform (The memory and processor are understood to be generic computer equipment. See MPEP 2106.05(f).)
receiving, by a second node, a posterior distribution of a parameter in a local model of at least one first node (Receiving data is rejected under insignificant extra-solution activity. See MPEP 2106.05(g). Using a node to perform the operations is understood to be an mere instructions to apply an exception. See MPEP 2106.05(f).)
Step 2B:
comprise a memory, the memory stores instructions, a processor is configured to execute the instructions stored in the memory, and when the instructions are executed, the processor is configured to perform (The memory and processor are understood to be generic computer equipment. See MPEP 2106.05(f).)
receiving, by a second node, a posterior distribution of a parameter in a local model of at least one first node (Receiving data is well-understood, routine, and conventional activities as supported under Berkheimer Evidence “Receiving or transmitting data over a network”. See MPEP 2106.05(d)(ll).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because when considered separately and in combination, they do not add significantly more (also known as an “inventive concept”) to the exception.
Claim Rejections - 35 USC § 102
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 the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claim(s) 13 – 16 and 19 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Corinzia Variational Federated Multi-Task Learning.
Regarding Claim 13, Corinzia teaches:
A federated learning method, comprising: (Corinzia teaches Figure 1, which outlines the framework being used for federated learning.)
receiving, by a second node, a posterior distribution of a parameter in a local model of at least one first node, (Corinzia teaches Algorithm 1 on Page 3, which details a set of C (second client nodes) receiving a s(t) distribution of a parameter from a first node (server) in the federated model. s(t) is a posterior distribution (see Proposition 1 on Page 2, where s(t) is defined).)
and updating, by the second node, a prior distribution of a parameter in a federated model based on the posterior distribution of the parameter in the local model of the at least one first node, wherein the federated model is a machine learning model whose parameter obeys a distribution. (Corinzia teaches Algorithm 1, which teaches optimization (step 8) a prior distribution based on the parameter θ of the server (first node) of a federated learning system.)
Regarding Claim 14, Corinzia teaches all the limitations of Claim 13, and further teaches:
The method according to claim 13, wherein before the receiving, by a second node, a posterior distribution of a parameter in a local model of at least one first node, the method further comprises: (Corinzia teaches Algorithm 1, which teaches receiving by a second node (client) a posterior distribution s(t) in Step 6.)
selecting, by the second node, the at least one first node from a candidate node, wherein the second node enacts federated learning using a plurality of rounds of iterations, (Corinzia teaches Algorithm 1, which teaches step 5, selecting candidate node Ct that occurs for multiple round of operations, as seen by step 4. The node that performs the federated learning step may not be the same as the first node, as taught by 11, for which each step updates the s(t) that is sent to each node.)
the at least one first node is a node participating in a current round of iteration, and the candidate node is a node participating in the federated learning before the current round of iteration; (Corinzia teaches Algorithm 1, which teaches multiple rounds of iteration. The candidate node would’ve been a node in round T-1 while the current node is in round T, thereby Corinzia teaches the limitation.)
and sending, by the second node, the prior distribution of the parameter in the federated model to the at least one first node. (Corinzia teaches Algorithm 1 which teaches sending the server prior that is trained in step 8 and modified in step 9 to the server in step 11. The prior distribution is a prior for the parameter θ)
Regarding Claim 15, Corinzia teaches all the limitations of Claim 14, and further teaches: The method according to claim 14, wherein the selecting, by the second node, the at least one first node from a candidate node comprises:
selecting, by the second node, the at least one first node from the candidate node based on evaluation information sent by the candidate node to the second node, wherein the evaluation information indicates a degree of matching between the prior distribution of the parameter in the federated model and local training data of the candidate node, (Corinzia teaches Algorithm 1 which selects a second node after selection of the candidate node in the previous step, the selection of the second node performs after the updating of the server posterior by the evaluation information, thereby the selection of the second node is based on the completion of the evaluation information sent to the second node.)
or the evaluation information indicates a degree of matching between the local training data of the candidate node and a posterior distribution obtained by the candidate node through training based on the prior distribution of the parameter in the federated model, (Corizia teaches Page 2 Section 2.A, which teaches the sent model distribution that is updated based on the local training data f the candidate node. The algorithm is provided as Algorithm 1, where the priors compute optimized distributions based on the local training data and based on the posterior distribution of the parameter that is sent to the clients. “degrees of matching” has been understood to be step 7, as the computation of the server prior step modifies the data based on the local data to make it match closer to the desired result as described by the local data.)
or the evaluation information indicates a degree of matching between the prior distribution of the parameter in the federated model and the posterior distribution obtained by the candidate node through training based on the prior distribution of the parameter in the federated model.
