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 .
DETAILED ACTION
This action is responsive to the following communication: Amendment filed Sept. 22, 2026. This Action is made Final.
Claims 1-20 are pending in the case. Claims 1, 13 and 17 are independent claims.
Claim Rejections - 35 USC § 103
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.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Zong et al. (hereinafter Zong) “Communication Reducing Quantization for Federated Learning with Local Differential Privacy Mechanism” 2021 in view of BOROUJENI et al. (hereinafter BORO) U.S. Patent No. 2022/0101130 and further in view of Voelker et al. (hereinafter Voe) U.S. Patent No. 2021/0133568.
With respect to independent claim 1, Zong teaches a method, comprising:
receiving a first local model and a first random quantization instruction; training the first local model to obtain a model update amount (see e.g., Section II.B – “Model aggregation and broadcasting: The server decodes the received … and recovers the discrete value into the estimated value … Then the server performs model aggregation according to (2) and broadcasts the new global model”); performing, based on the first random quantization instruction, random quantization on the model update amount to obtain a quantization model update amount (see e.g., Section II.B – Step 3 and 4); encoding the quantization model update amount to obtain encoded data (see e.g., Section II.B – Step 3 and 4); sending the encoded data (see e.g., Section II.B – “Then the server performs model aggregation according to(2)and broadcasts the new global model”); and receiving, in response to sending the encoded data, a second local model and a second random quantization instruction (see e.g., Section II.B Step 2-5 are repeated until the training procedure converges or the loss is lower than a preset threshold).
Zong does not expressly show the features discussed below. However, BORO teaches a sever that receives and sends (see e.g., Fig. 3 and Para [55][56] – “the base station 110 may transmit a machine learning component to the UEs 120. As shown, the UEs 120 may each include a first communication manager 320. “) the first local model is a neural network parameter model (see e.g., Para [54]-[58]), and wherein the first random quantization instruction(see e.g., Para [72]-[77] [83]-[86]– “ the UE 120 may determine a quantized value, based at least in part on the feedback, using randomization with probabilities based at least in part on respective distances between one or more values of the feedback and a plurality of quantized digits.”) performing, based on the first random quantization instruction (see e.g., Para [76][83]) and using a quantizer (see e.g., Para [77] – UE selects discrete digits and maps to probability), random quantization on the model update amount in the first random quantization manner to obtain a quantization model update amount (see e.g., Para [88] – “the UE 120 may transmit, and the base station 110 may receive, the quantized value. The base station 110 may determine an update based at least in part on the quantized value. ”), wherein the quantizer maps the model update amount to a predefined data grid (see e.g., Para [77] – “the UE 120 may select −3, −1, 1, and 3 as the quantized digits” and input feedback probabilities are mapped to those digits) and performing iterative updating by repeating, using the second local model and the second random quantization instruction (see e.g., Para [89] – “the base station 110 may update the machine learning component based at least in part on the update. In some aspects, the base station 110 may determine a plurality of updates … In some aspects, example 400 may be recursive, where the base station 110 re-transmits an updated machine learning component for additional training by the UE 120 and/or other UEs in the federated learning.”), the training, the random quantization, the encoding, and the sending until a global model converges (see e.g., Para [65] – “The second communication manager 330 may update the global machine learning component using multiple rounds of updates from the UEs 120 until a global loss function is minimized. ”). Both Zong and BORO are directed to federated learning process. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having Zong and BORO in front of them to modify the system of Zong to include the above feature. The motivation to combine Zong and BORO comes from BORO. BORO discloses the motivation to transmit a quantized value based at least in part on the feedback and the quantized value is determined using randomization so that communication can be enhanced. (see e.g. Abstract and Para [3]-[10]). This motivation for combination also applies to the remaining claims which depend on this combination.
Zhong-BORO does not expressly show instruction that indicates a random quantization manner of a plurality of random quantization manners, the plurality of random quantization manners comprising a random step quantization manner and a random quantizer manner. However, Voe teaches similar feature (see e.g., Para [11][55][106] -“ Temporal dithering can be embodied in at least four different techniques for creating augmented neural nonlinearities, all of which involve using the time-step parameter ω to determine the magnitude of any multi-bit spike emitted by a neuron (spike magnitudes can also importantly be interpreted as an integer-valued “number of spikes” emitted by a neuron on a given timestep). The first technique of dithered quantization involves applying a neural activation function, multiplying by ω, updating the neuron's voltage variable, and rounding this variable down to the nearest integer to compute the magnitude of the spike (if any) at the current timestep. The second technique of stochastic rounding involves applying a neural activation function and multiplying by ω as before, but additionally computes a truncation error and compares this error to a random sample from the uniform distribution between 0 and 1 to determine whether to round up or down when determining the magnitude of the spike ...” The dithered quantization corresponds to “a random step quantization manner” and the technique of stochastic rounding corresponds to a random quantizer manner.).”) Both Zong and Voe are directed to quantizing neural network information. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having Zong and Voe in front of them to modify the system of Zong to include the above feature. The motivation to combine Zong and Voe comes from Voe. Voe discloses the motivation to use multiple alternative quantization algorithms so that selection can be made to better fit the system constraints. (see e.g. Abstract and [11][55][106]). This motivation for combination also applies to the remaining claims which depend on this combination.
