DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Style
In this action unitalicized bold is used for claim language, while italicized bold is used for emphasis.
Information Disclosure Statement
All information disclosure statements were submitted prior to the first action and are incompliance with the provisions of 37 C.F.R. § 1.97. Accordingly, they have been considered.
Applicant Reply
“The claims may be amended by canceling particular claims, by presenting new claims, or by rewriting particular claims as indicated in 37 CFR 1.121(c). The requirements of 37 CFR 1.111(b) must be complied with by pointing out the specific distinctions believed to render the claims patentable over the references in presenting arguments in support of new claims and amendments. . . . The prompt development of a clear issue requires that the replies of the applicant meet the objections to and rejections of the claims. Applicant should also specifically point out the support for any amendments made to the disclosure. See MPEP § 2163.06. . . . An amendment which does not comply with the provisions of 37 CFR 1.121(b), (c), (d), and (h) may be held not fully responsive. See MPEP § 714.” MPEP § 714.02. Generic statements or listing of numerous paragraphs do not “specifically point out the support for” claim amendments. “With respect to newly added or amended claims, applicant should show support in the original disclosure for the new or amended claims. See, e.g., Hyatt v. Dudas, 492 F.3d 1365, 1370, n.4, 83 USPQ2d 1373, 1376, n.4 (Fed. Cir. 2007) (citing MPEP § 2163.04 which provides that a ‘simple statement such as ‘applicant has not pointed out where the new (or amended) claim is supported, nor does there appear to be a written description of the claim limitation ‘___’ in the application as filed’ may be sufficient where the claim is a new or amended claim, the support for the limitation is not apparent, and applicant has not pointed out where the limitation is supported.’)” MPEP § 2163(II)(A).
Drawings
The drawings are objected to under 37 CFR 1.83(a). The drawings must show every feature of the invention specified in the claims. Therefore, a structure including all three claimed layers, and including the claimed relationships between layers, as well as each layer’s relationship to the knowledge graph must be shown or the feature(s) canceled from the claim(s). To be clear, the following should be shown:
a first layer and its relationship to the “fixed-dimensional graph embedding vector,” the “note and relationship type in a fused event knowledge graph,” including the nodes of the fused event knowledge graph.
The second layer and its relationship to the first layer and to the “fixed dimensional event embedding vector” and to the “sequence of event embeddings.”
The third layer and its relationship to the second layer and to the “sequence of event embeddings,” as well as any relationship to the “fixed-dimensional event embedding vector” that is output from the second layer.
While changes to the claims may change which claim elements must be illustrated, changes to the claims do not affect the requirement that drawings include every feature of the invention specified in the claims. Therefore, amendments that include additional claim features should be understood as resulting in a requirement to include corresponding features in the drawings.
The objection to the drawings will not be held in abeyance.
No new matter should be entered.
Corrected drawing sheets in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. The figure or figure number of an amended drawing should not be labeled as “amended.” If a drawing figure is to be canceled, the appropriate figure must be removed from the replacement sheet, and where necessary, the remaining figures must be renumbered and appropriate changes made to the brief description of the several views of the drawings for consistency. Additional replacement sheets may be necessary to show the renumbering of the remaining figures. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either “Replacement Sheet” or “New Sheet” pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. THE OBJECTION TO THE DRAWINGS WILL NOT BE HELD IN ABEYANCE.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 1-15 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for pre-AIA the inventor(s), at the time the application was filed, had possession of the claimed invention.
