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 .
This action is responsive to the Amendment filed on 7/28/2026.
Claims 1-20 are pending in the case.
Response to Arguments
Applicant's arguments and amendments with regards to the 35 U.S.C. § 101 rejection of claim(s) 1-20 have been fully considered and are not persuasive. The 35 U.S.C. § 101 rejection of claim(s) 1-20 is respectfully maintained.
Regarding claim 1, applicant argues,
(A) “For example, amended claim 1 recites, inter alia, a "module configured to obtain prediction probabilities from multiple, different layers of the multi-layer machine learning model." Applicant respectfully submits that, as a practical matter, one cannot obtain prediction probabilities from multiple layers of a multi-layer machine learning model entirely within the mind. Accordingly, Applicant respectfully submits that claim 1 cannot be directed to a mental process, and respectfully requests reconsideration”.
Examiner respectfully disagrees. Claim 1 does not limit the manner in which the information (i.e. probabilities from layers of the multi-layer machine learning model) is obtained, and the broad interpretation of the limitation includes performing the step in the mind.
(B) “Furthermore, with respect to the notion that the claims are directed to mathematical concepts, Applicant respectfully submits that while the claims may involve mathematical concepts, they are not directed to such concepts. For example, although embodiments of the claimed invention may aggregate outputs from layers of the multi-layer machine learning model, Applicant respectfully submits that the claims are nonetheless directed to improving machine learning outputs. Accordingly, Applicant respectfully submits that the claims are not directed to the abstract idea of mathematical concepts, and respectfully requests reconsideration and withdrawal”.
Examiner respectfully disagrees. Applicant’s amount to a general allegation that the claims are patent eligible without specifically pointing out how the language of the claims overcome the 35 U.S.C. § 101 rejection of claim(s) 1.
(C) Examiner notes that on Pages 9-12 of 23 of their Remarks, applicant has cited multiple paragraphs of their disclosure, and further argues
“Lastly, the claims embody the solution to the problems within the current state of the art identified in the specification by reciting, inter alia:
an autocontrastive decoding module configured to obtain prediction probabilities from
multiple, different layers of the multi-layer machine learning model through which[[as]] data propagates [[through]]in the multi-layer machine learning model, and aggregate in a
contrastive manner the prediction probabilities from the multiple, different layers of the
multi-layer machine learning model to provide a final output from the multi-layer
machine learning model.
Accordingly, as illustrated above, Applicant respectfully submits that the claims, as
amended, explicitly reflect the recited improvement of using outputs from multiple layers of a machine learning model to improve model outputs. Therefore, because claim 1 integrates its features into practical applications (specifically, practical applications of a machine learning
system) that apply, rely on, or use any alleged abstract ideas in a manner that imposes meaningful limits, claim 1 fails Prong Two of the USPTO's required analysis.
As such, for at least the reasons discussed above, claim 1, as well as similarly amended independent claims 9 and 13, are not directed to an abstract idea, nor are the claims dependent thereon at least by virtue of their dependency on non-abstract independent claims, and for the additional features that they recite. Applicant respectfully requests that the 35 U.S.C. § 101 rejections of these claims be reconsidered and withdrawn”.
Examiner respectfully disagrees. The broad interpretation of claim 1, merely requires the final output layer to contrastively aggregate the probabilities from multiple layers of a model, which does not constitute an invention where….“may have integrated the exception into a practical application [include]: an additional element reflects an improvement in the functioning of a computer, or an improvement to other technology or technicalfield;... an additional element implements a judicial exception with, or uses a judicial exception in conjunction with, a particular machine or manufacture that is integral to the claim; an additional element effects a transformation or reduction of a particular article to a different state or thing; and an additional element applies or uses the judicial exception in some other meaningful way beyond generally linking the use of the judicial exception to a particular technological environment, such that the claim as a whole is more than a drafting effort designed to monopolize the exception.”.
Examiner further notes that the claim(s) recite using a previously trained machine learning model as a tool.
Therefore Examiner asserts that amended claim 1 is patent ineligible and remains rejected under 35 U.S.C. § 101.
Regarding claim(s) 2-20 applicant argues that the claims are eligible under 35 U.S.C. § 101 for reasons similar to those references with regards to claim 1
For reasons similar to those discussed above with regards to claim 1 examiner asserts that
Claims 2-20 are also patent ineligible under 35 U.S.C. § 101.
Applicant's arguments and amendments with regards to the 35 U.S.C. § 112(b) rejection of claim(s) 6 and 10, for the recitation of “multi-exit model” have been fully considered and are persuasive. The 35 U.S.C. § 112(b) rejection of claim(s) 6 and 10, for the recitation of “multi-exit model” is respectfully withdrawn.