Regarding Claim 16, Corinzia teaches all the limitations of Claim 14, and further teaches:
The method according to claim 14, wherein the selecting, by the second node, the at least one first node from a candidate node comprises:
selecting, by the second node, the at least one first node from the candidate node based on a difference between a historical posterior distribution of the candidate node and the prior distribution of the parameter in the federated model, wherein the historical posterior distribution is a posterior distribution that is of the parameter in the local model and that is obtained by the candidate node before the current round of iteration. (Corizinia teaches Algorithm 1, in which the selection of the client is selected after the difference (ratio) between a historical posterior distribution of the candidate node and the prior distribution of the parameter in the federated model (see Step 9, in which the updated posterior s(t) is compared to the historical posterior s(t-1) which is obtained by the candidate node from the previous round of iteration.)
Regarding Claim 19, Corinzia teaches all the limitations of Claim 13, and further teaches: The method according to claim 13, wherein the prior distribution of the parameter in the federated model comprises a plurality of local prior distributions, and the plurality of local prior distributions are in a one-to-one correspondence with a plurality of Bayesian models. (Corizia teaches the federated prior “Following a Bayesian approach, we assume a prior distribution over all network parameters p(θ, φ1, . . . , φK).” [Page 2] in the Server that comprises a plurality of prior distributions across multiple parameters and is used to create multiple local prior distributions, as seen in Algorithm 1. Furthermore, the system itself is understood to be a Bayesian Model, as described by Section 2.A, which teaches the clients being part of a Bayesian network (thereby each a Bayesian Model) which have local prior distributions.)
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.
Claim(s) 1, 7 - 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Corinzia Variational Federated Multi-Task Learning and further in view of Lalitha Fully Decentralized Federated Learning
Regarding Claim 1, Corinzia teaches:
A federated learning method, comprising: (Corinzia teaches Figure 1, which outlines the framework being used for federated learning.)
receiving, by a first node from a second node, a (Corinzia teaches Algorithm 1 on Page 3, which details a set of C (first nodes) receiving a s(t) distribution of a parameter from a second node (server) in the federated model. s(t) is a posterior distribution (see Proposition 1 on Page 2, where s(t) is defined).)
wherein the federated model is a machine learning model whose parameter obeys a distribution; (Corinzia teaches Equation(s) 1 and 2 on Page 2, which details the distribution over all parameters for which the selected parameter is part of.)
and performing, by the first node, (Corinzia teaches Algorithm 1 on Page 3, which include steps 6 – 9 which teaches joint optimization (reads on training))
training based on the prior distribution of the parameter in the federated model and local training data of the first node, to obtain a posterior distribution of a parameter in a local model of the first node. (Corinzia teaches Algorithm 1 on Page 3 which teaches the trained model which is trained by the first node being sent to the server and applied (step 11) to develop a posterior distribution of the parameter.)
Corinzia does not distinctly disclose sending a prior distribution, rather it is described as a posterior distribution.
However, Lalitha teaches sending have the nodes start with a prior distribution.
Lalitha teaches Section 3, which teaches step 2 –“Each user i performs a local Bayesian update on ρ (k−1) i to form b (k) i using the following rule.” [Page 3] which is detailed in Assumption 3: “Assumption 3. For all users i ∈ [N] we assume • The prior beliefs ρ (0) i (θ) > 0 for all θ ∈ ΘM.” [Page 3].
