DETAILED ACTION
This action is responsive to the Applicant’s response filed 4/16/26.
As indicated in Applicant’s response, claims 1, 9-11, 13-14, 20 have been amended. Claims 1-20 are pending a next office action.
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.
Claim 1 is rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. Claim(s) 1 is/are directed to Abstract Idea. The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the 2-step analysis as follows.
Step I: claim 1 is directed to a Process category of subject matter.
Step II-A:
Per prong one: the elements recited as selecting (modules) and identifying modules (from text from a drawn model), generating a definition from the drawn model and generating code may be considered steps than can be performed conceptually pen/paper or a human mind - Mental Process – MPEP 2106.04(c ) -- whereas the “by utilizing a neural network” does not change the fundamental nature of the conceptual steps being performed as “neural networks” themselves are built upon mathematical models or algorithms also included as subgroups of the Judicial Exception identified by the MPEP. (§ 2106.04(a1))
Per prong two: The claim does not improve the underlying functioning of the computer(processor) itself or a specific technical field (MPEP 2106.04(d1)); instead it merely uses a computer, neural network as a tool to automate a software-based task. The step of establishing a connection between modules from a drawn model can be integrated as subset of the conceptual steps identified in prong one, whereas the step of “executing a task” by utilizing a AI model amounts to a broad, generic post-solution activity that fails to provide a meaningful technical expertise or limitation, nor does it specify how the execution improves any technical field, or entails a inventive step above a mere “executing” expressed in a broad sense.
The claim based on the above fails to integrate the Abstract Idea into a practical Application.
Step II-B:
The additional elements such as computer method, processor, modules, code and generic mention of a “neural network” - MPEP 2106.05 (d) - can be seen as a mere effect of appending generic HW or numerical tools to the Abstract Idea, which cannot amount to presence of an inventive concept. No teachings are provided to demonstrate that a neural network as claimed particularly selects and connects the modules toward solving a technical bottleneck in SW compilation or deployment for optimizing memory or reducing footprint, or improving execution speed.
From the standpoint of arrangement of elements in combination – MPEP 2106.05(e) the sequence of elements follows the standard flow of designing and running software (selection, identifying, using tool and generating definition, code and executing code, and as such, no non-conventional or unexpected synergy is being perceived so that this synergic arrangement stands out in the realm of SW automation.
The additional elements fail to render the Abstract Idea of claim 1 significantly more than itself.
Claim 1 is directed to a Judicial Exception under the 35 USC 101 statute.
Claim 13 is rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. Claim(s) 13 is/are directed to Abstract Idea. The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the 2-step analysis as follows.
Step I: claim 13 is directed to a process type of category.
Step IIA:
Prong one: the elements recited as extracting text form a drawn AI model, generating a model definition and generating code for node correlating with the text can be viewed as activities that can be performed via pen/paper (code or model definition) using a human mind performing conceptual derivation (extraction) or analytic (correlation between drawing, text and programming). The neural network presented in a generic, broad term fails to change the nature of the abstracted method, when neural network as mathematical tool is a well understood means. The claim is directed to a Mental process type Exception that uses Mathematical Concepts as a means – MPEP 2106.04(c ); MPEP 2106.04 (a1)
Prong two:
The element recited in claim 13 as receiving drawn model, and executing task based on model definition and modules code are all pre or post-activity of no significance that can transform the conceptual activities identified in prong One of the Abstract Idea. The claimed method in all, merely uses a computer, neural network as a tool to automate a software-based task merely uses a computer, drawn text, neural network as a tool to automate a software-based task; and applying tools to support activities construed as Abstract Idea - – MPEP 2106.04 (d1) - cannot integrate the Abstract idea into a practical application.
Step II-B:
The additional elements such as modules, code and generic mention of a “neural network” for “executing a definition … to perform a task”- MPEP 2106.05 (d) - can be seen as a mere effect of appending generic HW or numerical tools to the Abstract Idea or rather as a way of applying of a Abstract Idea using conventional SW engineering concepts, computer and mathematical-based means; and this cannot provide an inventive concept to the method, as no part of the additional elements (from their high level of generality) is depicting a non-conventional limitation being imparted into the Abstract idea.
Per MPEP § 2106.05(e), the sequence of the additional elements as recited, simply follows the standard, logical workflow of designing and running software. No non-conventional or unexpected synergy is recited that takes it out of the realm of generic automation. In all the additional elements cannot render the Abstract Idea significantly more than itself.
Claim 20 is rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. Claim(s) 13 is/are directed to Abstract Idea. The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the 2-step analysis as follows
Step I: this claim is directed to a product/apparatus category.
Step II:
Per prong One:
This claim recites identify a task, search for a modules and content, extract text from a drawn model, select a set of modules (based on characteristics and text), generate a AI model and model definition for the AI model, where the act of “by utilizing the neural network” associated with these steps is expressed in a very high level of generality. Per MPEP 2106.04(a), the above activities can be construed as being performed via use of pen/paper (generate a model or a definition) based on conceptually performed acts of identifying, searching, extracting or selection from text or characteristics of drawn design or model, whereas “using a neural network “limitation amounts to a generic well-known tool by which to apply the steps of the mental process or pen/paper derivation. The claim is thus directed to a Mental Processes Abstract Idea subset of a Judicial Exception.
