DETAILED ACTION
1. This office action is in response to the Application No. 19328206 filed on 03/17/2026. Claims 7 and 14 has been cancelled. Claims 1-6 and 8-13 are presented for examination and are currently pending.
Response to Arguments
2. The arguments regarding the prior art rejection has been considered but are moot because a new secondary reference has been added to address the newly added limitations.
The 101 rejection is maintained and adjusted to reflect the amended claims.
On pages 6-7 of the remarks regarding the 101 rejection, the Applicant argued
that “Step 2A, Prong One: The Claims Do Not Recite a Judicial Exception. The Examiner asserts that the recited operations constitute mental processes because they involve "observation" and "making a judgement." Applicant respectfully disagrees. The mental process grouping is properly limited to operations that can practically be performed in the human mind. As the USPTO's August 2025 Memorandum to Technology Centers 2100, 2600, and 3600 expressly instructs, examiners may not expand the mental process category to cover limitations that cannot practically be performed mentally. As amended, Claim 1 recites operations that are computationally intensive by their nature and have no practical human mental analog: converting a cognition event to a vector space representation; transforming the vector space representation into a geometric representation on a continuous, differentiable mathematical space whose shape is determined by timing delays and connection weights across a plurality of connections; processing, by a geometric reasoning engine, the geometric representation by following geodesic paths of the thought manifold; and dynamically adjusting timing delays and connection weights to change the geometric shape of the thought manifold in response to processed information. Claim 1 now expressly requires a geometric reasoning engine performing geodesic path computation on a continuous differentiable manifold. While humans can perform differential calculations using pen and paper, they cannot practically compute geodesic paths over a continuous differentiable manifold, solve the underlying differential geometric equations, or dynamically adjust connection weights across a plurality of manifold data points with a speed anywhere close to what would be needed to imitate cognition. Manual calculations of this nature would require days or weeks to produce a single trajectory in any geometric space of sufficient complexity to process a cognitive event and produce a cognitive output as those terms are defined in the specification”.
The argument above is not persuasive because according to the detailed analysis of the claimed invention in the 101 rejection limitations like “convert the cognition event to a vector space representation of the cognition event”, “transform the vector space representation into a geometric representation on a thought manifold” or “applying a non-linear manifold learning operation that preserves semantic relationships between data points as geometric properties of the thought manifold ...” are all abstract ideas that encompasses processes that can be performed in the human mind.
Furthermore, according to 2106.04(a)(2)(III), “Nor do the courts distinguish between claims that recite mental processes performed by humans and claims that recite mental processes performed on a computer. As the Federal Circuit has explained, “[c]ourts have examined claims that required the use of a computer and still found that the underlying, patent-ineligible invention could be performed via pen and paper or in a person’s mind.” Here, while actual computer network infrastructure may be involved in the underlying invention, it is still performable by the human mind.
In addition, the 101 detailed analysis in this Office Action is in line with the August 2025 memorandum.
On page 7 of the remarks, the Applicant argued that “Further, the specification makes explicit that geodesic computation requires solving the equation
PNG
media_image1.png
30
240
media_image1.png
Greyscale
operations that are computational, not cognitive. The Examiner's characterizations (e.g., "observing the vector space and making a judgement on the geometric representation") improperly reduce concrete computational operations to informal mental analogs, which is precisely the analytical error the August 2025 guidance cautions against.”
The above argument is not persuasive because the equation mentioned above by the Applicant is not claimed. It appears the Applicant is arguing what is not claimed.
On pages 7-8 of the remarks, the Applicant argued that “Lastly, the examiner's classification of the limitations regarding the thought manifold as falling under the judicial exception of "mathematical concepts" is misplaced. Those limitations do not recite a mathematical concept, but merely involve a mathematical concept. MPEP §2106.04.II.A.1 states: Examiners should accordingly be careful to distinguish claims that recite an exception (which require further eligibility analysis) and claims that merely involve an exception (which are eligible and do not require further eligibility analysis). ... The claims do not claim an invention that operates according to some mathematical principle (e.g., "operate in accordance with F=ma"); they describe an environment in which a portion of the invention operates which happens to involve geometric math. Therefore, the following limitations do not recite abstract ideas: thought manifold comprises a continuous, differentiable mathematical space having a geometric shape (Mathematical concepts directed to a continuous, differentiable mathematical space that has a geometric shape); the thought manifold comprises a plurality of data points within the continuous, differentiable mathematical space (Mathematical concepts directed to a continuous, differentiable mathematical space of plurality of data points). The August 2025 guidance further instructs that a claim does not "recite" an abstract idea merely because it involves mathematical concepts, unless the claim explicitly sets forth equations or formulas as claim limitations. Claim 1 does not recite any mathematical formula as a claim limitation. It recites structural and operational features of a specific system. Accordingly, the claims do not recite a judicial exception and the rejection should be withdrawn at Step 2A, Prong One”.