Applicant's arguments and amendments with regards to the 35 U.S.C. § 112(b) rejection of claim(s) 6-8, 10-12, for the recitation of “linear amateur head” and “linear expert head”
have been fully considered and are not persuasive. The 35 U.S.C. § 112(b) rejection of claim(s) 6-8, 10-12 for the recitation of “linear amateur head” and “linear expert head” is respectfully maintained.
Regarding claim(s) 6-8, 10-12, applicant argues
“Applicant respectfully submits that the as-filed specification provides support for use of the aforementioned terms. For example, FIG. 5 and paragraphs [0059] - [0062] of the specification illustrate and recite
PNG
media_image1.png
202
640
media_image1.png
Greyscale
PNG
media_image1.png
202
640
media_image1.png
Greyscale
PNG
media_image2.png
127
642
media_image2.png
Greyscale
PNG
media_image3.png
133
642
media_image3.png
Greyscale
As noted above, "a language model head is a layer where the model output (a vector) is projected linearly to obtain a next token prediction from the model." With additional reference to FIG. 5, Applicant respectfully submits that this linear head is taken at an intermediate (amateur) layer at step 510 and from a final (expert) layer at 512. In other words, the amateur head connotes an intermediate output while the expert head connotes a final output. Furthermore, obtaining these outputs at multiple layers results in a "multi-exit" machine learning model in which outputs exit at different layers thereof. Accordingly, Applicant respectfully submits that the terms of the claims are known and supported”
Examiner respectfully disagrees with applicant’s arguments regarding for the recitation of “linear amateur head” and “linear expert head”. Examiner notes that applicant’s disclosure does not provide a definition for amateur and expert heads, and as noted by applicant ”linear head is taken at an intermediate (amateur) layer at step 510 and from a final (expert) layer at 512. In other words, the amateur head connotes an intermediate output while the expert head connotes a final output”.
It is unclear
what makes an intermediate layer an amateur head, i.e. how an intermediate layer that is a linear head differs from an intermediate layer that is a linear amateur head, and
what makes final output layer an expert head, i.e. how a final output layer that is a linear head differs from a final output layer that is a linear expert head,
rendering the claims indefinite.
Applicant's arguments and amendments with regards to the 35 U.S.C. § 102 and 103 rejection of claim(s) 1-20 have been considered, but are not persuasive. Applicant argues that the amended claims are allowable due to the following:
Regarding claim(s) 1 and 13 applicant argues
PNG
media_image4.png
466
653
media_image4.png
Greyscale
PNG
media_image5.png
317
649
media_image5.png
Greyscale
Examiner respectfully disagrees. Goyal [19, 20, 22, 25, 27] discloses decoder receives prediction probabilities from multiple layers, output is formed using the probabilities and gating mechanism provides different inference strategies including mixture of weights for different probabilities, to provide diverse summaries (contrastive),and
Goyal [20, 22, 29] discloses machine learning model may include decoders which have multiple layers including hidden layers (intermediate), which feed to later layers including the final layer to provide output, i.e. the decoder layers are different layers of the same model and the final output layer may aggregate their outputs, and
Goyal therefore sufficiently teaches obtain prediction probabilities from multiple, different layers of the multi-layer machine learning model through which data propagates in the multi-layer machine learning model, and aggregate in a contrastive manner the prediction probabilities from the multiple, different layers of the multi-layer machine learning model to provide a final output from the multi-layer machine learning model.
Regarding claim(s) 2-20 applicant argues that the claims are allowable for reasons similar to those references with regards to claim 1 above.
For reasons similar to those discussed above with regards to claim 1, examiner asserts that the previously cited art sufficiently teaches the amended claims 1-20.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier.
Claim limitations
"autocontrastive decoding module configured to obtain prediction probabilities from multiple, different layers of the multi-layer machine learning model as data propagates through the multi-layer machine learning model, and aggregate in a contrastive manner the prediction probabilities from the multiple, different layers of the multi-layer machine learning model to provide a final output from the multi-layer machine learning model", in claim 1, and
"autocontrastive decoding module configured to redistribute a prediction probability distribution of the transformer-based machine learning model by maximizing a difference between log-probabilities of a final layer and one or more intermediate layers of the transformer- based machine learning model", in claim 9,
have been interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because they use a generic placeholder
"autocontrastive decoding module" coupled with functional language "configured to obtain prediction probabilities from multiple, different layers of the multi-layer machine learning model as data propagates through the multi-layer machine learning model, and aggregate in a contrastive manner the prediction probabilities from the multiple, different layers of the multi-layer machine learning model to provide a final output from the multi-layer machine learning model", in claim 1, and
"autocontrastive decoding module" coupled with functional language "configured to redistribute a prediction probability distribution of the transformer-based machine learning model by maximizing a difference between log-probabilities of a final layer and one or more intermediate layers of the transformer- based machine learning model", in claim 9.