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to modify the sending of posterior distributions as taught by Corinzia to instead send a distribution defined as a prior distribution as taught by Lalitha for the improvement of separating the local prior distribution from the public distribution that is meant to be shared with the neighbors to perform a local analysis. This improvement is taught by Lalitha in Section 3 on page 3, where step 2 the users receive prior distributions and perform updates which then goes into step 5 which involves the users making an estimate using the private distributions.
Regarding Claim 7, Corinzia as modified by Lalitha teaches all the limitations of Claim 1, and does teach: The method according to claim 1, wherein the prior distribution of the parameter in the federated model comprises a plurality of local prior distributions, and the plurality of local prior distributions are in a one-to-one correspondence with a plurality of Bayesian models; (Corizia teaches: “The server maintains a posterior distribution that represents the plausibility of the shared parameters. In one step of the algorithm, the posterior is communicated to the clients before the training starts, while during training the clients update the posterior given the likelihood of their local data. Finally, the posterior update is sent back to the central server” on page 1 which teaches the creation of the local prior distributions (prior as the combination with Lalitha teaches sending a prior distribution) that are based on a Bayesian model as taught in Claim 1.)
and the performing, by the first node, training based on the prior distribution of the parameter in the federated model and local training data of the first node, to obtain a posterior distribution of a parameter in a local model of the first node comprises: (Corizia teaches: “The server maintains a posterior distribution that represents the plausibility of the shared parameters. In one step of the algorithm, the posterior is communicated to the clients before the training starts, while during training the clients update the posterior given the likelihood of their local data. Finally, the posterior update is sent back to the central server” on page 1 which teaches updating (reads on training) the distribution based on the local data.)
determining, by the first node, a prior distribution of the parameter in the local model of the first node based on degrees of matching between the local training data and the plurality of local prior distributions; (Corizia teaches Page 2 Section 2.A, which teaches the sent model distribution that is updated based on the local data. The algorithm is provided as Algorithm 1, where the priors compute optimized distributions based on the local training data and based on the a prior distribution in the one of the set of the sent client distributions. “degrees of matching” has been understood to be step 7, as the computation of the server prior step modifies the data based on the local data to make it match closer to the desired result as described by the local data.)
and performing, by the first node, training based on the prior distribution of the parameter in the local model and the local training data, to obtain the posterior distribution of the parameter in the local model. (Corizia teaches Page 2 Section 2.A, which teaches the sent model distribution that is updated based on the local data. The algorithm is provided as Algorithm 1, where the priors compute optimized distributions based on the local training data and based on the a prior distribution in the one of the set of the sent client distributions. The training step is step 8, where the determined prior is optimized by the client.)
Regarding Claim 8, Corinzia as modified by Lalitha teaches all the limitations of Claim 7, and does teach:
The method according to claim 7, wherein federated learning comprises a plurality of rounds of iterations, (Corinzia teaches Algorithm 1, which teach T, the number of iterations to perform federated learning)
and the posterior distribution of the parameter in the local model is a posterior distribution that is of the parameter in the local model and that is obtained through a current round of iteration; (Corinzia teaches Algorithm 1, which teaches generating a distribution of the distribution in the client (reads on local model) that is of the parameter θ for the current round of iteration)
and the determining, by the first node, a prior distribution of the parameter in the local model of the first node based on degrees of matching between the local training data and the plurality of local prior distributions comprises: (Corizinia teaches determining a distribution based on degrees of matching between the local data, as seen in Claim 7)
determining, by the first node, the prior distribution of the parameter in the local model of the first node based on differences between a historical posterior distribution and the plurality of local prior distributions, wherein the historical posterior distribution is a posterior distribution that is of the parameter in the local model and that is obtained by the first node before the current round of iteration. (Corizinia teaches receiving the historical model s(t-1) in step 9 of Algorithm 1 which is a posterior distribution (see Proposition 1 on Page 2) that contains the parameter and is store din the local model and is created before the current round updates.)