The above argument is not persuasive because the claim limitations like “thought manifold comprises a continuous, differentiable mathematical space having a geometric shape” or “thought manifold comprises a continuous, differentiable mathematical space having a geometric shape” etc. are clearly mathematical concepts which are classified as abstract ideas.
Furthermore, according to MPEP 2106.04(a)(2)(I): “It is important to note that a mathematical concept need not be expressed in mathematical symbols, because "[w]ords used in a claim operating on data to solve a problem can serve the same purpose as a formula”.
On page 9 of the remarks, the Applicant argued that “ Even assuming arguendo that any claim limitation recites an abstract idea, the claims as a whole integrate that idea into a practical application. The August 2025 guidance instructs that at Step 2A, Prong Two, the claim must be evaluated as a whole for whether it reflects a technological solution to a technological problem, including whether it improves the functioning of a computer system or other technology. Claim 1 does exactly this. The specification identifies a specific technical problem: existing AI systems operating in discontinuous, anisotropic, topologically fractured vector spaces are incapable of coherent geometric reasoning. The claimed system solves this problem through a specific architectural solution, transformation of vector space representations onto a continuous, differentiable manifold and processing by a geometric reasoning engine along geodesic paths, that produces an improved form of computational cognition unavailable in prior systems. This is a specific improvement to computer-implemented Al architecture, not a generic application of an abstract idea to a computer. Evaluation of claims as a whole means that the claims means that the examiner must consider all elements of the claims, including limitations that recite an abstract idea, additional elements, and any routine or conventional components in determining whether integration into a practical application exists: ...”.
Furthermore, on page 10 of the remarks, the Applicant argued that “Thus, the Examiner's Prong Two analysis evaluates the non-abstract elements individually and dismisses them piecemeal as extra-solution activity or environmental linking. This approach is inconsistent with the governing guidance, which requires evaluation of the claim as a whole. Considered as a whole, the claim recites a specific system architecture with defined components performing defined operations to achieve a concrete technical result: a cognitive output produced by geometric reasoning on a thought manifold whose shape is dynamically modified by the processing itself. The claims are integrated into a practical application and are not directed to an abstract idea”.
The Above argument is not persuasive because the Applicant’s argument are directed to the abstract ideas. The claimed limitations “transformation of vector space representations onto a continuous, differentiable manifold”, “geodesic path” are abstract ideas. The “geometric reasoning engine” is an additional element.
The limitation “geometric reasoning engine” under the broadest reasonable interpretation, is an instruction to implement the abstract ideas of “the geometric representation on the thought manifold by following one or more geodesic paths along the geometric shape”. As a result, the claim by the Applicant that these abstract ideas (“transformation of vector space representations onto a continuous, differentiable manifold”, “geodesic path”) improves the computation of cognitive systems is not persuasive.
Furthermore, the Applicant needs to include additional details by incorporating in the argument how the highlighted additional elements in the 101 rejection in addition to the abstract ideas improves the technological field or provides a solution that improves the functioning of the computer.
In addition, the Examiner has evaluated the claimed limitation’s abstract ideas and additional elements individually and as a whole in the detailed 101 analysis in this Office Action. However, the claimed limitations do not integrate the abstract ideas into practical application.
On page 10 of the remarks, the Applicant argued that “Even under Step 2B, the claims recite significantly more than any alleged judicial exception. As amended, Claim 1 recites a geometric reasoning engine performing geodesic path computation on a continuously evolving thought manifold whose geometric shape is defined by and modified through the timing delays and connection weights of its connections. The Examiner has not identified any evidence that this combination of features, a geometric reasoning engine, geodesic path processing on a dynamically reconfigured manifold, and manifold shape modification through connection weight and timing delay adjustment, is well-understood, routine, or conventional. Under MPEP 2106.05(d), the Examiner bears the burden of providing such evidence, and none has been provided. Furthermore, the claims do not preempt the alleged abstract ideas. Geometric reasoning, manifold mathematics, and machine learning can all be practiced in numerous ways that do not involve the specific architecture claimed, including conventional neural networks, vector-space-based AI systems, and statistical learning models, none of which would infringe Claim 1. The absence of preemption further supports eligibility”.