Per prong Two:
The elements recited as processor, repositories, modules, task, drawn AI model amounts to context or field of use in which the Abstract Idea operates, in which computer-based settings or tools allow the conceptually performed acts of identifying, searching, extracting or selection to be carried out. The acts of “generating” the AI model and “executing” the identified task are rather viewed as insignificant post-activities to the Abstract Idea that comprise identifying, searching, extracting or selection; that is, merely applying a abstract Idea in a particular SW, computer field of use without explicit showing of non-conventional limitations to particularly distinguish the computer application/field into a technically inventive transformation fails to integrate the Abstract idea into a practical application. See MPEP 2106.05 (a)(b)(c)(g)(h)
Step IIB:
The additional elements such as modules, code, processor, repositories, drawn model, text content, and generic mention of a “utilizing neural network” for “executing a task”- MPEP 2106.05 (d) - can be seen as a mere effect of appending generic HW or numerical tools or field of use to the Abstract Idea, which, as a way of applying of a Abstract Idea using conventional SW engineering concepts, computer and mathematical-based means, the claim in all cannot provide an inventive concept to the method, as no part of the additional elements (from their high level of generality) is depicting a non-conventional limitation being imparted into the Abstract idea.
Per MPEP § 2106.05(e), the sequence of the additional elements as recited, simply follows the standard, logical workflow of designing and running software. No non-conventional or unexpected synergy is recited that takes it out of the realm of generic automation. In all the additional elements cannot render the Abstract Idea significantly more than itself.
Step IIB analysis of dependent claims.
Claims 2-3 recite update to parameter and operation of a module in response to an edit to the AI model; but this update can be construed as post-activity resulting from a generating step of the Abstract idea; thus amounts a insignificance step susceptible to convert the Abstract idea significantly more than itself.
Claims 4-5 recite visually render graph of AI model including pre-defined modules, custom modules for inclusion in the visual representation, but display result obtained from activities of a mental process can be view a insignificant post-activity that fails to transform the Abstract idea significantly more than itself.
Claims 6 and 8 recite activities of identifying a module for replacement and modifying a AI model based on change in a task; but since these activities are subsequent to the AI model generating, these are mere post-activities that cannot convert the Abstract Idea into a practical application nor can they add significantly more to the mental process of step IIA.
Claims 7 and 9 recite conducting a search from repositories to identify a replacement and act of receiving a drawn AI model from manually drawn modules. These activities are construed as part of the mental process or pre-activities to this mental process, hence cannot add significantly more to the mental process of step IIA
Claims 10 and 12 recite extracting text from a drawn model and importing modules from a search environment into a collection module; these steps amount at best to pre-activities to the abstract Idea of step IIA, and as well-understood techniques, these elements (extract, import) cannot represent non-conventional improvement to this computer field using design and repositories.
Claims 11 and 14 recite generating a different model definition corresponding to the drawn model and modules thereon, and generating model definition via obtaining candidate modules from a repository. The acts of generating definition is construed as a human process that are either conceptual or carried out with pen/paper, whereas the act of generating model definition based on search of candidate modules does not establish a inventive technicality to mere human processes that are either conceptual or carried out with pen/paper as set forth with step IIA.
Claims 15 and 16 recite selection of a property via a interface and displaying of generated code via the interface. Selection can be viewed as a integral aspect of a mental process and displaying can be viewed as insignificant post-activity that fails to render the Abstract Idea significantly more than itself.
Claims 17, 18 and 19 recite enabling selection of a module for replacement, provision of an option to adjust an intensity parameter associated with the AI model, and a digital canvas to enabling drawing. The selection can be construed as a mental process and the options provisioned to adjust and provision of a canvas to draw amount to one or more tools being appended to the Abstract Idea, yet such provision fails to transform this Abstract idea significantly more than itself.
In all, claims 1,13, 20 are deemed non-eligible under the 35 USC § 101 statute.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 8-11, 13, 15, 18 is/are rejected under § 35 U.S.C. 103 as being unpatentable over Javaheripi et al, USPubN: 2020/0125960 (herein Javaheripi) in view of Huang et al, USPubN:
2023/0405820 (herein Huang), and Amer et al, USPubN: 2019/0304156 (herein Amer).
As per claim 1, Javaheripi discloses a system, comprising: a memory; and a processor, wherein the processor is configured to:
facilitate, by utilizing a neural network (neural network architecture - para 0005; framework
104 may be desired neural network architecture - para 0035), selection of a plurality of modules (one or more neurons e.g. basic unit of computation in a neural network - para 0031) for inclusion (graph representation - para 0005; transform neural network architecture 710 - Fig .7) in an artificial intelligence model;
establish, by utilizing the neural network (see framework from above; using a neural network as a baseline model - para 0043), a connection between each module selected from the plurality of modules with at least one (see Figs 2-3; para 0005-0009; connection may be selected – para 0007; Fig. 4; modification module … probability selection module … create new connections, remove prior connections – para 0034; connection … selected using the predetermined probability – para 0060; claim 5, pg. 7 ) other module (see selection from above; where each node is basic unit of computation - para 0031; each node belonging to a layer edges to all nodes no nodes with more than one layer may be connected - para 0046) selected from the plurality of modules,
wherein at least one module of the plurality of modules is identified from text extracted from (graph extraction module … receive a description of the neural network ... data files and/or collection of data files – para 0033 – Note1: text extracted from a NN file description for generating a graph based on the description reads on text extracted from drawn artificial intelligence model expressed in text) a drawn artificial intelligence model;
generate, by utilizing a neural network (see framework from above; using a neural network as a baseline model - para 0043) and based on the connection (see Figs 2-3; para 0005-0009), a model definition (e.g. network may be defined as small world - para 0040; baseline neural network model may shift towards the small-world equivalent or counterpart- para 0046; rewired graph, newly generated graph - para 0049) of the drawn artificial intelligence model (see Note1) and including the at least one module identified from the text (refer to text extraction from above),
A) Javaheripi does not explicitly disclose processor configured to generate a model definition of a drawn artificial intelligence model in terms of
(i) generating code for each module selected from the plurality of modules; and
(ii) execute a task by utilizing the artificial intelligence model via the model definition generated via the code for each module selected from the plurality of modules.