Regarding Claim 9, Corinzia as modified by Lalitha teaches all the limitations of Claim 8, and does teach: The method according to claim 8, wherein the prior distribution of the parameter in the local model is a prior distribution in the plurality of local prior distributions that has a smallest difference from the historical posterior distribution;
or the prior distribution of the parameter in the local model is a weighted sum of the plurality of local prior distributions, and weights respectively occupied by the plurality of local prior distributions in the weighted sum are determined by the differences between the historical posterior distribution and the plurality of local prior distributions. (Corizinia teaches Algorithm 1, which teaches computing the delta that is calculated based on the ratio between the historical posterior distribution (s(t-1)) and the local distribution s(t) and aggregates (calculating a weight sum) in step 10 based on the plurality of prior distributions. This becomes the future prior distribution for the client, as seen in step 11.)
Regarding Claim 10 , Corinzia as modified by Lalitha teaches all the limitations of Claim 8, and does teach: The method according to claim 1, wherein the method further comprises: sending, by the first node, the posterior distribution of the parameter in the local model to the second node. (Corinzia teaches Algorithm 1, for which the client (first node) sends the distribution of the parameter to the local model. This is explained on Page 3: “In the general setting, it can be computed as the un-normalized pdf ∆i = s (t) i (θ) s (t−1) i (θ) . The client i communicates the delta ∆i to the main server”)
Regarding Claim 11 , Corinzia as modified by Lalitha teaches all the limitations of Claim 8, and does teach: The method according to claim 1, wherein the prior distribution of the parameter in the federated model is a probability distribution of the parameter in the federated model, or a probability distribution of the probability distribution of the parameter in the federated model. (Corinzia teaches Algorithm 1, which teaches the new server prior (step 7) is a probability distribution of the parameter θ in the federated model system.)
Regarding Claim 12, Corinzia as modified by Lalitha teaches all the limitations of Claim 8, and does teach: The method according to claim 1, wherein the first node and the second node are respectively a client and a server in a network. (Corinzia teaches Algorithm 1, which directly outlines a first node (client) and a second node (server).)
Claim(s) 2, 3, and 4 is/are rejected under 35 U.S.C. 103 as being unpatentable over Corinzia Variational Federated Multi-Task Learning and further in view of Lalitha Fully Decentralized Federated Learning and further in view of Sozinov Human Activity Recognition Using Federated Learning
Regarding Claim 2, Corinzia as modified by Lalitha teaches all the limitations of Claim 1, but does not teach:
The method according to claim 1, wherein the method further comprises:
determining, by the first node, an uncertainty degree of the local model based on the posterior distribution of the parameter in the local model;
and sending, by the first node, the posterior distribution of the parameter in the local model to the second node when the uncertainty degree of the local model meets a first preset condition.
However, Sozinov teaches:
determining, by the first node, an uncertainty degree of the local model based on the posterior distribution of the parameter in the local model; (Sozinov teaches Section IV.E which teaches calculating a test accuracy (reads on uncertainty degree) based on the local dataset (which includes a local distribution).)
and sending, by the first node, the posterior distribution of the parameter in the local model to the second node when the uncertainty degree of the local model meets a first preset condition. (Sozinov teaches Section IV.E teaches rejecting the nodes (referenced as clients) which have a test accuracy that is lower the preset condition (referenced as accuracy threshold t). The nodes that are kept then, are the nodes that are equal to (meet) or higher than the preset condition.
As taught by Sozinov on page 1103 Section 2.A “When all clients are done with several epochs of the SGD, the server collects all local models and updates the global model Mr+1.”, which teaches that the nodes are sent to the second node after the updates are performed (see Algorithm 2 on Page 1107 which demonstrates the step). It is understood then, that the local models and the distributions stored within are sent to the second node in cases when they are not rejected (and thereby meet a first preset condition, which is the condition of not being lower then the accuracy thereshold t).)
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to combine the creation of the local model which has a posterior distribution of a parameter as taught by Corinzia as modified by Lalitha with the uncertainty calculation and rejection as taught by Sozinov for the improvement of removing erroneous nodes in the federated learning. This improvement is taught on Page 1106 of Sozinoz: “To address the problem of erroneous clients in the federated learning, we propose a rejection algorithm based on the test accuracy of the each individual client in the federated learning.”