On page 10 of the remarks, the Applicant argued that “For the foregoing reasons, the §101 rejection should be withdrawn in its entirety. The claims do not recite a judicial exception at Step 2A, Prong One, because the recited operations cannot practically be performed in the human mind and therefore fall outside the mental process grouping under both the governing guidance and the USPTO's August 2025 Memorandum. Even assuming arguendo that a judicial exception is recited, the claims integrate it into a practical application at Step 2A, Prong Two, by providing a specific architectural solution to the identified technical problem of incoherent reasoning in discontinuous vector spaces. And even under Step 2B, the claims recite significantly more than any alleged exception, as the Examiner has provided no evidence that the claimed combination of a geometric reasoning engine, geodesic path processing, and dynamically reconfigured manifold geometry is well-understood, routine, or conventional. Applicant respectfully requests that the Examiner withdraw the §101 rejection and advance the claims to allowance”.
The claim by the Applicant that the Office Action analyzes the combination of feature like a geometric reasoning engine, geodesic path processing on a dynamically reconfigured manifold, and manifold shape modification through connection weight and timing delay adjustment as a well-understood, routine, or conventional is not persuasive. The 101 rejection did not classify any of the above features argued by the Applicant as a well-understood, routine, or conventional.
According to the detailed analysis of the 101 rejection in this Office Action, the limitation “geometric reasoning engine” under the broadest reasonable interpretation is an additional element which is directed to mere instructions to implement the judicial exception (i.e., one or more geodesic paths along the geometric shape of the thought manifold). The limitations “one or more geodesic paths along the geometric shape ...” and “changes the geometric shape of the thought manifold to represent the influence of the information contained in the cognition event on the thought manifold by changing one or more of the timing delays or connection weights between two or more of the data points” are all abstract ideas. As a result, none of the limitations argued by the Applicant is classified as a well-understood, routine, or conventional.
In addition, the limitations “geodesic path processing on a dynamically reconfigured manifold” argued by the Applicant is not claimed. It appears the Applicant is arguing what is not claimed.
It is important to note, the judicial exception alone cannot provide the improvement. The improvement can be provided by one or more additional elements, MPEP 2106.05(a). Here, the argument that the claimed invention provides a specific solution to the technical problem of incoherent reasoning in discontinuous vector spaces is not persuasive because the Applicant has not provided detailed explanation of how the additional elements highlighted in the 101 rejection including the abstract ideas as a whole solve the incoherent reasoning in discontinuous vector spaces.
In conclusion, the additional elements in the 101 rejection analysis do not integrate the abstract ideas into practical application or significantly more than a judicial exception. As a result, the 101 rejection is maintained.
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.
3. Claims 1-6 and 8-13 are rejected under 35 U.S.C 101 because the claimed invention is directed towards an abstract idea without significantly more.
Step 1
Independent claim 1 is directed to a system, and falls into one of the four statutory categories.
Step 2A, Prong 1
Claim 1 recites the following abstract ideas:
convert the cognition event to a vector space representation of the cognition event (Mental process directed to conversion of an event to a vector. This may include observation of the event and making a judgement of the vector pace representation);
transform the vector space representation into a geometric representation on a thought manifold (Mental process directed to transformation of the vector space into geometric representation. This may include observing the vector space and making a judgement on the geometric representation of the thought manifold) wherein:
thought manifold comprises a continuous, differentiable mathematical space having a geometric shape (Mathematical concepts directed to a continuous, differentiable mathematical space that has a geometric shape);
by applying a non-linear manifold learning operation that preserves semantic relationships between data points as geometric properties of the thought manifold, such that the resulting geometric representation supports geodesic path computation thereon (Mental process directed to applying non-linear manifold learning which preserves data points as geometric properties. This can be performed with the use of pen and paper),
the thought manifold comprises a plurality of data points within the continuous, differentiable mathematical space (Mathematical concepts directed to a continuous, differentiable mathematical space of plurality of data points) and
the geometric shape is determined at least in part by the timing delays and connection weights of the plurality of connections (Mental process directed to determining the geometric shape using timing delays and connection weights. This include observing the time delays and connection weights and making a judgement on the geometric shape based on the observed time delays and connection weights); and
process the geometric representation on the thought manifold by following one or more geodesic paths along the geometric shape of the thought manifold, each geodesic path comprising one or more of the connections between the plurality of data points on the thought manifold, each path comprising one or more of the connections between the plurality of data points on the thought manifold (Mental process directed to the process of the geometric representation on the thought manifold by following paths along the geometric shape and geodesic path. This include observation of the paths along the geometric shape of the thought manifold), and
each path being determined at least in part by the influence of the geometric representation on the plurality of data points (Mental process directed to determining a path. This include observation of the geometric representation on the plurality of data points and making a judgement on the path based on the observed geometric representation on the plurality of data points);
changes the geometric shape of the thought manifold to represent the influence of the information contained in the cognition event on the thought manifold by changing one or more of the timing delays or connection weights between two or more of the data points (Mental process directed to making changes of the thought manifold to represent an influence in the information in the cognitive event. This include observation of the geometric shape of the thought manifold and making a judgement to adjust the geometric shape in the cognitive event based on the observed information).