Generally: separately listed claim elements are construed as distinct components, all claim terms must be given weight, and there is presumed to be a difference in meaning and scope when different words or phrases are used in separate claims. Since different term or phrases are presumed to differ in scope and each term or phrase in the claims must find clear support in the description, a description of a single element in the Specification may fail to support multiple claim terms. “[C]laims must ‘conform to the invention as set forth in the remainder of the specification and the terms and phrases used in the claims must find clear support or antecedent basis in the description so that the meaning of the terms in the claims may be ascertainable by reference to the description.’ 37 C.F.R. § 1.75(d)(1).” Phillips v. AWH Corp., 415 F.3d 1303, 1316 (Fed. Cir. 2005) (as cited in MPEP § 2111). Further, a lack of lack of detail in the Specification describing how a claimed result is achieved can support a finding that the Applicant was not in possession of the claimed invention at the time of filing, notwithstanding verbatim support. “It is not enough that one skilled in the art could write a program to achieve the claimed function because the specification must explain how the inventor intends to achieve the claimed function to satisfy the written description requirement. See, e.g., Vasudevan Software, Inc. v. MicroStrategy, Inc., 782 F.3d 671, 681-683, 114 USPQ2d 1349, 1356, 1357 (Fed. Cir. 2015) (reversing and remanding the district court’s grant of summary judgment of invalidity for lack of adequate written description where there were genuine issues of material fact regarding "whether the specification show[ed] possession by the inventor of how accessing disparate databases is achieved"). If the specification does not provide a disclosure of the computer and algorithm in sufficient detail to demonstrate to one of ordinary skill in the art that the inventor possessed the invention a rejection under 35 U.S.C. 112(a) or pre-AIA 35 U.S.C. 112, first paragraph, for lack of written description must be made.” MPEP § 2161.01(I). “An original claim may lack written description support when (1) the claim defines the invention in functional language specifying a desired result but the disclosure fails to sufficiently identify how the function is performed or the result is achieved[.] See Ariad Pharms., Inc. v. Eli Lilly & Co., 598 F.3d 1336, 1349-50 (Fed. Cir. 2010) (en banc). The written description requirement is not necessarily met when the claim language appears in ipsis verbis in the specification. ‘Even if a claim is supported by the specification, the language of the specification, to the extent possible, must describe the claimed invention so that one skilled in the art can recognize what is claimed. The appearance of mere indistinct words in a specification or a claim, even an original claim, does not necessarily satisfy that requirement.’” MPEP § 2163.03.
The independent claims fail to adequately describe the overall structure of the claimed “three-layered prediction model.” The only drawing of the model is box 160 on figure 1. That is, the only drawing of the claimed model appears to be a rectangle labeled “Prediction Model (for actions 1,…,x),” incorporated into the feedback loop of control scheme. Clearly this is insufficient to show possession of the three-layered model recited in the claims. The Specification also omits any structural description of the claimed layers. The term “layers” is a common of art applied to neural networks. But the Specification does not mention a neural network, a graph neural network, or any operations indicating the claimed layers are related as part of a neural networks. No other machine learning architecture is described in the Specification that gives any indication of the overall structure of the claimed layers. The Specification indicates that “the prediction layer may be configured to generate the prediction output by using a transformer-based sequence model.” Spec. p. 4. But the Specification is also clear that the public would be responsible for inventing the architecture of the other layers and the overall architecture of the claimed three-layer model: “However, at this point it should be noted that the invention is not restricted to any particular model architectures for the three layers specified. Many graph embedding mechanisms (for the graph embedding layer), neighbor aggregation functions (for computing the event embeddings), and sequence models have been proposed in literature, and those skilled in the art will be able to identify and implement suitable embodiments of these layers.” Spec. pp. 4-5 (emphasis added). Merely indicating the general function of each layer is at the discretion of the person of ordinary skill in the art and suggesting that one of ordinary skill in the art refer to “mechanisms . . . functions . . . and sequence models . . . proposed in the literature” to effectively create their own three-layer model supports a finding that the Specification has failed to include a sufficient description of the claimed subject matter. To be clear, the description does describe some aspects of each of the layers. According to the description, the “graph embedding layer” (first layer) may be trained to embed each node and relationship type in a fused event and knowledge graph, as embedding vectors. See Spec. ¶34. The “event embedding layer” (second layer) receives the graph embedding vectors from the graph embedding layer and assigns a “fixed-dimensional embedding vector” to each “event.” Spec. ¶35. The “prediction layer” (third layer) receives a sequence of event embeddings from the second layer and outputs a prediction. Spec. ¶37. The graph embedding layer is trained in an unsupervised manner to generate an embedding vector. Spec. ¶34, 39. The event embedding layer may be trained end to end in a supervised manner to generate an embedding vector. Spec. ¶35, 40. The prediction layer may be trained end to end in a supervised manner to generate predictions. Spec. ¶¶37, 40. None of this describes the structures of the layers themselves. Further, the claims recite “building up and training a three-layered prediction model” but the Specification fails to offer any explanation showing possession of any particular method of training the undescribed “layers,” much less training of a complete “three-layered prediction model,” as claimed. Reliance of the state of the art is reasonable where the Specification clearly references a particular, known technique. Graph neural networks that combine graphs and neural networks are known in the art. But the Specification does not indicate or imply that the claimed invention is a graph neural network or indicate how any particular graph neural network would be configured in a way that supports training of the claimed three-layer prediction model. That one of ordinary skill in the art could dig around the literature until they came up with a way of “building-up and training a three-layer model” may be sufficient for enablement. But merely invoking “literature” in the Specification does not replace providing the support required to show possession of the claimed invention. Ultimately, the Specification indicates general types of models that could potentially accomplish the operations attributed to each of the layers, without providing any particular overall model architecture or training scheme for the claimed three-layered prediction model. See Spec. ¶¶34-40. Without any example of the structure of the claimed “three-layer model” the description does not support the claims in their current form. “The written description requirement is not necessarily met when the claim language appears in ipsis verbis in the specification. ‘Even if a claim is supported by the specification, the language of the specification, to the extent possible, must describe the claimed invention so that one skilled in the art can recognize what is claimed. The appearance of mere indistinct words in a specification or a claim, even an original claim, does not necessarily satisfy that requirement.’” MPEP § 2163.03 (emphasis added).