Javaheripi discloses reconstructed network characterized with clustering of edges and nodes maintained with a low characteristic path length for build of a small-world NW according to which the iterative algorithm may be executed according to a modification having vertices removed by a probability p and reconnected to a new node selected according a desired redistribution, such that the graph execution terminates when all edges in the original graph have been considered once (para 0038). Hence, algorithm underlying a modified graph so that modified vertices and connectivity thereof are part of the executed algorithm in which all edges in the original graph have been considered once entails code associated with vertices/nodes and edge relationships of a given reconstructed NN graph being executed in accordance to topological reconfiguration that adopts a redistribution maintained with cluster of low path length intended for a given small-world network software (see Fig. 2, 3B, 4).
Amer discloses generating of composition graph from extracting relationships from a video/text language parser to add/remove nodes based on probability distribution underlying a probabilistic model and user prompting of an artificial intelligence transform scene (para 0127), using a graph module to parse training data or textual description (para 0187-0189) into a spatio-temporal event graph (para 0006-0007) depicting event frames, each identifying the "who", "to whom", "what, "where", "when", "did what" mapped among each other- e.g. for configuring a scene or animation event, using a generative Network model to train, using a training (Fig. 9) or animation module in association with I/O communications identified from the composition graph (para 0182-0185), and executing the abstracted animation of motion, operations, by actors of the actions in form of a machine learning model (para 0190-0191), where analytics and operations over the composition graph such as input processing, actions by graph module, agents module, classification task (para 0152) and animation module are implemented via computing devices (para 0140-0142) or classification task as executables made into one or more modules, functional units, pre-installed app such as computer program and code (para 0202).
Hence, organization of parsed textual input into an animation/composition graph, comprising events each identifying action, actor and context of the action for use by a training model that use executable of modules, functional units, pre-installed app to execute classification task, rendering behaviors or operation of the NN framework
Huang discloses a framework for predicting outcome of objects, particularly as a grasp
Neural Network (Fig. 1-2, Fig. 6) in connection with an OpenVINO ("Open Visual Inference and Neural network Optimization") toolkit enabling implementation of logic to perform operations facilitating Neural network applications that include tasks on vision emulation, speech/action recognition, language processing, object detection, classification etc, assisted with Open software and Libraries support system (para 0132-0133), the OpenVINO having a model optimizer to reduce layers of the model, resize the model input, modify batch size, and number of model layers, or re-quantifying the model representation (para 0134), and an inferencing engine equipped with library classes to implement one or more API functions to process input/output formats or intermediate representation of the model or execute the model (para 0135) in accordance with various functionalities (para 0137) associated with neural network model operations like in TensorFlow, PyTorch frameworks as variants effectuated with OpenVINO, using grouping of resources (para 0140) as processing units or node computing resources for processing data, I/O associated with inferencing applications (para 0577); e.g. GPUs of a AI system for executing visualization tasks (para 0578), compute services, preprocessing data and post-processing outputs of machine learning models, to perform inferencing or AI-based processing tasks (para 0579-0581)
Hence, selecting a group of compute units or processing resources to perform a AI-based visualization tasks, data recognition, I/O processing, graphics services, preprocessing data and
post-processing outputs from predictive engine as part of implementing executables/functionalities for compute resources or cluster nodes of a AI system or predictive model under a neural network framework is recognized.
Therefore, based on formation of a graph created via an API to provide definition of NN
action and identify properties, relationship among the compute units thereon as part of describing
a neural network and constant finetuning (para 0026) of the model in Javaheripi system for fast
training and execution of neural networks (para 0005, 0014, 0024; fast neural network training and
execution - para 0032) using in part selective probability distribution aiming faster convergence speed (para 0044), it would have been obvious for one of ordinary skill in the art before the
effective filing date of the invention to implement the neural network models in Javaheripi so that
utilizing a neural network framework and based on the selection (of the property for each module
and the connection), a model definition for the NN is generated and deployed by generating code for each module selected from the plurality of modules - as shown in Huang and in Amer executables for event frames - enabling thereby to execute a task - as in Huang AI-based processing tasks -- associated with the artificial intelligence model/engine expressed via the generated model definition, with the execution made possible via the code generated for each module selected from the plurality of modules - as in Amer - to carry out a particular NN-based application; because
frameworks for developing and deploying a inferencing or Artificial Intelligence model using Neural network architecture mostly rely or operate on basis of taking input data or textual description of the target model and converting significant construct or extracted concepts into a structured, visual hierarchy, numerical formulation, or graph representation - see Javaheripi: para 0033 - formed of basic units or nodes linked by edge relationships, per developer's UI activities using of this visual representation in conjunction with analysis of properties and meta-attributes of the graph basic units, node elements under tools and UI provisioning by the Neural Network framework would enable further refining of the functional configuration of the model, according to selective rearrangement by the user so that improved or redefined instance of the NN model over its initial version in terms of optimized inter-node relationships or path length, call function consolidation would enable a more compact code to be generated in the deployment of the NN model, the realization thereof by the NN framework being able to execute tasks, operations underlying inter-related elements visualized in the optimized model, providing thereby a most efficient payload for actions to be executed in relation to fulfilling functionalities of application or rendering of service flow by the NN model for which the graph-based framework is intended to achieve.