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
A review of the specification shows that the following appears to be the corresponding structure described in the specification for the 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph limitation: according to [0029] and [0048], "autocontrastive decoding module” in claims 1 and 9, are implemented as hardware/ processor(s) executing software modules to perform the instructions of the modules
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
Claim Rejections - 35 USC § 112
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 6-8, 10-12, 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.
Claim(s) 6, 7, 10, 11, each recite(s) “linear amateur head” and “linear expert head”.
“Amateur head” and “Expert head” are not well-known terms in the art, and it is unclear what makes a head “amateur” versus “expert”, rendering the claim(s) indefinite.
Examiner notes that applicant’s disclosure does not provide a definition for amateur and expert heads, and as noted by applicant ”linear head is taken at an intermediate (amateur) layer at step 510 and from a final (expert) layer at 512. In other words, the amateur head connotes an intermediate output while the expert head connotes a final output”.
It is unclear
what makes an intermediate layer an amateur head, i.e. how an intermediate layer that is a linear head differs from an intermediate layer that is a linear amateur head, and
what makes final output layer an expert head, i.e. how a final output layer that is a linear head differs from a final output layer that is a linear expert head,
rendering the claims indefinite.
Claim(s) 7-8 and 11-12 do not contain claim limitations that cure the indefiniteness of claim(s) 6 and 10 respectively, and therefore are also indefinite under 35 U.S.C. 112(b).
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(s) 1-20 is/are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1:
Claim(s) 13-20 is/are method type claim. Claim(s) 1-12 is/are system type claim(s). Therefore, claims 1-20 is/are directed to either a process, machine, manufacture or composition of matter.
Independent claim(s):
Step 2A Prong 1:
Regarding claim(s) 1 this/these claim(s) recite(s) obtain prediction probabilities from multiple, different layers of the multi-layer machine learning model … in the multi-layer machine learning model, and aggregate in a contrastive manner the prediction probabilities from the multiple, different layers of the multi-layer machine learning model to provide a final output from the multi-layer machine learning mode,
Regarding claim(s) 9 this/these claim(s) recite(s) redistribute a prediction probability distribution of the transformer-based machine learning model by maximizing a difference between log-probabilities of a final layer and one or more intermediate layers of the transformer- based machine learning model,
Regarding claim(s) 13, this/these claim(s) recite(s) obtaining prediction probabilities from multiple, different layers of the multi-layer machine learning model as the data propagates through the multi-layer machine learning model; and aggregating in a contrastive manner the prediction probabilities from the multiple, different layers of the multi-layer machine learning model to provide a final output from the multi-layer machine learning model.
The above limitations appear to be practically implementable in the human mind and is understood to be a recitation of a mental process and math.
Step 2A Prong 2:
Regarding claim(s) 1, 9 and 13, this judicial exception is not integrated into a practical application.
Additional elements:
Regarding claim(s) 1 and 9, this/these claim(s) recite(s) a system for machine learning, the system comprising: ... an autocontrastive decoding module.
The modules have been interpreted to be implemented as generic hardware/ processor(s) executing software modules to perform the instructions of the modules (generic computer processor component to apply the exception).
Regarding claim(s) 1, this/these claim(s) further recite(s)
different layers of the multi-layer machine learning model through which data propagates (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic model with multiple layers to make a prediction).
Regarding claim(s) 9, this/these claim(s) further recite(s)
a transformer-based machine learning model through which data propagates (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic transformer-based model with multiple layers to make a prediction).
Regarding claim(s) 13 this/these claim(s) further recite(s) providing data as input
a multi-layer machine learning model through which the data propagates (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic model with multiple layers to make a prediction).
The additional element(s) as disclosed above alone or in combination do not integrate the judicial exception into practical application as they are generic computer functions being implemented with generic computer elements in a high level of generality to perform the disclosed abstract idea above.
Therefore, the claim(s) is/are directed to an abstract idea.
Step 2B:
Regarding claim(s) 1, 9 and 13, this/these claim(s) do/does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Additional elements:
Regarding claim(s) 1 and 9, this/these claim(s) recite(s) a system for machine learning, the system comprising: ... an autocontrastive decoding module.