All dependent claims are rejected as containing the limitations of the claims from which they depend.
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-15 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor, or for pre-AIA the applicant regards as the invention.
Generally: separately listed claim elements are construed as distinct components, that all claim terms must be given weight, there is presumed to be a difference in meaning and scope when different words or phrases are used in separate claims, and repeated and consistent descriptions in the specification indicate the proper scope of a claimed term. “[C]laims must ‘conform to the invention as set forth in the remainder of the specification and the terms and phrases used in the claims must find clear support or antecedent basis in the description so that the meaning of the terms in the claims may be ascertainable by reference to the description.’ 37 C.F.R. § 1.75(d)(1).” Phillips v. AWH Corp., 415 F.3d 1303, 1316 (Fed. Cir. 2005) (as cited in MPEP § 2111). Therefore, use of two different terms in the claims that both rely on the description of a single structure in the Specification may render at least one term indefinite because there is no way to determine which term should be construed in view of the description of the single structure.
All independent claims recite “the event sequence used as input” without antecedent basis. It is not clear whether the event sequence used as input refers to the “events” included in the “fused event and knowledge graph,” thereby requiring that the events be sequential, or if the event sequence is meant to introduce a new claim element, and the use of the definite article is merely a typo.
The independent claims fail to adequately describe the overall structure of the claimed “three-layered prediction model” including the structures of any of the layers within the model. The only drawing of the model is box 160 on figure 1. That is, the only drawing of the claimed model appears to be a rectangle labeled “Prediction Model (for actions 1,…,x),” incorporated into the feedback loop of control scheme. Clearly this is insufficient to show possession of the three-layered model recited in the claims. The Specification also omits any structural description of the claimed layers. The term “layers” is a common term used in machine learning and would support a specific structure if recited, for instance, in reference to a neural network. But the Specification does not mention a neural network or any operations indicating the claimed layers are related to neural networks, or a graph neural network. Nothing in the Specification describes operations that would be understood by one of ordinary skill as clearly referring to layers of a neural network. No other machine learning architecture is described in the Specification that gives any indication of the structure of the claimed layers. The Specification names three layers. According to the description, the “graph embedding layer” (first layer) may be trained to embed each node and relationship type in a fused event and knowledge graph, as embedding vectors. See Spec. ¶34. The “event embedding layer” (second layer) receives the graph embedding vectors from the graph embedding layer and assigns a “fixed-dimensional embedding vector” to each “event.” Spec. ¶35. The “prediction layer” (third layer) receives a sequence of event embeddings from the second layer and outputs a prediction. Spec. ¶37. The graph embedding layer is trained in an unsupervised manner to generate an embedding vector. Spec. ¶34, 39. The event embedding layer may be trained end to end in a supervised manner to generate an embedding vector. Spec. ¶35, 40. The prediction layer may be trained end to end in a supervised manner to generate predictions. Spec. ¶¶37, 40. None of this describes the structures of the layers themselves. Further, the specification fails to explain any particular method of training the undescribed “layers” that would be consistent with relying on the state of the art to show how one of ordinary skill in the art should interpret the term “layer” within this application. While combining neural networks with graphs is known in the art and such graph neural networks would not be indefinite, if claimed, the Specification does not describe this combination or use terms that would imply the claimed invention is a version of a graph neural network. “The meaning of every term used in a claim should be apparent from the prior art or from the specification and drawings at the time the application is filed. Claim language may not be "ambiguous, vague, incoherent, opaque, or otherwise unclear in describing and defining the claimed invention." In re Packard, 751 F.3d 1307, 1311, 110 USPQ2d 1785, 1787 (Fed. Cir. 2014). Applicants need not confine themselves to the terminology used in the prior art, but are required to make clear and precise the terms that are used to define the invention whereby the metes and bounds of the claimed invention can be ascertained.” MPEP § 2173.05(a) (emphasis added).