Step 2A, Prong 2
Claim 1 recites the following additional elements:
receive a cognition event from a cognitive edge source (this limitation is directed to insignificant extra solution activity of data transmission. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(g));
by a geometric reasoning engine (This limitation is directed to the instructions to apply the judicial exception. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)));
wherein the processing of the geometric representation on the thought manifold:
produces a cognitive output associated with the cognition event (this limitation is directed to insignificant extra solution activity of data transmission. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(g)); and
a plurality of connections between the data points each of the plurality of connections comprising a timing delay and a connection weight; and (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)),
Step 2B
Claim 1 recites the following additional elements:
receive a cognition event from a cognitive edge source (this limitation is directed to insignificant extra solution activity of data transmission and it is well understood routine and conventional. This does not amount to significantly more than judicial exception. See MPEP 2106.05(d)(II), example i);
by a geometric reasoning engine (This limitation is directed to the instructions to apply the judicial exception. This does not amount to significantly more than judicial exception. See MPEP 2106.05(f)));
wherein the processing of the geometric representation on the thought manifold:
produces a cognitive output associated with the cognition event (this limitation is directed to insignificant extra solution activity of data transmission and it is well understood routine and conventional. This does not amount to significantly more than judicial exception. See MPEP 2106.05(d)(II), example i); and
a plurality of connections between the data points (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h)),
4. Dependent claim 2 is directed to a system, and falls into one of the four statutory categories.
Claim 2 do not recite any abstract ideas.
Claim 2 recites the following additional elements:
the thought manifold is implemented on a neuromorphic platform (this limitation is directed to mere instruction to apply an exception. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(f));
the data points are neurons of the neuromorphic platform (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h));
the one or more connections between data points are connections between neurons of the neuromorphic platform (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)); and
the one or more paths are pathways between the neurons of the neuromorphic platform (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 2 recites the following additional elements:
the thought manifold is implemented on a neuromorphic platform (this limitation is directed to mere instruction to apply an exception. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(f));
the data points are neurons of the neuromorphic platform (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h));
the one or more connections between data points are connections between neurons of the neuromorphic platform (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h)); and
the one or more paths are pathways between the neurons of the neuromorphic platform (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h)).
5. Dependent claim 3 is directed to a system, and falls into one of the four statutory categories.
Claim 3 do not recite any abstract ideas.
Claim 3 recites the following additional elements:
wherein the neuromorphic platform comprises a spiking neural network (this limitation is directed to mere instruction to apply an exception. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(f)).
Claim 3 recites the following additional elements:
wherein the neuromorphic platform comprises a spiking neural network (this limitation is directed to mere instruction to apply an exception. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(f)).
6. Dependent claim 4 is directed to a system, and falls into one of the four statutory categories.
Claim 4 do not recite any abstract ideas.
Claim 4 recites the following additional elements:
wherein the neuromorphic platform comprises a neuromorphic chip which acts as a physical embodiment of the thought manifold (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 4 recites the following additional elements:
wherein the neuromorphic platform comprises a neuromorphic chip which acts as a physical embodiment of the thought manifold (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h)).
7. Dependent claim 5 is directed to a system, and falls into one of the four statutory categories.
Claim 5 do not recite any abstract ideas.
Claim 5 recites the following additional elements:
wherein the neuromorphic chip is configured as a spiking neural network (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(h)).
Claim 5 recites the following additional elements:
wherein the neuromorphic chip is configured as a spiking neural network (this limitation is directed to generally linking the use of a judicial exception to a particular technological environment or field of use. This limitation does not amount to significantly more than the judicial exception. See MPEP 2106.05(h)).