Regarding Claim 3, Corinzia as modified by Lalitha and as modified by Sozinov teach all the limitations of Claim 2, and Sozinov further teaches: The method according to claim 2, wherein the uncertainty degree of the local model is measured based on at least one of the following: a variance of the posterior distribution of the parameter in the local model, a convergence speed of the posterior distribution of the parameter in the local model, or inferential accuracy of the posterior distribution of the parameter in the local model. (Sozinoz teaches: “Figure 3 shows the convergence and the test accuracy of DNN and softmax regression models trained for HAR classification”. [Page 1108] As seen in the graphs below, convergence and test accuracy are understood to be representing the same element (as they aren’t separated and figure 3 is meant to provide information for both of them, otherwise.) This is understood to be the convergence speed of the models, as Figure 3 shows the rounds it takes for convergence for the local models.)
PNG
media_image1.png
531
658
media_image1.png
Greyscale
The motivation for this combination can be found in Claim 2.
Regarding Claim 4, Corinzia as modified by Lalitha teach all the limitations of Claim 1, and further teaches:
The method according to claim 1, wherein the method further comprises:
(Corinzia as modified by Lalitha teaches the local model comprising at least one parameter and the first parameter is at least one parameter, see Claim 1 above.)
Corinzia as modified by Lalitha does not distinctly disclose:determining, by the first node, an uncertainty degree of a first parameter in the local model based on a posterior distribution of the first parameter
and sending, by the first node, the posterior distribution of the first parameter to the second node when the uncertainty degree of the first parameter meets a second preset condition.
However, Sozinov teaches:
determining, by the first node, an uncertainty degree of a first parameter in the local model based on a posterior distribution of the first parameter
(Sozinov teaches Section IV.E which teaches calculating a test accuracy (reads on uncertainty degree) based on the local dataset (which includes a local distribution of the first parameter).)
and sending, by the first node, the posterior distribution of the first parameter to the second node when the uncertainty degree of the first parameter meets a second preset condition. (Sozinov teaches Section IV.E teaches rejecting the nodes (referenced as clients) which have a test accuracy that is lower the preset condition (referenced as accuracy threshold t). The nodes that are kept then, are the nodes that are equal to (meet) or higher than the preset condition.
As taught by Sozinov on page 1103 Section 2.A “When all clients are done with several epochs of the SGD, the server collects all local models and updates the global model Mr+1.”, which teaches that the nodes are sent to the second node after the updates are performed (see Algorithm 2 on Page 1107 which demonstrates the step). It is understood then, that the local models and the distributions stored within are sent to the second node in cases when they are not rejected (and thereby meet a first preset condition, which is the condition of not being lower then the accuracy thereshold t).)
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to combine the creation of the local model which has a posterior distribution of a parameter as taught by Corinzia as modified by Lalitha with the uncertainty calculation and rejection as taught by Sozinov for the improvement of removing erroneous nodes in the federated learning. This improvement is taught on Page 1106 of Sozinoz: “To address the problem of erroneous clients in the federated learning, we propose a rejection algorithm based on the test accuracy of the each individual client in the federated learning.”
Claim(s) 5 is/are rejected under 35 U.S.C. 103 as being unpatentable over Corinzia Variational Federated Multi-Task Learning and further in view of Lalitha Fully Decentralized Federated Learning and further in view of Sozinov Human Activity Recognition Using Federated Learning and finally in view of Mohri Agnostic Federated Learning
Regarding Claim 5, Corinzia as modified by Lalitha and as modified by Sozinov teach all the limitations of Claim 4, but does not distinctly teach: The method according to claim 4, wherein the uncertainty degree of the first parameter is based on a variance of the posterior distribution of the first parameter.
However, Mohri teaches the limitation.
See Page 20, Lemma 9 which teach an a bound for σw (Desired Variance of w) based on σI (Variance of a distribution). It is understood that the uncertainty degree is a bounding that helps choose the distribution of the parameter, as seen in Page 21 in which Mohri teaches “the right choice of the stochastic variance of w depends on the application” which teaches the variance of w modifies in different cases.