All independent claims substantially recite “wherein the second layer is an event embedding layer that assigns to each event of the process instance a fixed-dimensional event embedding vector, and wherein the third layer is a prediction layer that receives as input a sequence of event embeddings from the second layer and that generates as output a prediction of an unknown property of the event sequence used as input.” It is not clear whether the “fixed-dimensional event embedding vector” output by the “second layer [that] is an event embedding layer” refers to the same element as the “sequence of event embeddings” received by the third layer from the second/event embedding layer. Different terms generally indicate separate claim elements, but an “event embedding vector” output by an “event embedding layer” sound very similar to a “sequence of event embeddings.” Further, if the terms are different, there is no clear connection between the second and third layer, which seems counter to the inventive concept of a “three-layered prediction model.” The sequence of event embeddings could also refer to a sequence of embedding vectors, but again, interpreting a sequence of “event embeddings” as a sequence of “event embedding vectors” would ignore the use of different language in the two terms. Since there are at least three ways the claim language could be read and none of which is clearly more reasonable than the others, the claim language is indefinite.
Claim 3 recites “the event embedding layer and the prediction layer are trained end-to-end in a supervised manner.” It is not clear whether each layer individually is trained end to end, or if both layers are trained end to end as a unit. Given the lack of explanation of the layer structure in the Specification, reference to the Specification cannot be used to reconcile this ambiguity.
Claim 9 recites “wherein the prediction layer generates the prediction output by using a transformer-based sequence model.” This language appears to be inconsistent with the plain meaning of “layer” or with the plain meaning of “transformer” or both. It is not clear how to interpret this claim language to reconcile the inconsistency. Generally, a “transformer based sequence model” would be inconsistent with a single “layer” of a neural network. It is not clear whether this claim is directed to a single layer version of a transformer based sequence model, or if “layer” is being used in reference to potentially any number of layers interconnected in any fashion. The claim language could also be read as a layer that is separate from the transformer model. Since it is not clear whether the layer is a transformer or if the layer somehow communicates with or creates a transformer, the language is indefinite.
All dependent claims are rejected as containing the limitations of the claims from which they depend.
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claim 1, 3-5, 7-8, and 14-15 are anticipated under 35 U.S.C. § 102(a)(1) by Estaban (Predicting the Co-Evolution of Event and Knowledge Graphs, 2016).
Claim 1 (Currently Amended): A computer-implemented method for event sequence forecasting of a process instance, the method comprising: building up and training a three-layered prediction model including a first, a second and a third layer, (Estaban teaches “Thus, we train a separate prediction model which estimates future events based on the latent representations of previous events in the event tensor and the latent representations of the involved generalized entities in the KG tensor.” Estaban P. 1. The three-layered aspect of the prediction model is addressed below, where each layer is recited.)
wherein the first layer is a graph embedding layer that assigns a fixed-dimensional graph embedding vector to each node and relation type (“Here we consider a slight extension to the subject-predicate object triple form by adding the value (es; ep; eo; Value) where Value is a function of s; p; o and can be the truth value of the triple or it can be a measurement. Thus (Jack, likes, Mary; True) states that Jack likes Mary, and (Jack, hasBloodTest, Cholesterol; 160) would indicate a particular blood cholesterol level for Jack. Note that es and eo represent the entities for subject index s and object index o.” Esteban p. 2, col. 1. “In representation learning, one assigns an r-dimensional latent vector to the entity e denoted by ae = (ae;0; ae;1; : : : ; ae;r)T.” Esteban p. 2 col. 2.) in a fused event and knowledge graph that contains available structural information including events, knowledge graph nodes, and links between the events and the knowledge graph nodes, (“Here we consider a slight extension to the subject-predicate-object triple form by adding the value (es; ep; eo; Value) where Value is a function of s; p; o and can be the truth value of the triple or it can be a measurement. Thus (Jack, likes, Mary; True) states that Jack likes Mary, and (Jack, hasBloodTest, Cholesterol; 160) would indicate a particular blood cholesterol level for Jack. Note that es and eo represent the entities for subject index s and object index o.” Esteban p. 2 col. 2. See also Esteban Fig. 1 and description of the KG and event arrival. The “first layer” reads on the operations resulting in the embedding of the knowledge graph.)