8. Dependent claim 6 is directed to a system, and falls into one of the four statutory categories.
Claim 6 recites the following abstract ideas:
the one or more paths are derived from calculations of geometric paths between the data points on the geometric space of the thought manifold (Mental process directed to deriving paths from calculation of geometric paths. This may include observation of the data points between the geometric space and making a judgement on derivation of the paths).
Claim 6 recites the following additional elements:
wherein: the thought manifold is implemented as a digital representation in standard computing technology (this limitation is directed to amount to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not integrate the abstract idea into a practical application. See MPEP 2106.05(h));
the data points and the one or more connections between data points are
stored in a relational database (this limitation is directed to the storage of data points and connections which is insignificant extra solution activity of mere data gathering. This limitation does not integrate the abstract idea into a practical application. See MPEP 2106.05(g)); and
Claim 6 recites the following additional elements:
wherein: the thought manifold is implemented as a digital representation in standard computing technology (this limitation is directed to amount to generally linking the use of a judicial exception to a particular technological environment or field of use. This does not amount to significantly more than judicial exception. See MPEP 2106.05(h));
the data points and the one or more connections between data points are
stored in a relational database (this limitation is directed to the storage of data points and connections which is insignificant extra solution activity of mere data gathering and it is well understood routine and conventional. This limitation does not amount to significantly more than judicial exception. See MPEP 2106.05(d)(II), example i); and
9. Independent claim 8 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 8, it is substantially similar to claim 1, and is rejected in the
same manner and reasoning applying.
10. Dependent claim 9 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 9, it is substantially similar to claim 2, and is rejected in the
same manner and reasoning applying.
11. Dependent claim 10 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 10, it is substantially similar to claim 3, and is rejected in
the same manner and reasoning applying.
12. Dependent claim 11 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 11, it is substantially similar to claim 4, and is rejected in
the same manner and reasoning applying.
13. Dependent claim 12 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 12, it is substantially similar to claim 5, and is rejected in
the same manner and reasoning applying.
14. Dependent claim 13 is directed to a method, and falls into one of the four statutory categories.
With regards to claim 13, it is substantially similar to claim 6, and is rejected in
the same manner and reasoning applying.
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.
15. Claims 1-5 and 8-12 are rejected under 35 U.S.C. 103 as being unpatentable over Maslennikov et al. ("Nonlinear dynamics and machine learning of recurrent spiking neural networks." Physics-Uspekhi 192 (2022): 1089-1109.) in view of Law et al. (US20240126811 filed 01/17/2023)
Regarding claim 1, Maslennikov teaches a computer system … (In addition, the recurrent neural network in Fig. 1a belongs to the class of so-called reservoir computing systems (pg. 1024, right col., second para.); It has been found that recurrent spiking networks can be a reservoir capable of learning tasks related to the autonomous and stimulus-induced generation of various space-time patterns (pg. 1020, left col., first para.); The SpiNNaker system simulates spiking neural networks in real time by connecting more than 1 million ARM processors, which enable simulation of 1 billion neurons with a biologically realistic number of connections (1,000±10,000 synapses per neuron) and with a time resolution of 1 ms, pg. 1033, left col., last para.) to:
receive a cognition event (An alternative approach, top-down, starts from the analysis of specific transformations of input stimuli into output responses performed by a neural network (pg. 1036, left col., last para.). The Examiner notes input stimuli is a cognition event) from a cognitive edge source (The recurrent architecture inherent in most parts of the brain is characterized by the presence of feedback loops, which allows the network not only to be directly influenced by input stimuli but also to respond to its own behavior in the past. Spiking dynamics, i.e., the ability to generate action potentials or spikes, is one of the most important properties of neurons, which underlies communication between neurons and the processing of large amounts of data, pg. 1021, left col., third para. The Examiner notes the edge source is the brain);
convert the cognition event to a vector space representation of the cognition event (the model-network neurons receive two independent noisy signals as inputs, simulating sensory sensations about the predominant color and direction of movement of the cloud of dots, respectively. The network also receives a context input in the form of a 2D vector corresponding to the context signal in the experiment, which tells the network what specifically should be determined: color or direction of movement (Fig. 2b). After training, the network reports its decision in response to incoming stimuli, generating a binary response (pg. 1026, left col., second para.); The corresponding system, schematically presented in Fig. 1a, consists of a layer of input neurons, the recurrent network itself, and a layer of output neurons (pg. 1024, left col., second to the last para.); The system of rate neurons is described by the equation
PNG