The modules have been interpreted to be implemented as generic hardware/ processor(s) executing software modules to perform the instructions of the modules (generic computer processor component to apply the exception).
Regarding claim(s) 1, this/these claim(s) further recite(s) multi-layer machine learning model (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic model to make a prediction).
Regarding claim(s) 9, this/these claim(s) further recite(s)
a transformer-based machine learning model (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic transformer-based model to make a prediction).
Regarding claim(s) 13 this/these claim(s) further recite(s) providing data as input
a multi-layer machine learning model (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic model to make a prediction).
The additional element(s) as disclosed above in combination of the abstract idea are not sufficient to amount to significantly more than the judicial exception as they are generic computer functions being implemented with generic computer elements in a high level of generality to perform the disclosed abstract idea above.
Therefore, the claim(s) is/are not patent eligible.
Step 2A Prong 1, Dependent claims:
Regarding claim(s) 4 and 16, this/these claim(s) recite(s) aggregate the first prediction probabilities from at least one of the intermediate layers with the second prediction probabilities from the final output layer, while changing the second prediction probabilities from the final output layer according to the first prediction probabilities obtained from at least one of the intermediate layers,
Regarding claim(s) 7, 11 and 20, this/these claim(s) recite(s) map an output of the at least one of the intermediate layers to a first next-token probability distribution, and wherein the linear expert head is configured to map an output of the final output layer to a second next-token probability distribution,
Regarding claim(s) 8 and 12, this/these claim(s) recite(s) contrast the first next-token probability distribution with the second next-token probability distribution to provide a token-level probability distribution as the final output from the transformer-based machine learning model,
Regarding claim(s) 19, this/these claim(s) recite(s) redistributing a prediction probability distribution ...by maximizing a difference between log-probabilities of the final output layer and at least one of the intermediate layers
The above limitations appears to be practically implementable in the human mind and is understood to be a recitation of a mental process and math.
Step 2A Prong 2, Dependent claims:
Regarding claim(s) 2 and 14, this/these claim(s) recite(s) wherein the multi-layer machine learning model comprises one or more intermediate layers and a final output layer (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic multi-layered model to make a prediction),
Regarding claim(s) 3 and 15, this/these claim(s) recite(s) obtain (Adding insignificant extra-solution activity (receiving information) to the judicial exception - see MPEP 2106.05(g). Furthermore, MPEP 2106.05(d)(II) indicate that merely “Receiving or transmitting data” buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) is a well‐understood, routine, conventional function when it is claimed in a merely generic manner (as it is in the present claim). Thereby, a conclusion that the claimed obtaining step is well-understood, routine, conventional activity is supported under Berkheimer),
first prediction probabilities from at least one of the intermediate layers, and obtain second prediction probabilities from the final output layer (These limitations appear to be directed to the specification of where information is to be received from, and is understood to be generally linking the use of the judicial exception to a particular technological environment or field of use, which is not indicative of integration into a practical application. MPEP 2106.05(h)).
Regarding claim(s) 5 and 17, this/these claim(s) recite(s) wherein the multi-layer machine learning model comprises a transformer-based machine learning model (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic multi-layered transformer-based model to make a prediction).
Regarding claim(s) 6, 10, 18 this/these claim(s) recite(s) wherein at least one of the one or more intermediate layers comprises a linear amateur head, and wherein the final output layer comprises a linear expert head, making the transformer-based machine learning model a multi-exit model (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic multi-exit model with generic recitation of linear amateur head and linear expert head in model layers).
Step 2B, Dependent claims:
Regarding claim(s) 2 and 14, this/these claim(s) recite(s) wherein the multi-layer machine learning model comprises one or more intermediate layers and a final output layer (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic multi-layered model to make a prediction),
Regarding claim(s) 3 and 15, this/these claim(s) recite(s) obtain (Adding insignificant extra-solution activity (receiving information) to the judicial exception - see MPEP 2106.05(g). Furthermore, MPEP 2106.05(d)(II) indicate that merely “Receiving or transmitting data” buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) is a well‐understood, routine, conventional function when it is claimed in a merely generic manner (as it is in the present claim). Thereby, a conclusion that the claimed obtaining step is well-understood, routine, conventional activity is supported under Berkheimer),
first prediction probabilities from at least one of the intermediate layers, and obtain second prediction probabilities from the final output layer (These limitations appear to be directed to the specification of where information is to be received from, and is understood to be generally linking the use of the judicial exception to a particular technological environment or field of use, which is not indicative of integration into a practical application. MPEP 2106.05(h)).