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to modify the uncertainty degree calculation as taught by Corinzia as modified by Lalitha and as modified by Sozinov with the calculation based on a variance as taught by Mohri for the improvement of bounding the variance of the distribution and allowing the system to choose based on the variance. This improvement is taught by Mohri: “Hence, the right choice of the stochastic variance of w depends on the application. If all domains are roughly equally weighted, then we have R(Λ) ≈ 1 p and the PERDOMAIN-variance is a more favorable choice. Otherwise, if σ 2 O (w) is small, then the WEIGHTED-stochastic gradient is more favorable.” [Page 21] in which the variance is used to select what measurement is most favorable.
Claim(s) 6 is/are rejected under 35 U.S.C. 103 as being unpatentable over Corinzia Variational Federated Multi-Task Learning and further in view of Lalitha Fully Decentralized Federated Learning and further in view of Sozinov Human Activity Recognition Using Federated Learning and in view of Blundell (U.S. Publication US20200005152A1)
Regarding Claim 6, Corinzia as modified by Lalitha teach all the limitations of Claim 1, but does not teach: The method according to claim 1, wherein the method further comprises: determining, by the first node, an uncertainty degree of the local model based on the posterior distribution of the parameter in the local model;
when the uncertainty degree of the local model meets a first preset condition, determining, by the first node, an uncertainty degree of a first parameter in the local model based on a posterior distribution of the first parameter, wherein the local model comprises at least one parameter, and the first parameter is any of the at least one parameter;
and sending, by the first node, the posterior distribution of the first parameter to the second node when the uncertainty degree of the first parameter meets a second preset condition.
However Blundell teaches:
The method according to claim 1, wherein the method further comprises: determining, by the first node, an uncertainty degree of the local model based on the posterior distribution of the parameter in the local model;
when the uncertainty degree of the local model meets a first preset condition, determining, by the first node, an uncertainty degree of a first parameter in the local model
Blundell teaches in paragraph 0067 calculating a measure for the values of the parameters of the posterior distribution and checking if it falls below a predetermined threshold. It then repeats the calculation steps as taught in Paragraph 0068 which leads back to the step 308, thereby teach an iterative calculation of an uncertainty degree of a model that matches a first threshold and then calculating the measure again.
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to combine the posterior distribution and federated learning system of Corinzia as modified by Lalitha with the step of literately calculating an uncertainty degree of a distribution as taught by Blundell for the improvement of being able to check when the training of the distribution/model is complete. This improvement is taught by Blundell: “The system determines whether the training is complete (308)” [0067] which directly teaches the intention of the step is to check when the system has performed its operation.
However, Sozinov teaches:
determining, by the first node, an uncertainty degree of a first parameter in the local model based on a posterior distribution of the first parameter, wherein the local model comprises at least one parameter, and the first parameter is any of the at least one parameter (Sozinov teaches Section IV.E which teaches calculating a test accuracy (reads on uncertainty degree) based on the local dataset (which includes a local distribution).)
and sending, by the first node, the posterior distribution of the first parameter to the second node when the uncertainty degree of the first parameter meets a second preset condition. (Sozinov teaches Section IV.E teaches rejecting the nodes (referenced as clients) which have a test accuracy that is lower the preset condition (referenced as accuracy threshold t). The nodes that are kept then, are the nodes that are equal to (meet) or higher than the preset condition.
As taught by Sozinov on page 1103 Section 2.A “When all clients are done with several epochs of the SGD, the server collects all local models and updates the global model Mr+1.”, which teaches that the nodes are sent to the second node after the updates are performed (see Algorithm 2 on Page 1107 which demonstrates the step). It is understood then, that the local models and the distributions stored within are sent to the second node in cases when they are not rejected (and thereby meet a first preset condition, which is the condition of not being lower then the accuracy thereshold t).)
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to combine the creation of the local model which has a posterior distribution of a parameter as taught by Corinzia as modified by Lalitha with the uncertainty calculation and rejection as taught by Sozinov for the improvement of removing erroneous nodes in the federated learning. This improvement is taught on Page 1106 of Sozinoz: “To address the problem of erroneous clients in the federated learning, we propose a rejection algorithm based on the test accuracy of the each individual client in the federated learning.”
