Notice of Pre-AIA or AIA Status
1. 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
2. This Non-Final Office Action is responsive to Applicants’ after-final reply received 11/19/25. Claims 1-9 remain pending of which claims 1 and 6-7 are independent.
3. This Action has the effect of withdrawing and replacing the Advisory Action mailed 12/17/25, which Applicants have asserted is improper given the substance of their after-final reply.
In this Action, the Zafar reference that was previously detailed in the Advisory Action and its 892 form (mailed 12/17/25) is now being asserted in an obviousness rejection, as provided below. As the reference was previously presented and provided by the Examiner in the communication mailed 12/17/25, the Examiner is not providing an additional 892 form with essentially the same substance again at this time.
4. In view of Applicants’ arguments provided via the after-final reply, the Examiner is withdrawing the rejections under 35 U.S.C. 101 previously presented in the Final Office Action dated 8/28/25.
Claim Rejections - 35 USC § 103
5. 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.
6. 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.
7. Claims 1-2 and 6-9 are rejected under 35 U.S.C. 103 as being unpatentable over Non-Patent Literature “Beginner’s Guide to Capsule Networks” (“Zafar”) in view of Non-Patent Literature “Normalization” (“FreeText Library”).
Regarding claim 1, ZAFAR teaches An information processing apparatus (Zafar, as directed to implementing a capsule network as used for example to recognizing objects in a scene (discussed generally on pages 1-2), the training and inference aspects of the capsule network as described necessarily involves execution on a computer (where the Examiner understands the routing discussion on pages 3-5 to detail the training and inference of the capsule network)) comprising:
a memory that stores a machine learning model of a vector neural network type (Zafar’s computer, implicit to the teachings provided therein, would be understood to feature a memory to store the model as trained and as used in inference); and
one or more processors that execute an arithmetic operation using the machine learning model (Zafar’s computer, implicit to the teachings provided therein, would be understood to feature processor elements to perform the various training and inference computations discussed on pages 2-5 for example), wherein the machine learning model has a plurality of vector neuron layers each including a plurality of nodes (page 3, under the heading “Routing by agreement”, teaching that the capsule network has several layers of capsules which the Examiner understands to be comprised of neurons based on the discussion on pages 2-5),
when one of the plurality of vector neuron layers is referred to as an upper layer and a vector neuron layer below the upper layer is referred to as a lower layer (the discussion on pages 2-5 makes reference to layer l and layer l+1, which the Examiner believes reads on adjacent upper and lower layers as recited),
the one or more processors are configured to execute outputting one output vector by using output vectors from the plurality of nodes of the lower layer as an input for each node of the upper layer (page 4, under the heading “The dynamic routing algorithm”, teaching that activations in layer l+1 are based on activations in layer l, a determination that involves the calculation of an output vector for the lower layer l), the outputting including:
when any node of the upper layer is referred to as a target node (a calculation for v in a particular capsule in the layer l+1, as discussed on page 4, is a calculation for a target node in an upper layer in accordance with the dynamic routing algorithm as taught),
(a) obtaining a prediction vector based on a product of the output vector of each node of the lower layer and a prediction matrix (page 4’s calculation of the prediction vector as a product of lower level capsule outputs and a corresponding weight (i.e., see step 1 on page 4)),
(b) obtaining a sum vector based on a linear combination of the prediction vectors obtained from each node of the lower layer (step 2 on page 4: “the output vector is the weighted sum of all the prediction vectors given by the capsules of layer l for the capsule j”),
(c) obtaining a normalization coefficient (coupling coefficient, ci, per pages 4-5, as determined for each capsule) by normalizing a norm of the sum vector (page 5, all capsules in a layer are subject to a softmax function so that the sum is 1, i.e., “a norm of the sum vector” as recited, and further per page 4, it is taught that the output vector (which is a sum vector per page 4 step 2) is subject to a squashing function to get an activation vj for the capsule j (where the squashing function is a further normalizing of the sum vector, which is already subject to normalization via the softmax, i.e., to put it simply: a norm of a norm value)).