As per claim 8, Javaheripi discloses system of claim 1, wherein the processor is further
configured to automatically modify the artificial intelligence model based on a change in the task
(e.g. speech recognition model that needs updating (fine-tuning) to comply with the new samples
from the previously seen dataset - para 0026).
As per claims 9-11, Javaheripi does not explicitly disclose system of claim 1, wherein the processor is further configured to
receive a drawn artificial intelligence model comprising manually drawn modules;
extract text from each block in the drawn artificial intelligence model, and
generate a different model definition corresponding to the drawn artificial intelligence model and including the at least one module from the plurality of modules correlating
with the text.
Javaheripi discloses extracting equivalents in a high-level description to corresponding neural network architecture to configure edges and nodes topology of the original NN model (para 0045; description of a neural network architecture - claim 10, pg. 7; para 0033) the description being provided by the user (para 0032; high-level description of the user's desired neural network – para 0035) in form of description file (refer to Note1)
Amer discloses parsing predefined data structure from textual information to match a story, a event, or constituents of a story in terms of parsed information depicting "who", "did what", "to whom or what", "when", "where" context or predicates (para 0083) using NLP techniques or language parser (para 0037, 0043) to extract the textual, word format depicting training data set in form of groups, epochs (para 0069) or predicates ("who", "did what", "to whom or what", "when", "where" - para 0062-0063; para 0092, 0104) by a tracking module into composition of a graph (para 0070); hence natural text input structured in word block format indicative of textual predicates entails manually generated text or drawn scene information format into a NL parser module to extract word, or equivalents to respective portions of a graph representation of a model is recognized.
Thus, it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to implement processing capabilities of a graph module in Javaheripi so that a processor implementation therefor would be configured to
receive a drawn artificial intelligence model or description file (per Note1) comprising drawn modules;
extract text - as per Amer - from each block or keyword in the drawn description of artificial intelligence model, and
generate a different model definition - as per Amer formation of a composition graph - corresponding to the drawn artificial intelligence model and including the at least one module from the plurality of modules correlating with the text - per a NLP in Amer; because
use of a particular module for extracting group of text or keyword of significance, or scene
describing block from initial drawn text or descriptive stream as part of input information collected
into a graph composition module of Javaheripi's NN framework would enable a particular text, depiction in portable format on information block to be assessed, verified and accordingly deemed equivalent to a node or module of a graph and accordingly selected for representing a node module of the graph, such as to consolidate the modular formation of SW structure hierarchy or logical arrangement expected to realize the NN functionality by the framework.
As per claim 13, Javaheripi discloses a method, comprising:
receiving, by utilizing a neural network (refer to claim 1), a drawn artificial intelligence model (see Note1);
extracting, by utilizing the neural network, text (refer to claim 1) from each block (directed graph based on the description … one or more nodes …connected via one or more edges – para 0033) in the drawn artificial intelligence model;
generating, by utilizing the neural network (refer to claim 1) and based on the graph of the artificial intelligence model (see above), a model definition (refer to claim 1) for the drawn artificial intelligence model and including at least one module from a plurality of modules (Figs 2-3; para 0005-0009; each node belonging to a layer edges to all nodes no nodes with more than one layer may be connected - para 0046) correlating with the text (refer to extracted text per para 0033) by generating code (refer to rationale A of claim 1) for an artificial intelligence model; and
executing, by utilizing the neural network, the model definition for the artificial
intelligence model to perform a task (refer to rationale A of claim 1).
As per claim 15, Javaheripi discloses method of claim 13, further comprising enabling selection of at least one property (measure characteristic in a plurality of probabilities may be selected - para 0031; probability selection module - para 0034) of the artificial intelligence model via an interface (plot indicating coefficient (C), small world property, path length L,probability p – para 0018; Fig. 3c) of an application associated with the neural network.
As per claim 18, Javaheripi discloses method of claim 13, further comprising providing an
option to adjust an intensity level for reducing operations (distributed training reduce an overall
training time - para 0028) or parameters (path length L, coefficient L may be reduced- para 0044;
characteristic path length may be reduced - para 0046) associated with the artificial intelligence
model.
Claims 2-3 is/are rejected under§ 35 U.S.C. 103 as being unpatentable over Javaheripi et al,
USPubN: 2020/0125960 (herein Javaheripi) in view of Huang et al, USPubN: 2023/0405820 (herein Huang), and Amer et al, USPubN: 2019/0304156 (herein Amer) further in view of Domquast et al, USPubN: 2016/0094649 (herein Domquast) and Hoydis et al, USPubN: 2023/0403100 (herein Hoydis)
As per claims 2-3, Javaheripi does not explicitly disclose system of claim 1, wherein the processor is further configured to
update a parameter for at least one module of the plurality of modules of the artificial intelligence module after adding an additional module to or removing a module from the artificial
intelligence model;
update an operation for at least one module of the plurality of modules of the artificial intelligence model after adding an additional module to or removing a module from the artificial
intelligence model.
Javaheripi discloses graph rewriting in terms of rearrangement ("rewire") made to a neural etwork model graph representation by removing, substituting, adding new edge or connections between basic units of the representative graph using a constant upgrade to characteristics, parameters and probability relationships to the involved nodes portion of the graph resulting from instance of the rewriting or finetuning the model (para 0031, 0034).