wherein the second layer is an event embedding layer that assigns to each event of the process instance a fixed-dimensional event embedding vector, and (“At each time step events form triples which form a sparse triple graph and which specifies which facts become available. The event tensor is a four-way tensor Z with (es; ep; eo; et; Value) and tensor elements zs;p;o;t. We have introduced the generalized entity et to represent time. Note that the characteristics of the KG tensor and the event tensor are quite different.” Esteban p. 3. col. 1. The “second layer” reads on operations resulting in the embedding of the event tensor.)
wherein the third layer is a prediction layer that receives as input a sequence of event embeddings from the second layer and that generates as output a prediction of an unknown property of the event sequence used as input. (“The key idea of the paper is that events are predicted using both latent representations of the KG and latent representations describing recently observed events. In the prediction model we estimate future entries in the event tensor Z. The general form is:
PNG
media_image1.png
200
400
media_image1.png
Greyscale
where the first version uses a single function and the latter uses a different function for each (p; o)-pair.3 Here, args is from the sets of latent representations from the KG tensor and the event tensor. An example of a prediction model is
PNG
media_image2.png
200
400
media_image2.png
Greyscale
where the prediction is based on the latent representations of subject, object and predicate from the KG-tensor and of the time-specific representations from the event tensor.” Esteban p. 3. The “fpredict” reads on the “third layer.” “Then we learn the latent representations of the entities that compose the tensor and use them to predict unobserved facts.” Esteban Abstract.)
Claim 3 (Currently Amended): The method according to claim 1, wherein the event embedding layer and the prediction layer are trained end-to-end in a supervised manner. (“Here, aes is the profile of the patient, calculated from the KG model. Being constant, aes assumes the role of parameters in the prediction model. aes;t describes all that so far has happened to the patient at the same instance in time t (e.g., on the same day). aes;t-1 describes all that happened to the patient at the last instance in time and so on. We model the functions by a multiway neural network with weight parameters W exploiting the great modeling flexibility of neural networks. See also Appendix on last page showing separate cost functions for the semantic KG model and for the prediction model. Knowledge of the occurrences at a given time implies labeled data, and therefore supervised training.)
Claim 4 (Currently Amended): The method according to claim 1, wherein an event history containing previous event sequences with known outcomes is used as training samples for training the event embedding layer. (“Here, aes is the profile of the patient, calculated from the KG model. Being constant, aes assumes the role of parameters in the prediction model. aes;t describes all that so far has happened to the patient at the same instance in time t (e.g., on the same day). aes;t-1 describes all that happened to the patient at the last instance in time and so on. We model the functions by a multiway neural network with weight parameters W exploiting the great modeling flexibility of neural networks.)
Claim 5 (Currently Amended): The method according to claim 1, wherein the event embedding layer determines the event embedding vectors by using a parametrized function that maps a graph neighborhood of any event in the fused event and knowledge graph to a fixed-dimensional vector. (See rejection of claim 1.)
Claim 7 (Currently Amended): The method according to claim 1, further comprising, upon arrival of a new event, using the trained prediction layer to dynamically update predictions. (“Our goal will be to predict the events that will happen in future time steps, using for that task both dynamic information from the previous event tensors and static information that is stored in the knowledge graph. Therefore, this architecture allows us to fuse static and dynamic information to predict future events.” Esteban Abstract.)