Zafar does not appear to teach the further limitation for (d) obtaining the output vector of the target node by dividing the sum vector by the norm and then multiplying the divided sum vector by the normalization coefficient. Rather, the Examiner relies upon previously-cited FREETEXT LIBRARY to teach what Zafar otherwise lacks, see e.g., FreeText Library’s page 1, teaching normalization of vectors, where the vector is divided by its magnitude to create a vector with a length of 1 (“unit vector”), and then a multiplication step to scale the unit vector is also taught (see, e.g., the discussion on the same page beginning with “An important application of normalization is to rescale a vector to a particular magnitude ...”).
Zafar and FreeText Library both relate to the handling of vector data as pertaining to comparable multi-layer neural networks. Hence, the references are similarly related and analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate FreeText Library’s normalization and scaling aspect into FreeText Library, with a reasonable expectation of success, such that the data as subject to FreeText Library’s scaling lends itself to more reliable handling regardless of any assumptions as to scaling, as would be generally understood in the state of the art (e.g., see the previously-referenced Lindgren NPL as mentioned in the prior Office Action in this same regard).
Regarding claim 2, Zafar in view of FreeText Library teach the information processing apparatus according to claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the normalization coefficient is obtained by normalizing the norm with a normalization function so that a total sum of the normalization coefficients in the upper layer becomes 1 (see Zafar as discussed above in relation to claim 1 and particularly the step (c)). The motivation for combining the references is the same as discussed above in relation to claim 1.
Regarding claim 6, the claim includes the same or similar limitations as claim 1 discussed above, and is therefore rejected under the same rationale.
Regarding claim 7, the claim includes the same or similar limitations as claim 1 discussed above, and is therefore rejected under the same rationale.
Regarding claim 8, Zafar in view of FreeText Library teach the information processing apparatus according to claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the normalization coefficient indicates a relative output intensity of the target node with respect to other nodes in the upper layer (Zafar’s weight and activation as computed, for each capsule, indicates the strength of that capsule’s signal in that layer relative to other capsules in the same layer). The motivation for combining the references is the same as discussed above in relation to claim 1.
Regarding claim 9, Zafar in view of FreeText Library teach the information processing apparatus according to claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the normalization coefficient is obtained by applying a normalization function to the norm of the sum vector, the normalization function incorporating information regarding all nodes in the upper layer (see Zafar as discussed above in relation to claim 1 and particularly the step (c)). The motivation for combining the references is the same as discussed above in relation to claim 1.
8. Claims 3-5 are rejected under 35 U.S.C. 103 as being unpatentable over Zafar in view of FreeText Library and further in view of WO 2019/083553 A1 (“Hinton”).
Regarding claim 3, Zafar in view of FreeText Library teach the information processing apparatus according to claim 1, as discussed above. The aforementioned references teach the further limitations wherein a plurality of the prediction matrices are prepared (Zafar’s application of per-capsule weight values, as discussed above in relation to claim 1 and particularly the step (a) – which cites to Zafar’s step 1 on page 4), a range of the plurality of nodes of the lower layer used for an arithmetic operation of the output vector of each node of the upper layer is limited by convolution using a kernel which has the plurality of prediction matrices as a plurality of elements (the weight values per capsule as taught by Zafar, e.g. as referenced just above, are part of the computation that determines activation of the capsule for sending signals to a next layer).
The references do not teach the limitation that the plurality of prediction matrices are determined by performing learning of the machine learning model and rather the Examiner relies upon HINTON to teach what Zafer etc. otherwise lack, see e.g., Hinton’s [63]-[64] discussing network parameters are adjusted as part of training the network, which is inclusive of the weight matrix instances per [0064]: “The parameters of the network 100 that are adjusted during the training include parameters of the neuron layers (e.g., the convolutional neuron layer 104), the capsule layers (e.g., the primary capsule layer 108, the convolutional capsule layer 112, and the class capsule layer 118), and the routing system 120” especially when read in view of [65]’s clarification that matrices are included as parameters.
