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 Actions
2. This Final Office Action is responsive to Applicants’ amendments and arguments, as received 4/29/26. Claims 1, 3-6, 9-14, 21-23, 25-27, 33, and 35-36 are pending, of which claims 1, 27, and 33 are independent.
3. Previously-presented claim objection to claim 10 (i.e., item 3 in the Non-Final Office Action dated 12/29/25) and claim rejection under 35 U.S.C. 112(b) to claim 35 (i.e., item 5 in the aforementioned Non-Final Office Action) are withdrawn, responsive to Applicants’ recent reply.
Claim Rejections - 35 USC § 103
4. 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.
5. 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.
6. Claims 1, 3-6, 9-11, 14, 22-23, 26-27, 33, and 36 are rejected under 35 U.S.C. 103 as being unpatentable over Non-Patent Literature “Cell-splitting grid: a self-creating and self-organizing neural network” (“Chow”) in view of Non-Patent Literature “How Lateral Interaction Develops in a Self-Organizing Feature Map” (“Sirosh”).
Regarding claim 1, CHOW teaches A method for constructing a neural network (Abstract on page 373: “A new model of self-creating and self-organizing neural network, cell-splitting grid (CSG), is presented. In this proposed CSG algorithm, the neurons and their connections are created and organized on a 2-D plane according to the input data distribution.”) comprising:
under control of a hardware processor in communication with a non-transitory memory
configured to store executable instructions (section 5, “Experimental Results”, beginning on page 383, in its first paragraph, discussing the application of the CSG algorithm, as one such applied algorithm among many others, to data to compare performance (see, e.g., Tables 1-2 on page 384), which the Examiner reasons would be understood by one of ordinary skill in the art to involve the use of a computer with essential computer elements such as a processor and memory (e.g., a memory to store the algorithms to be executed, such as CSG, and a processor to execute the algorithms) to obtain the performance data as detailed therein):
growing, from one single node, a plurality of layers of a neural network, wherein each layer comprises a plurality of nodes (CSG is understood to start with one neuron, which is subject to splitting (see, e.g., page 377: “The executing steps of the CSG algorithms are as follows: 1. Start from one neuron... 6. When the activation level of winner neuron c decreases to zero, perform the cell-splitting mechanism, i.e., to delete the neuron c and then generate four new neurons within the square region of original neuron c... ”), and see also FIG. 2 illustrating the same on the following page 378 and FIG. 3 on page 379, and where the resulting generated neurons are understood to be children having a different depth/layer, e.g. akin to the discussion provided in sections 4.1-4.2 on pages 382-383); and
self-organizing the plurality of layers of the neural network, using ...
... a learning rule implemented in a higher second layer of the plurality of layers of the neural network connected to the lower first layer of the plurality of layers of the neural network (page 377: “(1) During the processing in the CSG algorithm, the network itself determines the growth of new neurons according to the activation level. Since the network size is not pre-specified, CSG algorithm is flexible for different input data sets. (2) The weight adaptation in the CSG algorithm is adapted slightly within the winner neuron and its direct neighboring neurons. The learning rate is small and does not decrease to zero. This is called dynamic equilibrium.”, which is followed by a setting and adjustment policy for neuron activation level that drives when new neurons are generated therefrom, such that the CSG algorithm is detailed numerically 1-8 (pages 377-378) {the Examiner reasons that the algorithm as detailed and applied is akin to the application of a learning rule as applied that details when and how the network expands node by node and layer by layer}),
to alter inter-layer connectivity between the lower first layer and the higher second layer (weight initialization and weight adjustment via training, per pages 377-378, which the Examiner understands to define the connectivity between nodes in one layer and another),
wherein said growing and said self-organizing are performed over a plurality of iterations (CSG as applied to train a self-creating and self-organizing neural network is iterative, per sections 3.3-3.4, and recursive per section 4.1), and
wherein said growing is performed prior to said self-organizing in each of the plurality of iterations (based on the reference, it reasons that the network must grow new layers with new neurons before those new neurons in those new layers are subject to any self-organizing).