Regarding claim(s) 5 and 17, this/these claim(s) recite(s) wherein the multi-layer machine learning model comprises a transformer-based machine learning model (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic multi-layered transformer-based model to make a prediction).
Regarding claim(s) 6, 10, 18 this/these claim(s) recite(s) wherein at least one of the intermediate layers comprises a linear amateur head, and wherein the final layer comprises a linear expert head, making the transformer-based machine learning model a multi-exit model (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level application of a previously trained generic multi-exit model with generic recitation of linear amateur head and linear expert head in model layers).
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(s) 1-5, 13-17, is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Goyal (US 20230065155 A1).
Regarding claim 1, Goyal teaches a system for machine learning, the system comprising: a multi-layer machine learning model; and an autocontrastive decoding module configured to (Goyal [13-16, 32, 33, system processor executes software stored in memory to perform operations by a model including a decoder, Goyal [2, 18, 20] decoder provides diverse summaries (autocontrastive)); and
obtain prediction probabilities from multiple, different layers of the multi-layer machine learning model through which data propagates in the multi-layer machine learning model, and aggregate in a contrastive manner the prediction probabilities from the multiple, different layers of the multi-layer machine learning model to provide a final output from the multi-layer machine learning model (Goyal [19, 20, 22, 25, 27] decoder receives prediction probabilities from multiple layers, output is formed using the probabilities and gating mechanism provides different inference strategies including mixture of weights for different probabilities, to provide diverse summaries (contrastive), Goyal [20, 22, 29] machine learning model may include decoders which have multiple layers including hidden layers (intermediate), which feed to later layers including the final layer to provide output).
Regarding claim 2, Goyal teach(es) the invention as claimed in claim(s) 1 above.
Goyal further teaches wherein the multi-layer machine learning model comprises one or more intermediate layers and a final output layer (Goyal [20, 22] model can have multiple hidden (intermediate layers) and bottom layer(s) ((final output layer)).
Regarding claim 3, Goyal teach(es) the invention as claimed in claim(s) 2 above.
Goyal further teaches wherein the autocontrastive decoding module is further configured to obtain first prediction probabilities from at least one of the intermediate layers, and obtain second prediction probabilities from the final output layer (Goyal [19, 22] intermediate and final layers may each have probabilities used to determine output).
Regarding claim 4, Goyal teach(es) the invention as claimed in claim(s) 3 above.
Goyal further teaches wherein the autocontrastive decoding module is further configured to aggregate the first prediction probabilities from at least one of the intermediate layers with the second prediction probabilities from the final output layer, while changing the second prediction probabilities from the final output layer according to the first prediction probabilities obtained from at least one of the intermediate layers (Goyal [19, 22] intermediate and final layers may each have probabilities used to determine aggregated weighted output, bottom layers’ probabilities may depend on probabilities earlier layers).
Regarding claim 5, Goyal teach(es) the invention as claimed in claim(s) 4 above.
Goyal further teaches wherein the multi-layer machine learning model comprises a transformer-based machine learning model (Goyal [10, 17] model may have a transformer-based encoder).
Regarding claim 13, Goyal teaches a machine learning method, comprising (Goyal [13-16, 32, 33, system processor executes software stored in memory to perform operations by a model including a decoder, Goyal [2, 18, 20] decoder provides diverse summaries (autocontrastive));
providing data as input to a multi-layer machine learning model (Goyal [23] input may be received by the model);
obtaining prediction probabilities from multiple, different layers of the multi-layer machine learning model through which the data propagates in the multi-layer machine learning model; and aggregating in a contrastive manner the prediction probabilities from the multiple, different layers of the multi-layer machine learning model to provide a final output from the multi-layer machine learning model (Goyal [19, 20, 22, 25, 27] decoder receives prediction probabilities from multiple layers, output is formed using the probabilities and gating mechanism provides different inference strategies including mixture of weights for different probabilities, to provide diverse summaries (contrastive) Goyal [20, 22, 29] machine learning model may include decoders which have multiple layers including hidden layers (intermediate), which feed to later layers including the final layer to provide output).
Regarding claim 14 Goyal teach(es) the invention as claimed in claim(s) 13 above.
Goyal further teaches wherein the multi-layer machine learning model comprises one or more intermediate layers and a final output layer(Goyal [20, 22] model can have multiple hidden (intermediate layers) and bottom layer(s) ((final output layer)).
Regarding claim 15, Goyal teach(es) the invention as claimed in claim(s) 14 above.
Goyal further teaches obtaining first prediction probabilities from at least one of the intermediate layers; and obtaining second prediction probabilities from the final output layer(Goyal [19, 22] intermediate and final layers may each have probabilities used to determine output).
