DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . This action is in response to the instant application filed 05/24/2024. Claims 1-20 are pending.
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
Claim Rejections - 35 USC § 102
(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) 1-3, 5-11, 13-20 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Yi (FedLoRA: Model-Heterogeneous Personalized Federated Learning with LoRA Tuning).
Regarding claim 1, YI teaches a server configured to store a plurality of neural network models; and an electronic device configured to be connected to the server, comprising: a storage device configured to store an adapter master model; a communication interface configured to connect to the server; and a processor coupled to the storage device and the communication interface, and configured to output the adapter master model and input data to the server (section 1, A typical FL system consists of a central FL server and multiple participants (a.k.a., FL clients). The server broadcasts a global model to clients, who then further train the received model on their local data and upload the resulting model back to the server, Clients participating in FL can be mobile edge devices, clients upload the adapter models which is essentially outputting the input data and the adapter models that make up the master adapter model), wherein the server embeds the adapter master model into each of the plurality of neural network models (Figure 2, the adapter model is inserted into the client model right after the encoder), inputs the input data to the plurality of neural network models to generate output data by the plurality of neural network models embedded with the adapter master model, and transmits the output data to the electronic device (Figure 1, an output produced using the network along with the embedded adapter).
Regarding claim 2, Yi teaches wherein the adapter master model comprises a plurality of adapter models (Figure 2, the global adapter model comprises of the local adapter models), each adapter model of the plurality of adapter models comprises a weight matrix, and the server combines the weight matrix of each adapter model with an original weight matrix of a dense layer of a neural network model of the plurality of neural network models corresponding to the adapter model (Figure 1, Adapter uses weight matrices A and B which are combined with the weight matrix of the client model, Figure 4, the weight matrix is of a dense layer, FC1).
Regarding claim 3, Yi teaches wherein a number of parameters of the weight matrix is smaller than a number of parameters of the original weight matrix (Section 4.3, A low-rank adapter is an inherently “dimension-reduced” version of a local heterogeneous model, i.e., it contains far fewer parameters than the local heterogeneous model), and a dimension of the weight matrix is the same as a dimension of the original weight matrix (Figure 1, AB has the same dimensions as the original weight matrix).
Regarding claim 5, Yi teaches wherein the electronic device outputs training data to the server in advance, to input the training data to the plurality of neural network models embedded with the adapter master model, wherein the server trains the plurality of adapter models corresponding to the plurality of neural network models based on the training data (Alg 1, the clients (device) use the training data to train a local adapter and send the local adapter to the server in the previous iteration, then the server starts a new iteration where the clients input the training data into the local adapters to train them), and the server transmits at least one weight data generated by training the plurality of adapter models to the electronic device to cause the processor to update the adapter master model (Section 4.2, Eq. 10, After receiving the local homogeneous adapters, the server aggregates them like FedAvg to update the global adapter).
Regarding claim 6, Yi teaches wherein when the server is training the plurality of adapter models, the original weight matrices of the plurality of neural network models remain unchanged (Figure 3, image 2 Freeze Model, Train Adapter).
Regarding claim 7, Yi teaches wherein the server trains the plurality of adapter models based on a loss function, wherein the loss function is a result of a sum of products of multiplying a plurality of sub-loss functions output by respective neural network models of the plurality of neural network models embedded with the adapter master model by a plurality of corresponding coefficients, and a sum of the plurality of coefficients equals 1 (Eq 5, loss is the sum of l1 and l2 which is output by the adapter and the client model, the two losses are multiplied by coefficients that add up to 1).
Regarding claim 8, Yi teaches wherein, in response to the processor receiving an operation instruction, the processor notifies the server according to the operation instruction to remove the adapter master model (Alg 1, after every iteration the server removes the adapter master model from the client models and replaces it with an updated version in the next iteration).
Regarding claims 9-11, 13-16, 17-20, the claims recite essentially the same limitations as the claims above and as such are rejected for at least the same reasons.
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.
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) 4 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Yi as applied to claims 2 and 10 above, and further in view of Mahabadi (Parameter-efficient Multi-task Fine-tuning for Transformers via Shared Hypernetworks).
Regarding claim 4, Yi teaches the data processing system according to claim 2 but fails to teach wherein the adapter master model further comprises an encoding module, wherein the encoding module comprises a plurality of encoders, the plurality of encoders generate a plurality of feature parameters based on the input data, and the encoding module establishes a relation matrix based on the plurality of feature parameters, wherein inputs to the plurality of adapter models are generated based on the relation matrix.
Mahabadi teaches wherein the adapter master model further comprises an encoding module, wherein the encoding module comprises a plurality of encoders (Figure 1, The compact HYPERFORMER++ shares the same hypernetworks across all layers and tasks and computes the task embedding based on task, layer id, and position of the adapter module, the encoding module comprises hypernetworks h_LN and h_A), the plurality of encoders generate a plurality of feature parameters based on the input data, and the encoding module establishes a relation matrix based on the plurality of feature parameters, wherein inputs to the plurality of adapter models are generated based on the relation matrix (Section 4, projection matrices (Ul τ and Dl τ), Figure 1, the weights are matrices that are input into the layers of the adapter models).
Mahabadi and Yi are analogous to the claimed invention because they are in the field of using adapter models for language models. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have used the hypernetwork system in Mahabadi along with the system in Yi to train task specific adapters and enable the model to adapt to individual task (Mahabadi Abs).
Regarding claim 12, the claim recites essentially the same limitations as claim 4 and is rejected for at least the same reasons listed above.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NATNAEL A ASEGDEW whose telephone number is (571)270-0407. The examiner can normally be reached 7:30-5.
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, Kakali Chaki can be reached at (571) 272-3719. 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.
/NATNAEL A ASEGDEW/ Examiner, Art Unit 2122
/MICHAEL H HOANG/ PRIMARY EXAMINER, Art Unit 2122
Prosecution Insights