Domquast discloses configuration of node operations for performing data distribution associated with predictive storage techniques, via collection of graphical UI with diagrams illustrating the synchronization of operations or distribution mechanism by the nodes (Fig. 1), the mechanism including events or actions on read, write, delete, create on data (para 0045-0046), where an authority responsible for managing the parameters of the distribution between the nodes communicates (to participant users) collection of parameters resulting from actions of read, write, delete, create as part of redefining the file system associated with the storage(para 0054-0056), enabling a GUI user to follow the distribution from a node to a root to identify change and update the node or the root node according to parameters or metadata for each(para 0066-0067); e.g. to reflect a location information associated with a content being moved (para 0068), or updating vector relationships between node A and B due to a inter-node modification event (para 000075-0079), where collection of nodes as part of a predictive model that may be implemented as neural networks or other machine learning algorithms (para 0197-0198) for deriving a decision score, where refining the model can adjust NN weightings associated with I/O of the nodes in the distribution paradigm (para 0219)
Hence, configuration of a distribution scheme via a predictive NN model where tracking
changes between nodes performs responsive updates reflecting the changes in form of parameters or metadata update to each relationships event or topological modification due to action such as add, remove an element of the distribution scheme or nodes of machine learning/NN model implementing a predictive storage technique.
Implementing decoding schemes applied to a graph Neural network (GNN) application is
shown in Hoydis for rescaling block lengths in regard to improving dimensionality, performance of the algorithm (see Abstract), via processing in terms of conceptual diagrams of a rasterization or
graphics pipeline (Fig. 5), the GNN training to learn on message passing algorithm which may
represent node replacement or updates to trainable functions (para 0028), with the trainable parameters include factor node update parameters and variable node update parameters as part of
permutations (para 0040) observed from message or codewords associated with the code blocks of
the algorithm, the latter performing iterative process over symbol values and parameters of the graph nodes (claim 25, pg. 19) and updating one or more variable to factor node as well as updating factor node to variable node messages and/or updating variable nodes based on one or more trainable parameters (claim 26 pg. 19)
Hence updating parameters and variables associated with codewords pertinent to graph nodes of a NN algorithm on basis of permutations or rescaling block lengths, or node update as part of improving decoder software implementing a NN pipeline entails update state of a decoder or SW instruction of the algorithm.
Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to implement tracking by the NN environment of parameters and metadata of graph nodes associated with visual representation a NN model to a development interface in Javaheripi, so that responsive to an event or modification affecting a node, the NN framework processor is further configured to update a parameter - as shown in Domquast - update state of the decoder instruction - Hoydis - for at least one graph node or module of the plurality of modules of the artificial intelligence or NN node/module after a modification event identified as replacing a node - as in Hoydis - adding an additional module to or removing a module from the artificial intelligence model - as shown in Dorquast update; because
parametric assignment or variable settings of modules or neurons of a neural network graph is integral to the programming code or executables pertinent of operation of each module or neuron of this network algorithm, and changes to nodes of the NN representation in terms of their being
removed, added or substituted from the perspective of a tracker effect of an user development
framework require automated adjustment or update lest mismatch between parametric setting for
graph element prior to the change and those subsequent to the change would not only create
programmatic or compiler conflict; but also would not reflect runtime behavior of the NN graph
modules as intended by user action or an optimization scheme associated with the framework tool.
Claims 4-7, 12, 14, 17, 20 is/are rejected under§ 35 U.S.C. 103 as being unpatentable over
Javaheripi et al, USPubN: 2020/0125960 (herein Javaheripi) in view of Huang et al, USPubN:
2023/0405820 (herein Huang), and Amer et al, USPubN: 2019/0304156 (herein Amer), further in
view of Paini et al, USPubN: 2024/0143354 (herein Paini) and Browne et al, USPubN:
2018/0293517 (herein Browne)
As per claims 4-5, Javaheripi does not explicitly disclose system of claim 1,
(i) wherein the processor is further configured to visually render a graph for the artificial intelligence model including a visual representation of each module of the plurality of modules selected for inclusion in the artificial intelligence model.
(ii) wherein the plurality of modules are pre-defined modules, custom-generated modules, or a combination thereof.
Browne discloses AI system (Fig. 2A) with use of project file to import entity code or
already-compiled software into a AI model as these entities can be plugged as concept nodes (para
0096-0097) of a graph representation of the AI model/engine (para 0071-0072; para 0098), the
external entities of code packaged as reuse into different design contexts and retrievable from remote via import statement (para 0089-0093) and a URI specified by a python script, and configured as blocks of code interacting with each other via relationships and interfaces specified by scripts and ReSTful implementations (para 0099-0105) to form node stream graph into the AI engine (para 0106-0115), enabling the user to further configure and redefine the concept nodes in course of the training or simulation in accordance to code interaction and dependencies description (para 0150- 0151)
Paini also discloses analytic subsystem of a AI system to extract metadata from a dataset and relationships to match graph templates destined for quick learning and faster build of the graphical analytics (see Abstract), using a framework providing user customization options and transformer engine built upon auto-build templates (para 0075; Fig. 2; nodes in a graph - para 0030-0031), in that the pre-defined code template can be modified based on parameter change requirements and dynamic code synching with the AI engine(para 0073) using a self-learning stage of the AI system to receive templates (template skeleton) from pre-trained information repository, or predefined industry logic along with associated information/relationships which can be adapted for compiling code of the cognitive smart AI engine (para 0082; Fig. 7) where customization of the graph uses video guidance (para 0022) to visually explain formation and/or how nodes are to be logically grouped (para 0080) as part of the smart cognitive aspect of the Al transformer
Hence, pre-built code entities being imported from external repository and cognitive AI subsystem having visual assistance enabling interactive placement and selection of pre-build entity codes reusable as graph nodes as part of user customization and graph-analytics by a AI engine is
recognized.