Claim(s) 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Corinzia Variational Federated Multi-Task Learning in view of Sozinov Human Activity Recognition Using Federated Learning
Regarding Claim 17, Corinzia teaches all the limitations of Claim 13, but does not teach:
However, Sozinov teaches:
The method according to claim 13, wherein the local model comprises no parameter whose uncertainty degree does not meet a preset condition.
(Sozinov teaches Section IV.E which teaches calculating a test accuracy (reads on uncertainty degree) based on the local dataset (which includes a local distribution of the first parameter). Sozinov also teaches Section IV.E teaches rejecting the nodes (referenced as clients) which have a test accuracy that is lower the preset condition (referenced as accuracy threshold t), thereby having no local nodes with parameters whose uncertainty degree does not met a preset condition (the nodes rejected are lower then the condition.)
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to combine the creation of the local model which has a posterior distribution of a parameter as taught by Corinzia with the uncertainty calculation and rejection as taught by Sozinov for the improvement of removing erroneous nodes in the federated learning. This improvement is taught on Page 1106 of Sozinoz: “To address the problem of erroneous clients in the federated learning, we propose a rejection algorithm based on the test accuracy of the each individual client in the federated learning.”
Claim(s) 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Corinzia Variational Federated Multi-Task Learning and in view of Sozinov Human Activity Recognition Using Federated Learning
Regarding Claim 18, Corinzia teaches all the limitations of Claim 13, and further teaches: The method according to claim 13, wherein the at least one first node comprises a plurality of first nodes, and posterior distributions of parameters in local models of the plurality of first nodes each comprise a posterior distribution of a first parameter; (Corinzia teaches Algorithm 1, which teaches a plurality of active client nodes which receive the server posterior s(t) from the server in step 6. The first node (server) comprises a plurality of distributions that form the total prior distribution and thereby separate nodes under the server that contains the distribution.)
and the updating, by the second node, a prior distribution of a parameter in a federated model based on the posterior distribution of the parameter in the local model of the at least one first node comprises: (Corinzia teaches Algorithm 1, which teaches computing a server prior based on the posterior distribution.)
Corinzia does not distinctly disclose:
if a difference between the posterior distributions of the first parameters of the plurality of first nodes is greater than a preset threshold,
updating, by the second node, the prior distribution of the parameter in the federated model to split the first parameters into a plurality of parameters.
However, Sozinov teaches:
if a difference between the posterior distributions of the first parameters of the plurality of first nodes is greater than a preset threshold, (Sozinov teaches: “Then, using the test accuracy on each of the local test datasets and an accuracy threshold t, we reject clients that have the test accuracy below the threshold. For selecting t, we investigate local test accuracy on a normal client and compare it to the local test accuracy on an erroneous client.” [Page 1106] which teaches a comparison between a distribution of the parameters stored in the models based on local test accuracy, with nodes that are accepted being greater than a preset threshold, t. )
updating, by the second node, the prior distribution of the parameter in the federated model to split the first parameters into a plurality of parameters. (Sozinov teaches splitting the nodes that are lower than the preset threshold from those which are above the preset threshold, in which the updating and selection step is performed on the Clients and the split is performed as a result. See Algorithm 2 for the method of rejection.)
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to combine the creation of the local model which has a posterior distribution of a parameter as taught by Corinzia with the uncertainty calculation and rejection as taught by Sozinov for the improvement of removing erroneous nodes in the federated learning. This improvement is taught on Page 1106 of Sozinoz: “To address the problem of erroneous clients in the federated learning, we propose a rejection algorithm based on the test accuracy of the each individual client in the federated learning.”
Claim(s) 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Corinzia Variational Federated Multi-Task Learning and in view of Blundell (U.S. Publication US20200005152A1) and in view of Lalitha Fully Decentralized Federated Learning
Regarding Claim 20, Corinzia teaches:
A federated learning apparatus
receiving, by a first node from a second node, a (Corinzia teaches Algorithm 1 on Page 3, which details a set of C (first nodes) receiving a s(t) distribution of a parameter from a second node (server) in the federated model. s(t) is a posterior distribution (see Proposition 1 on Page 2, where s(t) is defined). Corinzia teaches Equation(s) 1 and 2 on Page 2, which details the distribution over all parameters for which the selected parameter is part of.)