media_image2.png
34
488
media_image2.png
Greyscale
where x ϵ RN is the state vector of the neural network nodes, which determine the rates of their activity through the activation function
PNG
media_image3.png
28
270
media_image3.png
Greyscale
pg. 1024, right col., first para.);
transform the vector space representation into a geometric representation on a thought manifold (Figure 2(c). Trajectories of the model system in two contexts and different intensities of input stimuli in the projection onto the choice axes, colors, and movement, pg. 1025; In the projection onto the three-dimensional subspace defined by these axes, the model dynamic qualitatively reproduce the key properties of the neurophysiological data from the experiment (Fig. 2c) (pg. 1026, left col., last para.); (d) Projection of the multidimensional activity of the neural network onto the space of three principal components, Figure 1(d), pg. 1024)
wherein: the thought manifold comprises a continuous, differentiable mathematical space having a geometric shape (It is noteworthy that system (11) describes the dynamics of the network in continuous time (pg. 1024, right col., second para.); Fig. 3(d) Singular points and trajectories of the phase space of the model system in perceiving two visual stimuli with a delay and making a decision about the match/mismatch of their categories, pg. 1026, Fig. 3);
the thought manifold comprises a plurality of data points within the continuous, differentiable mathematical space and a plurality of connections between the data points, each of the plurality of connections comprising a timing delay and a connection weight (At the beginning of the test, the trajectory is in a stable state of equilibrium (gray oblique cross); after the input stimulus is presented, the trajectory leaves its vicinity in the direction of one of the states that correspond to a particular category (red and blue circles) (Fig. 3d). Further, during the delay interval, a transition occurs to the neighborhood of stable or saddle points corresponding to two categories (red and blue stars) (pg. 1027, left col., second para.); At the same time, the recurrent-network weights define the direction (the so-called recurrent manifold) along which the trajectories move between the initial and final states, pg. 1028, left col., first para.); and
the geometric shape is determined at least in part by the timing delays and connection weights of the plurality of connections ((d) Projections of activity onto space of the three principal components (pg. 1027, Fig. 4); It has been found that two sets of stable equilibrium states are formed: Finit, corresponding to the beginning of the test, and Fterminal, corresponding to its end. The contextual stimulus initializes the network state in the vicinity of the equilibrium state Finit, then the input rectangular stimulus drives the trajectory out of its basin of attraction, and the trajectory relaxes to the vicinity of the stable equilibrium state Fterminal at a rate determined by the value of the contextual input stimulus, which affects the position of the points Finit and Fterminal along the direction called the input stimulus manifold, which is determined by the input weight vector, pg. 1028, left col., first para.); and
process by a geometric reasoning engine, the geometric representation on the thought manifold by following one or more paths along the geometric shape of the thought manifold, each path comprising one or more of the connections between the plurality of data points on the thought manifold (First, in the process of receiving sensory input, a kind of accumulation of evidence in favor of one of two alternative solutions occurs. The population response trajectory moves along the choice axis. Second, the larger the values of the color and direction signals, the further the trajectories go from the choice axis along the corresponding axes, pg. 1026, left col., last para. to pg. 1026, right col., first para.), and
each path being determined at least in part by the influence of the geometric representation on the plurality of data points ( … trajectories associated with different contexts are spatially isolated from each other. This behavior of the system is determined by the special phase space trajectories that emerged in the course of training (pg. 1026, right col., first para.); Under the action of the input signal, the trajectory moves away from the corresponding line, and the stronger on average the input signal, the further it moves, pg. 1026, right col., second para.);
wherein the processing of the geometric representation on the thought manifold ((a) Diagram of a trained recurrent neural network. (b) Example of target space-time pattern that the network generates as an output after successful training. (c) Corresponding dynamics of recurrent network neurons that underlie autonomous generation of the specified pattern, Fig. 1a-c, pg. 1024):
produces a cognitive output associated with the cognition event (Similar projection when the network converts short input stimuli into output responses on the corresponding output element, Fig. 1e, pg. 1024); and
changes the geometric shape of the thought manifold to represent the influence of the information contained in the cognition event on the thought manifold by changing one or more of the timing delays or connection weights between two or more of the data points (It should be noted that the points where the trajectory direction changes, which are highlighted in Fig. 1d in different colors, correspond to the activity peaks of the output neurons in Fig. 1b displayed in the same color (pg. 1025, left col., first para.); In particular, special trajectories can be identified in a multidimensional phase space that correspond to the execution of a particular function, and the distribution of weights, clustering of connections, and presence of modules in the structure of a trained network can be analyzed, pg. 102, left col., first para.).