With respect to dependent claim 2, the modified Zong teaches the first random quantization instruction comprises random step quantization instruction (see e.g., Algorithm 1, 1.15-22 and Du col. 8-10).
With respect to dependent claim 3, the modified Zong teaches performing the random quantization further comprises performing, by using a random step quantization method, the random quantization (see e.g., Algorithm 1, 1.15-22 Du col. 8 and 10).
With respect to dependent claim 4, the modified Zong teaches the random step quantization method comprises random quantization steps that are randomly and evenly distributed (see e.g., Algorithm 1, 1.15-22 and page 78 – “The dither vectors … are randomized from a uniform distribution over …”).
With respect to dependent claim 5, the modified Zong teaches performing the random quantization comprises further performing, by using a random quantizer, the random quantization (see e.g., Algorithm 1, 1.15-22 and Du col. 8 and 10).
With respect to dependent claim 6, the modified Zong teaches the random quantizer comprises an upward quantizer and a downward quantizer (Zong-Du does not expressly show the “upward quantizer” and “downward quantizer.” However, this recited feature is merely a design/implementation choice, therefore it would have been obvious for a person of ordinary skill in the art to try).
With respect to dependent claim 7, the modified Zong teaches the first random quantization instruction comprises a random quantizer instruction, and wherein performing the random quantization comprises performing, based on the random quantizer instruction, the random quantization (see e.g., Du col. 8 and 10 – “a dither noise vector v.sub.k,tϵR.sup.d may be added to the quantization to randomize the quantization effects”).
With respect to dependent claim 8, the modified Zong teaches performing the random quantization further comprises further performing, by using a random step quantization method, the random quantization (see e.g., Algorithm 1, 1.15-22 and Du col. 8 and 10 – “a dither noise vector v.sub.k,tϵR.sup.d may be added to the quantization to randomize the quantization effects”).
With respect to dependent claim 9, the modified Zong teaches the random step quantization method comprises random quantization steps that are randomly and evenly distributed (see e.g., Algorithm 1, 1.15-22 and page 78 – “The dither vectors … are randomized from a uniform distribution over …”).
With respect to dependent claim 10, the modified Zong teaches performing the random quantization further comprises further performing, by using a random quantizer, the random quantization (see e.g., Algorithm 1, 1.15-22 Du col. 8 and 10).
With respect to dependent claim 11, the modified Zong teaches the random quantizer comprises an upward quantizer and a downward quantizer (Zong-Du does not expressly show the “upward quantizer” and “downward quantizer.” However, this recited feature is merely a design/implementation choice, therefore it would have been obvious for a person of ordinary skill in the art to try).
With respect to dependent claim 12, the modified Zong teaches using a plurality of types of third random quantization instructions in the iterative updating (see e.g., Section II.B Step 2-5).
With respect to independent claim 13, the modified Zong teaches an apparatus, comprising: a memory configured to store instructions; and one or more processors coupled to the memory and configured to execute the instructions to:
deliver, to a plurality of terminals, a first global model is a neural network parameter model (see e.g., BORO Para [54]-[58] – The motivation to combine is discussed above with respect to claim 1), wherein the first global model comprises a plurality of local models, and wherein the plurality of local models are in one-to-one correspondence with the plurality of terminals(see e.g., Section II.B – Step 1); deliver, to each terminal of the plurality of terminals, a random quantization instruction (see discussion above with respect to claim 1) that indicates a random quantization manner of a plurality of random quantization manners, the plurality of random quantization manners comprising a random step quantization manner and a random quantizer manner (see e.g., Voe Para [11][55][106] -“ Temporal dithering can be embodied in at least four different techniques for creating augmented neural nonlinearities, all of which involve using the time-step parameter ω to determine the magnitude of any multi-bit spike emitted by a neuron (spike magnitudes can also importantly be interpreted as an integer-valued “number of spikes” emitted by a neuron on a given timestep). The first technique of dithered quantization involves applying a neural activation function, multiplying by ω, updating the neuron's voltage variable, and rounding this variable down to the nearest integer to compute the magnitude of the spike (if any) at the current timestep. The second technique of stochastic rounding involves applying a neural activation function and multiplying by ω as before, but additionally computes a truncation error and compares this error to a random sample from the uniform distribution between 0 and 1 to determine whether to round up or down when determining the magnitude of the spike ...” The dithered quantization corresponds to “a random step quantization manner” and the technique of stochastic rounding corresponds to a random quantizer manner.).” The motivation to combine is discussed above with respect to claim 1). receive, from the plurality of terminals, encoded data (see e.g., Section II.B – Step 4); decode the encoded data to obtain