Regarding the amendment, Chow does not teach the further mandatory limitation that the “self-organizing the plurality of layers of the neural network” is “using”, in addition to the above-addressed “learning rule”, also spatiotemporal waves in a lower first layer of the plurality of layers of the neural network, wherein each node in the lower first layer makes an excitatory connection with nodes within a local excitation radius and an inhibitory connection with nodes over a global inhibition radius, thus forming a ring of no connection between an excitation region and an inhibition region. Rather, the Examiner relies upon SIROSH to teach what Chow does not, see e.g., Sirosh:
Abstract discussing “A biologically motivated mechanism for self-organizing a neural network with modifiable lateral connections is presented. The weight modification rules are purely activity-dependent, unsupervised and local. The lateral interaction weights are initially random but develop into a “Mexican hat” shape around each neuron.” Based on the Introduction section, the Examiner understands the self-organizing feature map to be a trained algorithm where neuron’s are weighted as a function of inputs/stimulation. Further, the connection weights of these neurons, as subject to stimulation/excitation and modification, can assume a connection weight that is defined by short-range excitation and long-range inhibition. See Introduction section on page 1360. More detail is provided on the following page 1361, under the headings “The Neuron Model” and “The Network”, which clarify the short-range and long-range excitation/inhibition duality (section A) and where the feature map of these neurons grow into laterally-connected arrangement that is definable in terms of distances d and 3d + 1, for example, as corresponding to excitatory connections and inhibitory connections respectively.
The aforementioned references both relate to self-organizing neural networks that are stimulated to grow and organize as a function of their inputs/stimulation. Hence, they are similarly directed and therefore 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 an expression of the self-organization, as taught by Sirosh, into a framework such as Chow’s, with a reasonable expectation of success, such as to be more biologically realistic in simulating biological processes of interest in the state of the art, as Sirosh details in its Abstract and Introduction sections.
The Examiner notes that Sirosh was selected to read on the newly-made mandatory limitation, in combination with Chow, and in response to Applicants’ amendments and arguments. However, Applicants should take note that the Megardon NPL, also made of record in the 892 form, could also have been cited in Sirosh’s stead.
Regarding claim 3, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the growing, from the one single node, the plurality of layers of the neural network comprises dividing the one single node to generate a daughter node, of the at least one node, in the lower first layer (as cited above with respect to claim 1, Chow’s CSG algorithm is understood to start with one neuron, which is subject to splitting (see, e.g., Chow’s page 377: “The executing steps of the CSG algorithms are as follows: 1. Start from one neuron... 6. When the activation level of winner neuron c decreases to zero, perform the cell-splitting mechanism, i.e., to delete the neuron c and then generate four new neurons within the square region of original neuron c... ”), and see also Chow’s FIG. 2 illustrating the same on the following page 378 and Chow’s FIG. 3 on page 379, and where the resulting generated neurons are understood to be children having a different depth/layer, e.g. akin to the discussion provided in sections 4.1-4.2 on Chow’s pages 382-383).
Regarding claim 4, Chow in view of Sirosh teach the method of claim 3, as discussed above. The aforementioned references further teach the additional limitation
Chow teaches the method of claim 3, as discussed above, further comprising dividing the daughter node in the lower first layer to generate a further daughter node, of the daughter node of the one single node, in the lower first layer (as cited above with respect to claim 1, Chow’s CSG algorithm is understood to start with one neuron, which is subject to splitting (see, e.g., Chow’s page 377: “The executing steps of the CSG algorithms are as follows: 1. Start from one neuron... 6. When the activation level of winner neuron c decreases to zero, perform the cell-splitting mechanism, i.e., to delete the neuron c and then generate four new neurons within the square region of original neuron c... ”), and see also FIG. 2 illustrating the same on the following Chow’s page 378 and FIG. 3 on Chow’s page 379, and where the resulting generated neurons are understood to be children having a different depth/layer, e.g. akin to the discussion provided in sections 4.1-4.2 on Chow’s pages 382-383, and as such, the Examiner reasons that it is within Chow’s teachings to consider a neuron having a child neuron in a next layer after splitting, and then for that child neuron to further be split to create what would essentially be a grandchild to the first neuron).
Regarding claim 5, Chow in view of Sirosh teach the method of claim 3, as discussed above. The aforementioned references further teach the additional limitation
Chow teaches the method of claim 3, further comprising dividing the daughter node in the lower first layer to generate a further daughter node, of the daughter node of the one single node, in the higher second layer (as cited above with respect to claim 1, Chow’s CSG algorithm is understood to start with one neuron, which is subject to splitting (see, e.g., Chow’s page 377: “The executing steps of the CSG algorithms are as follows: 1. Start from one neuron... 6. When the activation level of winner neuron c decreases to zero, perform the cell-splitting mechanism, i.e., to delete the neuron c and then generate four new neurons within the square region of original neuron c... ”), and see also FIG. 2 illustrating the same on the following Chow’s page 378 and FIG. 3 on Chow’s page 379, and where the resulting generated neurons are understood to be children having a different depth/layer, e.g. akin to the discussion provided in sections 4.1-4.2 on Chow’s pages 382-383, and as such, the Examiner reasons that it is within Chow’s teachings to consider a neuron having a child neuron in a next layer after splitting, and then for that child neuron to further be split to create what would essentially be a grandchild to the first neuron).