Therefore, it would have been obvious for one of ordinary skill in the art before the effective
filing date of the invention to implement selection of node implementation in building a NN graph
in Javaheripi so that the AI platform processor is programmed to visually render a graph for the
artificial intelligence model having therein a visual representation of each module of the plurality of
modules selected for inclusion in the artificial intelligence model - as per the visual guidance in Paini - wherein presentation of the NN graph comprises plurality of modules are pre-defined modules, custom-generated modules, or a combination thereof - as set forth in Paini repository of customizable code entities; or from external store of pre-compiled entity code as in Browne; because
pre-built, custom code entities or pre-compiled modules stored remotely for use by diverse
host machines in terms of imported/reuse SW assets can be continually managed with relevant control or trusted registration protocol by those services particularly dedicated to administer and
update versioned instances of these assets, which would facilitate retrieval by target environment by a mere posted request or URI-based query while alleviate a target host environment for having to store reusable entities locally, and provision of externally maintained prebuilt modules as a non-contextual SW template or agnostic implementation resource as set forth above, would enable a local user to be able to perceive/identify - as set forth in Paini visual guidance- and select which particular code entities modules to edit or customize once such external resource is integrated as nodes of a graph representation of a model, in order to adapting one such initial/raw code entity into one or more purposeful and context-specific executable units forming the overall functionality of the model as endeavored by a host framework or a AI engine.
As per claims 6-7, Javaheripi does not explicitly disclose system of claim 1, wherein the processor is further configured to
identify at least one module of the plurality of modules of the artificial intelligence model for replacement;
conduct a neural architecture search in a plurality of repositories to identify at least one replacement module to replace the at least one module for replacement.
Browne discloses AI model configured to assess individual concept nodes imported from
remote and possibly replace the node without having to retrain the whole AI model (replaced with a new version, trained individually - para 0095)
Paini discloses storage of reuse entity code in repository as auto-built template or skeleton
templates (para 0082)
Amer discloses search and processing thereof (para 0165-0172) and selective ingestion of
exemplary scene component into graph representation of AI model using a generative network of a
ML system (para 0035) thereby forming a spatio-temporal relationships of model that performs
training over generated animations from a data store (para 0009)
Thus, based on Browne, Paini and Amer, conducting neural network search at a AI framework directed at one or more repositories to identify module or code entity for use in additively generating an AI graph or replacing a inter-node link or a node of the graph is recognized.
Therefore, based on possibility to reduce or modify node links or replacing node pairs (para
0041,0044-0045) as shown in Javaheripi graph "rewriting", it would have been obvious for one of
ordinary skill in the art before the effective filing date of the invention to implement the graph optimization in Javaheripi with processor programming so that, the optimization or rewriting process would be configured to identify at least one module of the plurality of modules of the artificial intelligence model for replacement - as in Browne or Javaheripi - then conduct a neural architecture search - as set forth in Amer - in a plurality of repositories- as per Paini and Amer - to identify at least one replacement module to replace - as in Browne or Javaheripi - the at least one module with the replacement; because
each module node being processed by a optimization algorithm entails a particular
functionality and implicated relationship with the other module of the NN graph so that upon a given consideration, by a optimization process, for inducing reduction of NN code payload by removing a edge, a node or substituting one node function with another function, parametric reconfiguration of the portion of neural network graph affected by the intended removal or replacement can sometimes be more effort intensive than conducting additional queries or rediscovery directed at one or more external storages in order to finding a prebuilt or existing code entity among plurality thereof that would most readily and immediately fit - as replacement module - for the intended programmatic portion of a graph in which one or more nodes has been contemplated for removal or replacement.
As per claim 12, Javaheripi discloses system of claim 1, wherein the processor is further
configured to import the plurality of modules (refer to search conducting for external storages per
rationale of claim 7) from a search space including a module collection (refer to obviousness of
claims 4-5 for prestored module or code entities from the teachings by Browne and Paini).
As per claim 14, Javaheripi does not explicitly disclose method of claim 13, further comprising generating the model definition for the artificial intelligence model by obtaining, via a neural architecture search, candidate modules for the artificial intelligence from a repository.
But this feature of performing a search from the NN architecture framework a search directed at a remote repository to seek and obtain candidate code entities or modules for use in generating a AI model would fall under the ambit of conducting a search to identify pre-defined, custom modules as addressed in claims 5-7.
Hence, obtaining, via a neural architecture search, candidate modules for the artificial
intelligence module from a repository as part of generating the model definition for the artificial
intelligence model would be deemed obvious for the same reasons set forth in claims 5-7.
As per claim 17, Javaheripi does not explicitly disclose method of claim 13, further
comprising enabling selection of the at least one module of the artificial intelligence model for
replacement by at least one other module.
But the identification of a candidate module destined to replace a module of a AI model falls under the ambit of the subject matter of claims 6-7; hence would have been obvious for the same reasons set forth in the rationale of claims 6-7.