Zafar and FreeText Library both relate to the handling of vector data as pertaining to comparable multi-layer neural networks. Hinton is similarly directed. Hence, the references are similarly related and analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate more detailed aspects of Hinton’s learning/training aspects into the comparable network of Zafar as modified per claim 1, with a reasonable expectation of success, such that a practiced approach to determining weights and the like per Hinton is applicable to the same/similar features per Zafar.
Regarding claim 4, Zafar in view of FreeText Library teach the information processing apparatus according to claim 1, as discussed above. The aforementioned references do not teach the additional limitations wherein the memory stores a known feature vector group obtained from an output of at least one specific layer of the plurality of vector neuron layers when a plurality of teacher data are input to the learned machine learning model (Hinton’s [64] discussing that “the class membership of the input is known” as pertaining to each input in a training data set, and accordingly for that to be known and used to evaluate a loss function as later discussed in the same paragraph, it follows that at minimum the known information for the teacher/training data is stored, available, etc. to be useful for retrieval and then comparison in that loss function evaluation, and as discussed in relation to claim 1 (Hinton: [18], [41], [56], [58], [65]), one would understand that the data itself can be vectors rather than single dimensional), and
the one or more processors are configured to perform an arithmetic operation of a similarity between a feature vector obtained from the output of the specific layer when new input data is input to the learned machine learning model and the known feature vector group (Hinton’s [64] discussing the loss function).
Zafar and FreeText Library both relate to the handling of vector data as pertaining to comparable multi-layer neural networks. Hinton is similarly directed. Hence, the references are similarly related and analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate more detailed aspects of Hinton’s learning/training aspects into the comparable network of Zafar as modified per claim 1, with a reasonable expectation of success, such that a practiced approach to determining weights and the like per Hinton is applicable to the same/similar features per Zafar.
Regarding claim 5, Zafar in view of FreeText Library and further in view of Hinton teach the information processing apparatus according to claim 4, as discussed above. The aforementioned references teach the additional limitations wherein the specific layer has a configuration in which vector neurons disposed in a plane defined by two axes of a first axis and a second axis, are disposed as a plurality of channels along a third axis in a direction different from the two axes (Hinton’s [4], [74], [78], [97] discussing that the dimensionality can be more than 1, e.g. see [41]’s discussion that neuron arrangement may be a three dimensional arrangement having length, width, and depth), and
the feature vector is one of
(i) a first type feature spectrum in which a plurality of element values of an output vector of a vector neuron at one plane position of the specific layer are arranged over the plurality of channels along the third axis (Hinton’s [4], [74], [78], [97] discussing that the dimensionality is can be more than 1, e.g. see [41]’s discussion that neuron arrangement may be a three dimensional arrangement having length, width, and depth),
(ii) a second type feature spectrum obtained by multiplying each of the element values of the first type feature spectrum by the normalization coefficient ([95] teaches the use of a normalized routing factor between capsules of different layers, where the normalization is applied per [96] to a sum (as discussed per [75], the sum itself involving the taking of an output from each node/capsule in a lower layer)), and
(iii) a third type feature spectrum in which the normalization coefficient at one plane position of the specific layer is arranged over the plurality of channels along the third axis.
The motivation for combining the references is the same as discussed above in relation to claim 1.
Conclusion
9. Any inquiry concerning this communication or earlier communications from the examiner should be directed to SHOURJO DASGUPTA whose telephone number is (571)272-7207. The examiner can normally be reached M-F 8am-5pm CST.
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, Tamara Kyle can be reached at 571 272 4241. 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.
/SHOURJO DASGUPTA/Primary Examiner, Art Unit 2144