Regarding claim 6, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the growing, from the one single node, the plurality of layers of the neural network comprises dividing the one single node to generate a daughter node, of the one single node, in the higher second layer (as cited above with respect to claim 1, Chow’s CSG algorithm is understood to start with one neuron, which is subject to splitting (see, e.g., Chow’s page 377: “The executing steps of the CSG algorithms are as follows: 1. Start from one neuron... 6. When the activation level of winner neuron c decreases to zero, perform the cell-splitting mechanism, i.e., to delete the neuron c and then generate four new neurons within the square region of original neuron c... ”), and see also FIG. 2 illustrating the same on the following Chow’s page 378 and FIG. 3 on Chow’s page 379, and where the resulting generated neurons are understood to be children having a different depth/layer, e.g. akin to the discussion provided in sections 4.1-4.2 on Chow’s pages 382-383).
Regarding claim 9, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein an architecture of the lower first layer and higher second layer comprises a pooling architecture, and/or wherein an architecture of two layers of the plurality of layers comprises a pooling architecture (Chow’s page 377, the Examiner understands the following as the pooling of activation inputs in one level/layer until a threshold that warrants a generation/split/divide action creating neurons in a next level/layer: “When the neurons are generated, the new neuron is endowed with an initial value as the activation level. When a neuron is activated, the activation level decreases by a constant value. This process continues until (activation level) of one neuron becomes zero and the neuron is split to generate its four offspring neurons.”).
Regarding claim 10, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein an architecture of the lower first layer and higher second layer comprises an expansion architecture, and/or wherein an architecture of two layers of the plurality of layers comprises an expansion architecture (Chow’s CSG algorithm is explicitly an expansion architecture in the generation of further neurons from existing ones in a next level/layer, see e.g. the discussions provided in sections 4.1-4.2 comparing it with other similar expansion architectures that feature parents/children constructs).
Regarding claim 11, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the lower first layer and/or the higher second layer comprises a square geometry or a rectangular geometry (the geometry taught is square, see e.g., Chow’s Figs. 1-4, for example).
Regarding claim 14, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the neural network comprises a spiking node, and/or wherein the neural network comprises a spiking neural network (section 3.3 starting on Chow’s page 376 discusses Chow’s CSG algorithm, which the Examiner understands to dynamically expand and organize on the basis of activation, with the expansion being specific to the areas of activation as shown per Figs. 1-4 in a manner that is the same as a spiking neural network and spiking nodes as generally understood in the state of the art).
Regarding claim 22, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein said self-organizing comprises applying structural training data to the lower first layer (Chow’s page 376, section 3.2, discussing the operative data to be “input data x distributed in n-dimensional space”).
Regarding claim 23, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the learning rule comprises a local learning rule, and/or wherein the learning rule comprises a dynamic learning rule (Chow’s page 377: “The weight adaptation in the CSG algorithm is adapted slightly within the winner neuron and its direct neighboring neurons. The learning rate is small and does not decrease to zero. This is called dynamic equilibrium.”, which the Examiner reasons is indicative of both local learning (subject to a type of neighbor-proximity) and is characterized explicitly as providing a dynamic learning aspect).
Regarding claim 26, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the hardware processor comprises a neuromorphic processor (Chow’s self-creating and self-organizing neural network is equivalent to a neuromorphic device/construct/processor as is generally understood in the state of the art).
Regarding claim 27, the claim includes the same or similar limitations as claim 1 discussed above, and is therefore rejected under the same rationale. The claim additionally recites a limitation for having executable instructions to perform a task using the neural network, which the Examiner believes is further taught: see, e.g., Chow’s page 375, section 2, first paragraph discussing the use of a trained neural network to “be used for data analysis.”
Regarding claim 33, the claim includes the same or similar limitations as claim 1 discussed above, and is therefore rejected under the same rationale.
Regarding claim 36, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the learning rule is a local learning rule (Chow’s page 377: “The weight adaptation in the CSG algorithm is adapted slightly within the winner neuron and its direct neighboring neurons.”, which the Examiner reasons is indicative of both local learning (subject to a type of neighbor-proximity)).
7. Claims 12-13 and 25 are rejected under 35 U.S.C. 103 as being unpatentable over Chow in view of Sirosh and further in view of CN 104700121 A (“Li”).