As per claim 20, Javaheripi discloses a device, comprising: a memory; and a processor;
wherein the processor is configured to identify, by utilizing a neural network (refer to claim 1), a task (just-in-time service, task - para 0026-0027; different functionalities, operations CONY, batch normalization, activation, - para 0047; linear operations by CONY - para 0050) to be completed by an artificial intelligence model (refer to claim 1);
wherein the processor is configured to search, by utilizing the neural network, for a plurality of modules (refer to rationale of claims 6-7) and content in a plurality of repositories (refer to rationale of claims 7, 12);
wherein the processor is configured to extract, by utilizing the neural network, text (refer to claim 1, see Note1) from a drawn artificial intelligence model of the content that is associated with the artificial intelligence model;
wherein the processor is configured to select, by utilizing the neural network, a set of candidate modules of the plurality of modules in the plurality of repositories (refer to rationale of claim 7) based on matching characteristics of the set of candidate modules (see search and identify from repositories in claim 7) with the task (see above) and based on the text (see Note1);
wherein the processor is configured to generate the artificial intelligence model (refer to claim 1) based on the portion of the content and the set of candidate modules (refer to the above), wherein a model definition is generated for the artificial intelligence model (refer to claim 1) based on the drawn artificial intelligence model (see above); and wherein the processor is configured to execute the task (refer to rationale A(ii) of claim 1) using the artificial intelligence model.
B) Javaheripi does not explicitly disclose
wherein the processor is configured to extract, by utilizing the neural network, text from each block in a drawn artificial intelligence model of the content that is associated with the task, the artificial intelligence model, or a combination thereof;
Javaheripi discloses a NN architecture being layered into neuron computation units performing within the network in accordance to classes or nature of their operations (para 0047), each neuron or node being a basic unit of computation of the network (para 0031) each computation representing a task of a graph cluster to be executed with the algorithm (para 0038) developed for a corresponding node/edge topology being modified to adapt a accelerated realization or distribution scheme; e.g. cluster of low path length intended for a given small-world network software (see Fig. 2, 3B, 4) – being deemed faster compared to realization of the original graph, the original graph derived from extracting text for file description of edge connected blocks of the original text model (files – para 0033)
Hence configuring edge-node cluster of a artificial model in which each block from the initial drawn artificial intelligence model is related to content of description text that is associated with a compute task among other tasks of the generated artificial intelligence model or model definition thereof from which executable algorithm is being generated is recognized – referred herein as (*).
Mapping of nodes of a model graph to corresponding functions of the executable version of the graph-represented model which is being derived from description text, schema or drawing was a well-understood practice.
Amer discloses generating of composition graph from extracting relationships from a video/text language parser to add/remove nodes based on probability distribution underlying a probabilistic model and user prompting of an artificial intelligence transform scene (para 0127), using a graph module to parse training data or textual description (para 0187-0189) into a spatio-temporal event graph (para 0006-0007) depicting event frames, each identifying the "who", "what", "to whom", "where", "when", "did what" mapped among each other. Hence, extracting functional elements from a raw input stream to match a given operation on an artificial intelligence animation model is recognized.
Huang discloses inferencing engine equipped with library classes to implement one or more
AP functions to process input/output or intermediate representation of the model (para 0135) in
accordance with various functionalities (para 0137) associated with plurality of a neural network model operations like in TensorFlow, PyTorch frameworks as variants effectuated with OpenVINO, using grouping of resources (para 0140) such as processing units or node computing resources for performing I/O associated with inferencing applications (para 0577); e.g. GPUs of a AI system for executing visualization tasks (para 0578), compute services, preprocessing data and post-processing outputs of machine learning models, to perform inferencing or AI-based processing tasks (para 0579-0581).
Hence, processing representation of a model to correlate library of classes or API functions
with particular processing units implementing a AI task entails mapping a portion of the AI model
with an API or library class to perform a function.
Paini discloses extract metadata from a dataset and relationships to match graph templates destined for quick learning and faster build of the graphical analytics (see Abstract), in providing user customization options and transformer engine built upon pre-defined templates (para 0075; Fig. 2; nodes in a graph-para 0030-0031), in that the raw template can be modified based on parameter change requirements and dynamic code synching with the AI engine (para 0073) using a self-learning stage of the AI system to receive templates (template skeleton); hence extracting relationships of input-dataset to acquire raw templates that are parameterized into a code for execution by an AI engine entails extracting, by utilizing the neural network, a functional portion (e.g. a predefined template) of the original input content that can correlate with the task or computation unit, of the artificial intelligence model.
Therefore, based on use of text extracting so that each node of graph representation described per a text content can correspond to a compute task to implement among tasks of a particular adaptation of AI model scheme as set forth above in Javaheripi executing algorithm set for a accelerated distribution scheme - per (*) from above, it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to implement the user-description in Javaheripi so that use of a extraction logic can achieve correlation between text or portions of the initial description - as in Amer - and a task, a function or operation by the NN model in order to associate a pre-built code entity as in Paini and Huang's Open system for accordingly generating a node hierarchy and relationships on basis of the extracted information; because
provision of externally maintained prebuilt modules as a non-contextual SW template or
agnostic implementation resource as set forth above coupled with an extraction module that extract
correspondence between a function or task described from a initial NN description as set forth
above and a prestored code entity as in Paini, or Huang, would enable a local to identify and
select which particular code entities modules to fetch or acquire from external resource and
integrate such code entities onto a graph representation of a model where each node is a basic
computation formed from the code entity, whose finetuning manipulation and parameterization by a framework user would be able to adapt one such initial/raw code entity into one or more purposeful and context-specific executable units or cluster forming the overall functionality of the model as endeavored by a host framework or an AI engine in Javaheripi.
Claims 16 and 19 is/are rejected under§ 35 U.S.C. 103 as being unpatentable over Javaheripi et al, USPubN: 2020/0125960 (herein Javaheripi) in view of Huang et al, USPubN: 2023/0405820 (herein Huang), and Amer et al, USPubN: 2019/0304156 (herein Amer) further in view of Brent et al, USPubN: 2023/0068660 (herein Brent) and Sun et al, CN 110377282 (translation) 8-17-2021, 10 pgs (herein Sun)
As per claims 16 and 19, Javaheripi does not explicitly disclose method of claim 13, further comprising
displaying the code generated for the artificial intelligence model via a user interface.
providing a digital canvas to enable drawing of blocks, connections, modules, or a combination thereof, associated with the artificial intelligence model.