Maslennikov does not explicitly teach a computer system configured to execute software instructions that cause the computer system to:, transform the vector space representation into a geometric representation on a thought manifold by applying a non-linear manifold learning operation that preserves semantic relationships between data points as geometric properties of the thought manifold, such that the resulting geometric representation supports geodesic path computation thereon, the geometric representation on the thought manifold by following one or more geodesic paths along the geometric shape of the thought manifold, each geodesic path comprising one or more of the connections between the plurality of data points on the thought manifold,
Law teaches a computer system configured to execute software instructions that cause the computer system to (In at least one embodiment, framework 100 includes a computer readable storage medium and/or code stored on said computer readable storage medium in a form of a computer program including a plurality of computer readable instructions executable by one or more processors [0153]):
transform the vector space representation into a geometric representation on a thought manifold (generate geometric coordinates corresponding to elements of data, wherein said geometric coordinates indicate each of dependencies among said data (e.g., between said elements of data) [0061]) by applying a non-linear manifold learning operation that preserves semantic relationships between data points as geometric properties of the thought manifold (In at least one embodiment, spacetime representation 106 encodes or otherwise represents one or more nodes of initial graph representation 104 as one or more events in a Minkowski space (e.g., a special case of a general relativity of a spacetime isometric to
PNG
media_image4.png
38
29
media_image4.png
Greyscale
1 d and a geometry induced on each fixed tangent space of a given Lorentz manifold) [0075]), such that the resulting geometric representation supports geodesic path computation thereon (In at least one embodiment, only distances (e.g., geodesics) between events represented by spacetime representation 106 are extracted 107 [0076]; In at least one embodiment, said extracted distances include Lorentzian distances (e.g., geodesics in Lorentzian space) [0077]),
and process, by a geometric reasoning engine, the geometric representation on the thought manifold by following one or more geodesic paths along the geometric shape of the thought manifold (In at least one embodiment, coordinates indicating a direction of a dependency of data include geometric coordinates ... In at least one embodiment, said coordinates are mapped to a spacetime in which said direction is taken as a time orientation of an arc or line segment between two events in said spacetime, such as a geodesic [0060]),
each geodesic path comprising one or more of the connections between the plurality of data points on the thought manifold (In at least one embodiment, a neighborhood is convex if a unique geodesic is definable between each pair of points within said neighborhood [0078]), and
each path comprising one or more of the connections between the plurality of data points on the thought manifold (In at least one embodiment, a relationship (e.g., a geodesic arc) between events is predicted based, at least in part, on nodes corresponding to said events belonging to respective convex normal neighborhoods [0081])
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Maslennikov to incorporate the teachings of Law for the benefit of using geometric coordinates which can be relied upon to accurately predict dependencies among said data and directions of said dependencies even when cycles (e.g., directed cycles) exist among said dependencies (Law [0062])
Regarding claim 2, Modified Maslennikov teaches the computer system of claim 1, Maslennikov teaches wherein: the thought manifold is implemented on a neuromorphic platform; the data points are neurons of the neuromorphic platform (Another key task in the construction of a neuromorphic system is the implementation of connections between elements, which are carried out in biological neural networks through a synaptic contact Ð a microscopic gap between two neurons, pg. 1032, right col., first para.);
the one or more connections between data points are connections between neurons of the neuromorphic platform; and the one or more paths are pathways between the neurons of the neuromorphic platform (It is assumed that the development of neuromorphic devices, in which the basic principles of the operation of neural networks of the brain, such as spike dynamics and synaptic plasticity, and which have a comparable number of neurons and a topology of connections close to reality, will result in an enhanced ability to self-organize, learn, and perform various classes of tasks, pg. 1032, left col., first para.).
Regarding claim 3, Modified Maslennikov teaches the computer system of claim 2, Maslennikov teaches wherein the neuromorphic platform comprises a spiking neural network (Nonlinear dynamics and machine learning of recurrent spiking neural networks, title).
Regarding claim 4, Modified Maslennikov teaches the computer system of claim 2, Maslennikov teaches wherein the neuromorphic platform comprises a neuromorphic chip which acts as a physical embodiment of the thought manifold (To build a complete Neurogrid system, 16 Neurocore chips are placed on a board and arranged in a tree structure, pg. 1033, left col., first para.).