Regarding claim 16, Goyal teach(es) the invention as claimed in claim(s) 15 above.
Goyal further teaches aggregating the first prediction probabilities from at least one of the intermediate layers with the second prediction probabilities from the final output layer, while changing the second prediction probabilities from the final output layer according to the first prediction probabilities obtained from at least one of the intermediate layers(Goyal [19, 22] intermediate and final layers may each have probabilities used to determine aggregated weighted output, bottom layers’ probabilities may depend on probabilities earlier layers).
Regarding claim 17, Goyal teach(es) the invention as claimed in claim(s) 14 above.
Goyal further teaches wherein the multi-layer machine learning model comprises a transformer-based machine learning model (Goyal [10, 17] model may have a transformer-based encoder).
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 (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 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 9, 19, 20, are rejected under 35 U.S.C. 103 as being unpatentable over Goyal (US 20230065155 A1), in view of Stahlberg (US 20230351125 A1).
Regarding claim 9, Goyal teaches a system for machine learning, the system comprising: a transformer-based machine learning model; and an autocontrastive decoding module configured to (Goyal [13-16, 32, 33, system processor executes software stored in memory to perform operations including a decoder, Goyal [17] model may have a transformer-based encoder, Goyal [2, 18, 20] decoder provides diverse summaries (autocontrastive)):
redistribute a prediction probability distribution of the transformer-based machine learning model by ... based on ... log-probabilities of a final layer and one or more intermediate layers of the transformer- based machine learning model through which data propagates (Goyal [19, 20, 22, 25-27] decoder calculates prediction log-probabilities from multiple layers, output is formed using the weighted (redistributed) log-probabilities and gating mechanism provides different inference strategies including mixture of weights for different probabilities, to provide diverse summaries (contrastive), Goyal [20, 22, 29] machine learning model may include decoders which have multiple layers including hidden layers (intermediate), which feed to later layers including the final layer to provide output).
Goyal does not specifically teach maximizing a difference between log-probabilities
However Stahlberg teaches redistribute a prediction probability distribution of the transformer-based machine learning model by maximizing a difference between log-probabilities of a final layer and one or more intermediate layers of the transformer- based machine learning model (Stahlberg [6, 11, 33, 34, 46, 65, 66] transformer based model may maximize difference between log-probabilities of different layers for better handling of ambiguity).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Stahlberg of redistribute a prediction probability distribution of the transformer-based machine learning model by maximizing a difference between log-probabilities of a final layer and one or more intermediate layers of the transformer- based machine learning model, into the invention suggested by Goyal; since both inventions are directed towards contrastive aggregation of prediction probabilities from different layers, and incorporating the teaching of Stahlberg into the invention suggested by Goyal would provide the added advantage of better handling of ambiguity, and the combination would perform with a reasonable expectation of success (Stahlberg [6, 11, 33, 34, 46, 65, 66]).
Regarding claim 19, Goyal teach(es) the invention as claimed in claim(s) 17 above.
Goyal further teaches redistributing a prediction probability distribution of the transformer-based machine learning model ... based on ... log-probabilities of the final output layer and at least one of the intermediate layers (Goyal [19, 20, 22, 25-27] decoder calculates prediction log-probabilities from multiple layers, output is formed using the weighted (redistributed) log-probabilities and gating mechanism provides different inference strategies including mixture of weights for different probabilities, to provide diverse summaries (contrastive)).
Goyal does not specifically teach maximizing a difference between log-probabilities
However Stahlberg teaches redistribute a prediction probability distribution of the transformer-based machine learning model by maximizing a difference between log-probabilities of a final layer and one or more intermediate layers of the transformer- based machine learning model (Stahlberg [6, 11, 33, 34, 46, 65, 66] transformer based model may maximize difference between log-probabilities of different layers for better handling of ambiguity).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Stahlberg of redistribute a prediction probability distribution of the transformer-based machine learning model by maximizing a difference between log-probabilities of a final layer and one or more intermediate layers of the transformer- based machine learning model, into the invention suggested by Goyal; since both inventions are directed towards contrastive aggregation of prediction probabilities from different layers, and incorporating the teaching of Stahlberg into the invention suggested by Goyal would provide the added advantage of better handling of ambiguity, and the combination would perform with a reasonable expectation of success (Stahlberg [6, 11, 33, 34, 46, 65, 66]).
Regarding claim(s) 20, Goya land Stahlberg teach(es) the invention as claimed in claim(s) 19 above.
Claim 10 further teaches linear amateur head and linear expert head.