quantization model update amounts of the plurality of terminals (see e.g., Section II.B – Step 5 Algorithm 1); aggregate the quantization model update amounts to obtain a second global model (see e.g., Page 76 and Algorithm 2); wherein additional quantization errors respectively introduced into the quantization model update amounts(see e.g., Voe [6] [54] – “Algorithms 2 and 3 are similar, but do not require maintaining an additional state-variable (i.e., the voltage) across time-steps. Instead, they sample random distributions to approximate this state-variable at each moment in time, thus distributing the quantization error over time (rather than tracking and pushing it onto the next time-step, as in Algorithm 1). Algorithm 4 specifies an alternative to Algorithm 1 that may be applied to any spiking neuron model instead of any nonlinear function.” The motivation to combine is discussed above with respect to claim 1) in accordance with the random quantization instruction cancel one another when the quantization model update amounts are aggregated (see e.g., Voe Para [6] [54] [87]-[89]– “but additionally computes a truncation error and compares this error to a random sample from the uniform distribution between 0 and 1 to determine whether to round up or down when determining the magnitude of the spike” “the underlying spiking neuron model is normalized such that it may only spike at most once across any time-step of length ω=1 (this is discussed elsewhere in more detail) … (a) Zero-Mean Error “); and deliver, to the plurality of terminals, the second global model to instruct the plurality of terminals to perform iterative updating until the second global model converges (see e.g., Section II.B – Steps 2-5).
With respect to dependent claim 14, the modified Zong teaches the random quantization instruction comprises a random step quantization instruction or a random quantizer instruction, and wherein the random step quantization instruction and the random quantizer instruction are for obtaining the quantization model update amounts by using a random step quantization method or a random quantizer (see e.g., Algorithm 1, 1.15-22 Du col. 8 and 10).
With respect to dependent claim 15, the modified Zong teaches the random step quantization method comprises random quantization steps that are randomly and evenly distributed (see e.g., Algorithm 1, 1.15-22 and page 78 – “The dither vectors … are randomized from a uniform distribution over …”).
With respect to dependent claim 16, the modified Zong teaches the random quantizer comprises an upward quantizer and a downward quantizer (Zong-Du does not expressly show the “upward quantizer” and “downward quantizer.” However, this recited feature is merely a design/implementation choice, therefore it would have been obvious for a person of ordinary skill in the art to try).
Claim 17 is rejected for the similar reasons discussed above with respect to claim 1.
With respect to dependent claim 18, the modified Zong teaches the first random quantization instruction comprises a random step quantization instruction or a random quantizer instruction, and wherein the one or more processors are further configured to execute the instructions to perform, based on the random step quantization instruction or the random quantizer instruction and by using a random step quantization method or a random quantizer, the random quantization (see e.g., Algorithm 1, 1.15-22 Du col. 8 and 10).
With respect to dependent claim 19, the modified Zong teaches the random step quantization method comprises random quantization steps that are randomly and evenly distributed (see e.g., Algorithm 1, 1.15-22 and page 78 – “The dither vectors … are randomized from a uniform distribution over …”).
With respect to dependent claim 20, the modified Zong teaches the random quantizer comprises an upward quantizer and a downward quantizer, and wherein a first quantity of upward quantizers is the same as a second quantity of downward quantizers (Zong-Du does not expressly show the “upward quantizer” and “downward quantizer.” However, this recited feature is merely a design/implementation choice, therefore it would have been obvious for a person of ordinary skill in the art to try).
It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. “The use of patents as references is not limited to what the patentees describe as their own inventions or to the problems with which they are concerned. They are part of the literature of the art, relevant for all they contain.” In re Heck, 699 F.2d 1331, 1332-33, 216 USPQ 1038, 1039 (Fed. Cir. 1983) (quoting In re Lemelson, 397 F.2d 1006, 1009, 158 USPQ 275, 277 (CCPA 1968)). Further, a reference may be relied upon for all that it would have reasonably suggested to one having ordinary skill the art, including nonpreferred embodiments. Merck & Co. v. Biocraft Laboratories, 874 F.2d 804, 10 USPQ2d 1843 (Fed. Cir.), cert. denied, 493 U.S. 975 (1989). See also Upsher-Smith Labs. v. Pamlab, LLC, 412 F.3d 1319, 1323, 75 USPQ2d 1213, 1215 (Fed. Cir. 2005); Celeritas Technologies Ltd. v. Rockwell International Corp., 150 F.3d 1354, 1361, 47 USPQ2d 1516, 1522-23 (Fed. Cir. 1998).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PEIYONG WENG whose telephone number is (571)270-1660. The examiner can normally be reached on Mon.-Fri. 8 am to 5 pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Matthew Ell, can be reached on (571) 270-3264. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://portal.uspto.gov/external/portal. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/PEI YONG WENG/Primary Examiner, Art Unit 2141