Regarding claim 12, Chow in view of Sirosh teaches the method of claim 1, as discussed above. The aforementioned references do not teach the further limitation wherein the lower first layer and/or the higher second layer comprises a non-rectangular geometry. Rather, the Examiner relies upon LI to teach what Chow etc. otherwise lacks, see e.g., Li’s [0004] discussing the use of spherical geometries in a self-organizing map.
Chow and Li both relate to self-organizing neural network structures, with an aim to improve performance. Hence, the references are similarly directed and therefore analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to vary the geometry as Li considers, with a reasonable expectation of success, as to try and improve the performance accordingly as mentioned per [0004].
Regarding claim 13, Chow in view of Sirosh and further in view of Li teach the method of claim 12, as discussed above. The aforementioned references teach the further limitation wherein the non-rectangular geometry comprises an annulus geometry, a spherical geometry, and/or disk geometry, wherein the disk geometry has a hyperbolic distribution (Li’s [0004] as discussed just above per claim 12). The motivation for combining the references is as discussed above in relation to claim 12.
Regarding claim 25, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references do not teach the further limitation comprising training a classifier connected to the plurality of layers and/or the neural network. Rather, the Examiner relies upon LI to teach what Chow etc. otherwise lacks, see e.g., Li’s [0029] discussing a classification method as related to its self-organizing map/network.
Chow and Li both relate to self-organizing neural network structures, with an aim to improve performance. Hence, the references are similarly directed and therefore analogous. Chow’s page 375, section 2, first paragraph, discusses the use of its self-created and self-organized trained neural network to “... for data analysis.” Li, as mentioned here, is more explicitly directed to classification, which the Examiner reasons is a type of data analysis. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply a self-created and self-organizing map/network, per Chow, for a data analysis task that is directed to classification, as is widely practiced in the state of the art, with a reasonable expectation of success.
8. Claim 21 is rejected under 35 U.S.C. 103 as being unpatentable over Chow in view of Sirosh and further in view of Non-Patent Literature “Recurrently connected and localized neuronal communities initiate coordinated spontaneous activity in neuronal networks” (“Lonardoni”).
Regarding claim 21, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references do not teach the further limitation further comprising generating the spatiotemporal waves based on noisy interactions between nodes of the first layer of the plurality of layers of the neural network. Rather, the Examiner relies upon LONARDONI to teach what Chow etc. otherwise lacks, see e.g., Lonardoni’s pages 6-7, sections titled “Simulations of spontaneous and pharmacologically manipulated neuronal culture activities” and “Simulated spontaneous spiking activity in the neuronal network model”, discussing the reproduction of observed pharmacologically manipulated neural network activity in a simulated spiking model, where the observations are inclusive of spatiotemporal patterns among the neurons that the Examiner reasons as equivalent to the recitation for “spatiotemporal waves based on noisy interactions” between nodes of a same layer.
Chow and Lonardoni both relate to self-organizing neural network structures. Hence, the references are similarly directed and therefore analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to aim to incorporate and/or implement a sort of rigorous reproduction of neural behavior as observed, per Lonardoni, in the sort of optimized comparable neural network contemplated by Chow with a reasonable expectation of success.
13. Claim 35 is rejected under 35 U.S.C. 103 as being unpatentable over Chow in view of Sirosh and further in view of Non-Patent Literature “Neuromorphic Computing: The Potential for High-Performance Processing in Space” (“Bersuker”).
Regarding claim 35, Chow in view of Sirosh teach the method of claim 1, as discussed above. The aforementioned references do not teach the further limitation wherein the spatiotemporal waves in a lower first layer of the plurality of layers further comprises at least one malfunctioning node. Rather, the Examiner relies upon BERSUKER to teach what Chow etc. otherwise lacks, see e.g., Bersuker’s page 8, right column in the second-to-last paragraph, discussing the capability of neuromorphic networks (such as Chow’s) to be fault tolerant.
Chow and Bersuker both relate to self-organizing neural network structures. Hence, the references are similarly directed and therefore analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to aim to incorporate and/or implement a neural network in the manner described by Chow to produce a fault tolerant network as engineered by Bersuker with a reasonable expectation of success.
Conclusion
14. The prior art made of record and not relied upon is considered pertinent to Applicants’ disclosure:
Non-Patent Literature “An adaptable neuromorphic model of orientation selectivity based on floating gate dynamics” (Gupta)
Non-Patent Literature “Limitations of short range Mexican Hat connection for driving target selection in a 2D neural field: activity suppression and deviation from input stimuli” (Megardon)
Non-Patent Literature “Multi-map self-organization for sensormotor learning: a cortical approach” (Menard)
15. Applicants’ amendment necessitated the new ground(s) of rejection presented in this Office Action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicants are 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.
16. 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
Prosecution Insights