Claim 8 (Currently Amended): The method according to claim 7, wherein updating predictions comprises: using the event embedding layer to compute an embedding of the event from the embeddings of the related nodes of the fused event and knowledge graph; and feeding the resulting embedding into the prediction layer to update the predictions about the event sequence. (See Esteban fig. 1. (“Fig. 1. The figure shows an example where the event tensor is predicted from the representations of the events in the last two time steps and from the KG representation. The dotted line indicate the transfer of observed events into the KG.”) “In our model, each change in the status of the KG is communicated via events. Thus each change in the KG first appears in the event tensor and predictions of events also implies predictions in the KG. The events that change the KG status are transferred into the KG and the latent representations of the KG, i.e., aes ; aep ; aeo, are re-estimated regularly (Figure 1).” Esteban p. 4.)
For rejections of claims 14 and 15, see rejection of claim 1.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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 2 and 6 are unpatentable under 35 U.S.C. § 103 in view of Estaban and Rossi (Temporal Graph Networks for Deep Learning On Dynamic Graphs, 2020).
Claim 2 (Original): The method according to claim 1, wherein the graph embedding layer is trained in an unsupervised manner. (The previously cited art does not expressly teach unsupervised training.
“TGN can be trained for a variety of tasks such as edge prediction (self-supervised) or node classification (semi-supervised).” Rossi p. 5. “The key contribution of this paper is a novel Temporal Graph Network (TGN) encoder applied on a continuous-time dynamic graph represented as a sequence of time-stamped events and producing, for each time t, the embedding of the graph nodes Z(t) = (z1(t); : : :; zn(t)(t)).” Rossi p. 2.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Rossi because unsupervised learning requires less human effort (i.e. less labeled data).)
Claim 6 (Currently Amended): The method according to claim 5, wherein the graph neighborhood of an event comprises a 2-hop neighborhood of the event. (The primary reference does not discuss hops in a neighborhood.
Rossi teaches “Temporal Graph Attention (attn): A series of L graph attention layers compute i’s embedding by aggregating information from its L-hop temporal neighborhood.” Rossi p. 4. “Each layer amounts to performing multi-head-attention (Vaswani et al., 2017) where the query (q(l)(t)) is a reference node (i.e. the target node or one of its L - 1-hop neighbors), and the keys K(l)(t) and values V(l)(t) are its neighbors. Finally, an MLP is used to combine the reference node representation with the aggregated information.” Rossi p. 4.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Rossi because this is a way of utilizing the greater relationship between events that occur more closely in time, within the data.
Claim 9 is unpatentable under 35 U.S.C. § 103 in view of Estaban and Coke (Contextualized Knowledge Graph Embedding; April 2020).
Claim 9 (Currently Amended): The method according to claim 1, wherein the prediction layer generates the prediction output by using a transformer-based sequence model. (The primary reference dies not teach transformer models.
Coke teaches “CoKE takes a sequence as input and uses a Transformer encoder to obtain contextualized representations. These representations are hence naturally adaptive to the input, capturing contextual meanings of entities and relations therein.” Coke Abstract.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Coke because this technique helps capture contextual meanings of relations in the data.)
Claims 10-11 are unpatentable under 35 U.S.C. § 103 in view of Estaban and Allahdadian (2022/0188410; filed 2020, different assignee)
Claim 10 (Currently Amended): The method according to claim 1, further comprising: re-training the third layer, or the second and third layer, or all three layers, as soon as an amount of new events that has arrived exceeds a critical mass defined by a configurable threshold. (The previously cited art does not teach retraining.