Goyal further teaches providing the prediction probability distribution of the transformer-based machine learning model which has been redistributed as the final output from the multi-layer machine learning model (Goyal [19, 20, 22, 25-27] decoder calculates prediction log-probabilities from multiple layers, output is formed using the weighted (redistributed) log-probabilities and gating mechanism provides different inference strategies including mixture of weights for different probabilities, to provide diverse summaries (final output)).
Claims 6-8, 10-12, 18, are rejected under 35 U.S.C. 103 as being unpatentable over Goyal (US 20230065155 A1), in view of Stahlberg (US 20230351125 A1) and Kouris (US 20230128637 A1).
Regarding claim(s) 6, Goyal teach(es) the invention as claimed in claim(s) 5 above.
Goyal further teaches wherein at least one of the intermediate layers comprises a ... amateur head, and wherein the final output layer comprises a ... expert head (Goyal [0025] gating mechanism may give more weight (expert) to some layer heads than others (amateur));
Goyal does not specifically teach linear amateur head, linear expert head, making the transformer-based machine learning model a multi-exit model.
However Stahlberg teaches linear heads (Stahlberg [6, 11, 33, 34, 46, 65, 66] receives prediction probabilities from multiple layers, output is formed using the probabilities of different layers using strategies for better handling of ambiguity, Stahlberg [41, 52] output(s) may be formed using linear heads to allow projection of information into appropriate space(s)).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Stahlberg of linear heads, into the invention suggested by Goyal,
so that the amateur head and expert head of Goyal are linear heads,
resulting in wherein at least one of the intermediate layers comprises a linear amateur head, and wherein the final output layer comprises a linear expert head;
since both inventions are directed towards forming an output using the probabilities of different layers using strategies, and incorporating the teaching of Stahlberg into the invention suggested by Goyal would provide the added advantage of better handling of ambiguity and allowing projection of information into appropriate space(s), and the combination would perform with a reasonable expectation of success (Stahlberg [6, 11, 33, 34, 46, 65, 66, 41, 52]).
Goyal and Stahlberg does not specifically teach making the transformer-based machine learning model a multi-exit model
However Kouris teaches making the ... machine learning model a multi-exit model (Kouris [67-69] model can be a multi-exit model, multi-exit strategy allows for faster training (approach results in no compromise in accuracy)).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Kouris of making the ... machine learning model a multi-exit model, into the invention suggested by Goyal and Stahlberg; since both inventions are directed towards generating an output based on predictions from multiple layers, and incorporating the teaching of Kouris into the invention suggested by Goyal and Stahlberg would provide the added advantage of faster training,
and the combination would perform with a reasonable expectation of success
(Kouris [67-69]).
Regarding claim(s) 7, Goyal, Stahlberg and Kouris teach(es) the invention as claimed in claim(s) 6 above.
Claim 6 further teaches linear amateur head and linear expert head.
Goyal further teaches wherein the ... amateur head is configured to map an output of the at least one of the intermediate layers to a first next-token probability distribution, and wherein the ... expert head is configured to map an output of the final output layer to a second next-token probability distribution (Goyal [22, 23, 36] hidden and final layer outputs may be associated with first and next token probability distributions respectively, Goyal [0025] gating mechanism may give more weight (expert) to some layer heads than others (amateur)).
Regarding claim(s) 8, Goyal, Stahlberg and Kouris teach(es) the invention as claimed in claim(s) 7 above.
Goyal further teaches wherein the autocontrastive decoding module is further configured to contrast the first next-token probability distribution with the second next-token probability distribution to provide a token-level probability distribution as the final output from the transformer-based machine learning model (Goyal [19, 22, 23] token level probability distributions are weighted differently for different layers, provides different inference strategies including mixture of weights for different probabilities, to provide diverse summaries (contrastive), Goyal [22, 23, 36] hidden and final layer outputs may be associated with first and next token probability distributions respectively);
Regarding claim(s) 10, Goyal and Stahlberg teach(es) the invention as claimed in claim(s) 9 above.
Goyal further teaches wherein at least one of the one or more intermediate layers comprises a ... amateur head, and wherein the final layer comprises a ... expert head (Goyal [0025] gating mechanism may give more weight (expert) to some layer heads than others (amateur));
Goyal does not specifically teach linear amateur head, linear expert head, making the transformer-based machine learning model a multi-exit model.