and performing, by the first node, (Corinzia teaches Algorithm 1 on Page 3, which include steps 6 – 9 which teaches joint optimization (reads on training))
training based on the prior distribution of the parameter in the federated model and local training data of the first node, to obtain a posterior distribution of a parameter in a local model of the first node. (Corinzia teaches Algorithm 1 on Page 3 which teaches the trained model which is trained by the first node being sent to the server and applied (step 11) to develop a posterior distribution of the parameter.) Corinzia does not distinctly teach:
may comprise a memory, the memory stores instructions, a processor is configured to execute the instructions stored in the memory, and when the instructions are executed, the processor is configured to perform
However, Blundell teaches: “Computers suitable for the execution of a computer program can be based on general or special purpose microprocessors or both, or any other kind of central processing unit. Generally, a central processing unit will receive instructions and data from a read-only memory or a random access memory or both. The essential elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices for storing instructions and data. The central processing unit and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.” [0091] which teach using a memory and processor to run computer programs.
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to use the federated learning system of Corinzia with a memory and a processor as taught by Blundell as it is a known method for running computer programs.
Corinzia as modified by Blundell does not distinctly disclose sending a prior distribution, rather it is described as a posterior distribution.
However, Lalitha teaches sending have the nodes start with a prior distribution.
Lalitha teaches Section 3, which teaches step 2 –“Each user i performs a local Bayesian update on ρ (k−1) i to form b (k) i using the following rule.” [Page 3] which is detailed in Assumption 3: “Assumption 3. For all users i ∈ [N] we assume • The prior beliefs ρ (0) i (θ) > 0 for all θ ∈ ΘM.” [Page 3].
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to modify the sending of posterior distributions as taught by Corinzia to instead send a distribution defined as a prior distribution as taught by Lalitha for the improvement of separating the local prior distribution from the public distribution that is meant to be shared with the neighbors to perform a local analysis. This improvement is taught by Lalitha in Section 3 on page 3, where step 2 the users receive prior distributions and perform updates which then goes into step 5 which involves the users making an estimate using the private distributions.
Claim(s) 21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Corinzia Variational Federated Multi-Task Learning and in view of Blundell (U.S. Publication US20200005152A1)
Regarding Claim 21, Corinzia teaches:
A federated learning apparatus, wherein the federated learning apparatus(Corinzia teaches Figure 1, which outlines the framework being used for federated learning.)
receiving, by a second node, a posterior distribution of a parameter in a local model of at least one first node, (Corinzia teaches Algorithm 1 on Page 3, which details a set of C (second client nodes) receiving a s(t) distribution of a parameter from a first node (server) in the federated model. s(t) is a posterior distribution (see Proposition 1 on Page 2, where s(t) is defined).)
and updating, by the second node, a prior distribution of a parameter in a federated model based on the posterior distribution of the parameter in the local model of the at least one first node, wherein the federated model is a machine learning model whose parameter obeys a distribution. (Corinzia teaches Algorithm 1, which teaches optimization (step 8) a prior distribution based on the parameter θ of the server (first node) of a federated learning system.)
Corinzia does not distinctly teach:
may comprise a memory, the memory stores instructions, a processor is configured to execute the instructions stored in the memory, and when the instructions are executed, the processor is configured to perform:
However, Blundell teaches: “Computers suitable for the execution of a computer program can be based on general or special purpose microprocessors or both, or any other kind of central processing unit. Generally, a central processing unit will receive instructions and data from a read-only memory or a random access memory or both. The essential elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices for storing instructions and data. The central processing unit and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.” [0091] which teach using a memory and processor to run computer programs.
Before the effective filing date of the invention it would have been obvious to one of ordinary skill in the art to use the federated learning system of Corinzia with a memory and a processor as taught by Blundell as it is a known method for running computer programs.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Jonathan M Bakhit whose telephone number is (571)272-0454. The examiner can normally be reached Monday - Thursday 8:00AM - 6:00PM EST.
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, Alexey Shmatov can be reached at (571) 270-3428. 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.
/ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123
J.M.B.
Examiner
Art Unit 2123
Prosecution Insights