Allahdadian teaches “[0095] Step 402 applies the reconstructive model to a new dataset that may be a very long stream of inputs or a very long sequence of batches such that concept drift occurs and feature(s) are suppressed to compensate for model decay. How much model decay is tolerable depends on the implementation. Eventually, step 403 may detect that a count of suppressed features (i.e. features removed from the unsuppressed subset of features) exceeds a retrain threshold such as a percentage or absolute amount of features. . . .[0096] For example, suppression of five features of a hundred features may be tolerable but suppression of more than ten of those features may break anomaly detection, usually by generating too many false negatives. In other words, too many suppressed features causes anomalies to be undetected. . . . [0097] Retraining the reconstructive model with recent inputs would recalibrate the reconstructive model so that the reconstructive model could operate without suppressing any features. In other words, retraining would eliminate the model decay. However, retraining may entail technical problems as follows. . . . [0098] Retraining may take hours, days, or weeks, depending on how large of a training corpus is needed. If the retrain threshold is too high such that many features need suppressing before the retrain threshold is exceeded, then retraining must begin immediately because the model has severely decayed, and immediate retraining may cause a service outage of immense duration. A somewhat lower retrain threshold is better for the following reasons.” Allahdadian ¶¶ 95-98. “[0101] In an embodiment shown with the dashed arrow, step 403 detects that the retrain threshold is exceeded, and step 406, as discussed later herein, immediately begins retraining. In another embodiment, steps 404-405 defer retraining as follows. [0102] As explained above, retraining is slow because a large training corpus is involved. Another slow phenomenon is accumulation of recent inputs after concept drift. As a corpus grows, historic inputs that occurred after concept drift eventually predominate over inputs that occurred before concept drift. . . . [0103] In other words, additional time may be needed for the corpus to adequately reflect the drift. Retraining too soon and before sufficient accumulation of drifted inputs may bias the training toward fitting outdated inputs from before the drift. Steps 404-405 prevent retraining too soon by deferring retraining as follows. . . . [0104] Even after step 403 detects that the retrain threshold is exceeded, step 404 continues to apply the reconstructive model to a growing amount of new data in the production environment. Eventually, step 405 detects that the accumulated amount of new data exceeds a sufficiency threshold, which means that the training corpus adequately reflects the scope and magnitude of input variations that can occur after concept drift.” Allahdadian ¶101. See also Allahdadian Fig. 4 showing different paths taken, one requiring a minimum amount of training data and the other omitting this requirement, the path taken based on the level of error (based on the inaccuracy of the features.)
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Allahdadian because this technique mitigates problems caused by concept drift while balancing against the resources required for more training.)
Claim 11 (Currently Amended): The method according to claim 10, wherein the configurable threshold defining the critical mass of new events is dynamically adjusted if a predictive accuracy drops below a certain threshold. (See rejection of claim 10.)
Claims 12-13 are unpatentable under 35 U.S.C. § 103 in view of Estaban and Hafner (Reinforcement learning in feedback control, 2010).
Claim 12 (Currently Amended): The method according to claim 1, further comprising: using the obtained predictions for automated decision making in a control loop. (The previously cited art does not teach a control loop.
Hafner teaches “Technical process control is a highly interesting area of application serving a high practical impact. Since classical controller design is, in general, a demanding job, this area constitutes a highly attractive domain for the application of learning approaches—in particular, reinforcement learning (RL) methods. RL provides concepts for learning controllers that, by cleverly exploiting information from interactions with the process, can acquire high quality control behaviour from scratch.” Hafner Abstract.
It would have been obvious to one of ordinary skill in the art before the effective filing date to deploy the three-layer model in a feedback control system because one of ordinary skill in the art would recognize that applying machine learning to a feedback control system can decrease the work required to manually design and/or improve control system.)
Claim 13 (Currently Amended): The method according to claim 12, wherein the control loop is a closed control loop (See Hafner Fig. 1.) that receives as input a desired outcome, (See Hafner Fig. 1, showing a setpoint.) that uses the trained prediction model (“Learning the complete control law from scratch is also the underlying scenario for the control tasks presented here—meant as a challenge for the reinforcement learning algorithms under examination. Of course, in a practical application, it is often advisable to use as much prior knowledge as available, and for example to start with the best controller that can be designed in a classical way and use the learning controller then to improve over the inadequacies of the designed controller.” Hafner p. 138.) to predict the outcome for several possible actions, leading to several expected outcomes, and that includes a control element that selects actions that lead to an expected outcome having the smallest difference to the desired outcome. (“The classical feedback-control loop describes the application-specific influence of a control device on a controlled process. Within this interaction loop the control device applies appropriate control actions, u, to bring the controlled process variables, y, in close proximity to external set-points or reference inputs, w. A deviation of the controlled process variables from the set-point can occur due to external disturbances on the process and/or to the external change of the set-point. The decision of the control device is based on information that is fed back from the process. In this article we refer to the feedback control schematic depicted in Fig. 1.” Hafner p. 140.)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PAUL M KNIGHT whose telephone number is (571) 272-8646. The examiner can normally be reached Monday - Friday 9-5 ET.
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 Bechtold can be reached on (571. 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.
PAUL M. KNIGHT
/PAUL M KNIGHT/
Primary Examiner, Art Unit 2148
Prosecution Insights