However Stahlberg teaches linear heads (Stahlberg [6, 11, 33, 34, 46, 65, 66] receives prediction probabilities from multiple layers, output is formed using the probabilities of different layers using strategies for better handling of ambiguity, Stahlberg [41, 52] output(s) may be formed using linear heads to allow projection of information into appropriate space(s)).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Stahlberg of linear heads, into the invention suggested by Goyal,
so that the amateur head and expert head of Goyal are linear heads,
resulting in wherein at least one of the intermediate layers comprises a linear amateur head, and wherein the final layer comprises a linear expert head;
since both inventions are directed towards forming an output using the probabilities of different layers using strategies, and incorporating the teaching of Stahlberg into the invention suggested by Goyal would provide the added advantage of better handling of ambiguity and allowing projection of information into appropriate space(s), and the combination would perform with a reasonable expectation of success (Stahlberg [6, 11, 33, 34, 46, 65, 66, 41, 52]).
Goyal and Stahlberg does not specifically teach making the transformer-based machine learning model a multi-exit model
However Kouris teaches making the ... machine learning model a multi-exit model (Kouris [67-69] model can be a multi-exit model, multi-exit strategy allows for faster training (approach results in no compromise in accuracy)).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Kouris of making the ... machine learning model a multi-exit model, into the invention suggested by Goyal and Stahlberg; since both inventions are directed towards generating an output based on predictions from multiple layers, and incorporating the teaching of Kouris into the invention suggested by Goyal and Stahlberg would provide the added advantage of faster training,
and the combination would perform with a reasonable expectation of success
(Kouris [67-69]).
Regarding claim(s) 11, Goyal, Stahlberg and Kouris teach(es) the invention as claimed in claim(s) 10 above.
Claim 10 further teaches linear amateur head and linear expert head.
Goyal further teaches wherein the ... amateur head is configured to map an output of the at least one of the intermediate layers to a first next-token probability distribution, and wherein the ... expert head is configured to map an output of the final layer to a second next-token probability distribution (Goyal [22, 23, 36] hidden and final layer outputs may be associated with first and next token probability distributions respectively, Goyal [0025] gating mechanism may give more weight (expert) to some layer heads than others (amateur)).
Regarding claim(s) 12, Goyal, Stahlberg and Kouris teach(es) the invention as claimed in claim(s) 11 above.
Goyal further teaches wherein the autocontrastive decoding module is further configured to contrast the first next-token probability distribution with the second next-token probability distribution to provide a token-level probability distribution for the transformer-based machine learning model (Goyal [19, 22, 23] token level probability distributions are weighted differently for different layers, provides different inference strategies including mixture of weights for different probabilities, to provide diverse summaries (contrastive), Goyal [22, 23, 36] hidden and final layer outputs may be associated with first and next token probability distributions respectively).
Regarding claim(s) 18, Goyal teach(es) the invention as claimed in claim(s) 17 above.
Goyal does not specifically teach adding a linear head to at least one of the intermediate layers to make the transformer- based machine learning model a multi-exit model.
However Stahlberg teaches adding linear heads (Stahlberg [6, 11, 33, 34, 46, 65, 66] receives prediction probabilities from multiple layers, output is formed using the probabilities of different layers using strategies for better handling of ambiguity, Stahlberg [41, 52] output(s) may be formed using linear heads to allow projection of information into appropriate space(s)).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Stahlberg of adding a linear head to at least one of the intermediate layers, into the invention suggested by Goyal;since both inventions are directed towards forming an output using the probabilities of different layers using strategies, and incorporating the teaching of Stahlberg into the invention suggested by Goyal would provide the added advantage of better handling of ambiguity and allowing projection of information into appropriate space(s), and the combination would perform with a reasonable expectation of success (Stahlberg [6, 11, 33, 34, 46, 65, 66, 41, 52]).
Goyal and Stahlberg does not specifically teach make the transformer- based machine learning model a multi-exit model
However Kouris teaches make the ...machine learning model a multi-exit model (Kouris [67-69] model can be a multi-exit model, multi-exit strategy allows for faster training (approach results in no compromise in accuracy)).
It would have been obvious to one of an ordinary skill in the art before the effective filing date of the claimed invention, to have incorporated the concept taught by Kouris of make the ... machine learning model a multi-exit model, into the invention suggested by Goyal and Stahlberg; since both inventions are directed towards generating an output based on predictions from multiple layers, and incorporating the teaching of Kouris into the invention suggested by Goyal and Stahlberg would provide the added advantage of faster training,
and the combination would perform with a reasonable expectation of success
(Kouris [67-69]).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SANCHITA ROY whose telephone number is (571)272-5310. The examiner can normally be reached Monday-Friday 12-8.
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, Usmaan Saeed can be reached at (571) 272-4046. 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.
SANCHITA ROY
Primary Examiner
Art Unit 2146
/SANCHITA ROY/Primary Examiner, Art Unit 2146
Prosecution Insights