Brent discloses a grid canvas (para 0379; canvas 502 - Fig. 5) as interface to support drawing
of diagrams and illustrating a flow process of a development which allow additional modification
(para 0124-0125) by the user via process menu (para 0381) enabling the user to draw or configure
block diagrams (Fig. 5) or a process outline steps (Fig. 6), where recognition of objects and layered
information by the GUI based environment also support neural network representation in terms of correlating images and environment objects as part of the layered image (Fig. 7) - e.g. features in data set modeled as dimension of a CNN model (para 0141) whose 3 dimensions can be visualized as region or sub-regions to be processed in depth by the convolutional NN (Fig. 10-11; para 0143-0144); where the canvas provides interactive detection of UI objects (movement of objects) defined with programming language in a visual field- listening to user touch, or triggered positioning (para 0079-0081) by which the user can explicitly or implicitly declare object class (para 0075-0076) underlying a field activable by a touch.
Hence, canvas for describing a NN model layer toward classification of information or
datasets and enabling, via a visual field presentation of programming language objects, a developer to activate upon the object thereby to declare of objects or classes of associated programming language is recognized
Sun discloses a web-based Neural Network framework, using a convolutional NN UI that
constructs a mapping relation between one or more HTML elements of the Web code per a display
effect and the corresponding element's (generated) source code (step 1, step 4 - pg. 3; step 1, pg. 6;
claim 2, pg. 9) enabling identification of the position of the HTML element on the display, and via
the UI, evaluation of similarity between the HTML code and the generated written code (step 1, step 3-4, pg. 4)
Thus, it would have been obvious at the time of the invention for one skill in the art to
implement the development UI in Javaheripi NN framework so that the UI would support drawing
and graphical editing of the NN model via provision of a digital canvas to enable drawing of blocks, connections, modules, or a combination thereof, associated with the artificial intelligence model as in Brent - as well as having visual field affording the developer to be presented with programming code objects or corresponding source code as in Sun, by which, via interactive effect, the developer would be able to declare, define or manipulate code constructs - as per the 'declare object' setting in Brent or per effect of verifying a via UI correspondence between web equivalent and CNN generated code as in Sun; because
depiction of a model using a drawing canvas equipped with digital grid would facilitate
generation of diagrams and block relationships required for a flow by the model such as a Neural
Network architecture in Javaheripi framework and provision of GUI elements enabling the user to
visualize object representing definition of object or classes of a programming language would
facilitate the endeavor of a developer toward investigating or selecting candidate programming code or constructs and consolidating a set of functions or basis code deemed suitable for further parameterization and adaptation into the execution stage that carries out rendering of a task or
sequence thereof as part of simulating and/or deploying a NN model or a web implementation of a CNN as in Sun.
Response to Arguments
Applicant's arguments filed 4/16/26 have been fully considered but they are not persuasive. Following are the Examiner’s observations in regard thereto.
(A) Applicants have submitted that with addition (in claims 1, 13, 20) of model definition of a drawn artificial intelligence model including a module identified from a corresponding text extracted from the drawn model, the NN architecture selection by Javaheripi, the Open Vino optimization by Huang and the scene parsing formation of graph in Amer are deemed deficient to support the position of the Office Action, notably when Amer scene parsing is not concerned with text extraction, when Huang is merely concerned with optimization (not about definition from a drawn model) (Applicants Remarks pg. 3-5). The newly added limitations have been addressed with a reconstructed ground of rejection in which Amer and Huang are not particularly geared to render obvious deriving text from a drawn model, but rather for another limitation in rationale A of claim 1. The allegation on merits for the added feature as raised by the Applicant therefore become MOOT or non-prima facie compliant.
(B) Applicants have submitted that claim 13 as amended is now narrower due to the extracting of text from a drawn artificial intelligence model, in that Javaheripi, Amer and Huang as mentioned above cannot be seen as meeting this more narrow feature (Applicants Remarks pg 5-6), and that claim 20 is also amended to be narrower in the same extent of claim 13, which is rendering the combination of references by the Office Action ineffective and that the obviousness rejection based on these references would have to be withdraw (Applicants Remarks pg.6 to pg. 7). The newly amended claim language in question has been addressed with a more recent and different ground of rejection that does not include extracting text of a drawn artificial intelligence model as a core for the obviousness analysis.
(C ) Applicants have submitted that the Office has treated the manually drawn model, text extraction and text-correlated module concepts in separate feature not as combination of features (Applicants Remarks pg. 7). The above remark fails to point out exactly where a claim language (i.e. the newly added one) is not properly prosecuted using the rules set by virtue of 102, 103 statute, and so, in direct reference to the actual grounds of rejection being presented to meet this added language.
For the above reasons, the claims as submitted stand rejected as set forth above.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Tuan A Vu whose telephone number is (571) 272-3735. The examiner can normally be reached on 8AM-4:30PM/Mon-Fri.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Chat Do can be reached on (571)272-3721.
The fax phone number for the organization where this application or proceeding is assigned is (571) 273-3735 ( for non-official correspondence - please consult Examiner before using) or 571-273-8300 ( for official correspondence) or redirected to customer service at 571-272-3609.
Any inquiry of a general nature or relating to the status of this application should be directed to the TC 2100 Group receptionist: 571-272-2100.
/Tuan A Vu/
Primary Examiner, Art Unit 2193
June 27, 2026
Prosecution Insights