Regarding claim 5, Modified Maslennikov teaches the computer system of claim 4, Maslennikov teaches wherein the neuromorphic chip is configured as a spiking neural network (Neurogrid, a mixed analog-digital system that can simulate in real time networks consisting of several million spike neurons and several billion synaptic connections … To build a complete Neurogrid system, 16 Neurocore chips are placed on a board and arranged in a tree structure, pg. 1032, left col., last sentence to pg. 1033, left col., first para.).
Regarding claim 8, claim 8 is similar to claim 1. It is rejected in the same manner and reasoning applying.
Regarding claim 9, claim 9 is similar to claim 2. It is rejected in the same manner and reasoning applying.
Regarding claim 10, claim 10 is similar to claim 3. It is rejected in the same manner and reasoning applying.
Regarding claim 11, claim 11 is similar to claim 4. It is rejected in the same manner and reasoning applying.
Regarding claim 12, claim 12 is similar to claim 5. It is rejected in the same manner and reasoning applying.
16. Claim 6 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Maslennikov et al. ("Nonlinear dynamics and machine learning of recurrent spiking neural networks." Physics-Uspekhi 192 (2022): 1089-1109) in view of Law et al. (US20240126811 filed 01/17/2023) and further in view of Gütig ("Spiking neurons can discover predictive features by aggregate-label learning." Science 351.6277 (2016): aab4113)
Regarding claim 6, Modified Maslennikov teaches the computer system of claim 1, Maslennikov does not explicitly teach wherein: the thought manifold is implemented as a digital representation in standard computing technology; the data points and the one or more connections between data points are stored in a relational database; and the one or more paths are derived from calculations of geometric paths between the data points on the geometric space of the thought manifold.
Gütig teaches wherein: the thought manifold (The mean response … of a neuron to a feature activity pattern y during and after learning was measured through dedicated batches of probe trials during which the learning rule was not engaged. The index y refers either to a particular feature … from the set of discrete features or to the feature activity pattern of a particular feature parameter … when feature activity patterns belonged to a continuous manifold, pg. aab4113-9, right col., section: Neural responses) is implemented as a digital representation in standard computing technology (I used the TIDIGITS speech corpus, which also contains sequences of digits of variable length (ranging from 2 to 7). With all utterances of single digits reserved to test performance, training was based on unsegmented digit sequences only. Specifically, I used the number of occurrences of a single target digit as aggregate-label to train neurons to discriminate between this target and the remaining distractor digits. I found that multispike tempotrons reliably learned to fire a single spike whenever their target digit occurred and to remain silent otherwise (Fig. 3A), pg. aab4113-3, middle col.);
the data points and the one or more connections between data points are stored in a relational database (In Fig. 3, I used continuous digit sequences from the TIDIGITS database to train multi-spike tempotrons with aggregate-label learning to detect one out of the 11 English digit words “zero,” “oh,” “one,” “two,” “three,” “four,” “five,” “six,” “seven,” “eight,” and “nine,” which served as the clue. For each training example, the neuron’s supervisory signal consisted only of the aggregate-label, the number of times that the clue occurred within the speech sequence. … In contrast, I defined the training set as the remaining 55 connected digit sequences from each speaker, pg. aab4113-12, left col., last para to middle col.); and
the one or more paths are derived from calculations of geometric paths between the data points on the geometric space of the thought manifold (RESULTS: To implement aggregate-label learning, I calculated how neurons should modify their synaptic efficacies in order to most effectively adjust their number of output spikes. Because a neuron’s discrete number of spikes does not provide a direction of gradual improvement, I derived the multi-spike tempotron learning rule in an abstract space of continuous spike threshold variables. In this space, changes in synaptic efficacies are directed along the steepest path, reducing the discrepancy between a neuron’s fixed biological spike threshold and the closest hypothetical threshold at which the neuron would fire a desired number of spikes, pg. 1041, middle col., last para.).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of Modified Maslennikov to incorporate the teachings of Gütig for the benefit of unsupervised neural networks to discover reoccurring constellations of sensory features even when they are widely dispersed across space and time (Gütig, abstract)
Regarding claim 13, claim 13 is similar to claim 6. It is rejected in the same manner and reasoning applying.
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 extension fee 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 date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MORIAM MOSUNMOLA GODO whose telephone number is (571)272-8670. The examiner can normally be reached Monday-Friday 8am-5pm 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, Michelle T Bechtold can be reached on (571) 431-0762. 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.
/M.G./Examiner, Art Unit 2148
/MICHELLE T BECHTOLD/ Supervisory Patent Examiner, Art Unit 2148
Prosecution Insights