DETAILED ACTION
This office action is in response to communication filed on August 11, 2026.
Response to Amendment
Amendments filed on August 11, 2026 have been entered.
Claims 1-4, 6, 10-13, 15 and 19-20 have been amended.
Claims 5, 8-9, 14 and 17-18 have been canceled.
Claims 1-4, 6-7, 10-13, 15-16 and 19-20 have been examined.
Response to Arguments
Applicant’s arguments, see Remarks (p. 10-11), filed on 08/11/2026, with respect to the objections to the claims have been fully considered. In view of the amendments to the claims addressing the informalities raised in the previous office action, the objections to the claims have been withdrawn. However, upon further consideration, new objections to the claims are presented to address additional informalities introduced by the amendments.
Applicant’s arguments, see Remarks (p. 11-18), filed on 08/11/2026, with respect to the rejection of the claims under 35 U.S.C. 101 have been fully considered but are not persuasive.
Applicant argues (p. 14-15) that The Office Action asserts that the claims are not integrated into a practical application under Step 2A of its § 101 analysis because, in part, the additional elements represent insignificant, extra-solution activities and generic computing elements. But this conclusion does not apply to amended claims 1, 10, and 19 since these claims have been amended to recite features at a much lower level of generality than the original claims and integrate the features into a practical application … Applicant submits that no reasonable interpretation of amended independent claims 1, 10, and 19 would characterize these claims, especially when considered as an ordered combination of the recited steps/elements, as containing features that merely represent insignificant, extra- solution activities and/or generic computing elements. Particularly, these newly claimed details incorporate the appending of an MLP to a vision transformer encoder, such that the claims recite the construction of a particularly recited technological element for performing various operations. For example, the newly recited vision transformer encoder with the MLP generates image-patch-sized time intervals to convert non-overlapping patches to tokens by linear projection and also generates latent feature vectors. Thus, these claimed features enable simultaneous learning in both the time and frequency domain so as to improve forecasting accuracy.
These arguments are not persuasive.
The examiner submits that the claimed invention, when considered as a whole, collects data, analyzes the data using a series of mental and/or mathematical steps, and outputs the results of the data collection and analysis (see rejection below), which under the current guidance:
“Even if the judicial exception is narrow (e.g., a particular mathematical formula or detailed mental process), the Court has held that a claim may not preempt that judicial exception” (see 2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence, “III. Update on Certain Areas of the USPTO’s Patent Subject Matter Eligibility Guidance Applicable to AI Inventions”, section “A. Evaluation of Whether a Claim Is Directed to a Judicial Exception (Step 2A)”);
“For data, mere “manipulation of basic mathematical constructs [i.e.,] the paradigmatic ‘abstract idea,’” has not been deemed a transformation. CyberSource v. Retail Decisions, 654 F.3d 1366, 1372 n.2, 99 USPQ2d 1690, 1695 n.2 (Fed. Cir. 2011) (quoting In re Warmerdam, 33 F.3d 1354, 1355, 1360, 31 USPQ2d 1754, 1755, 1759 (Fed. Cir. 1994))” (see MPEP 2106.05(c)); and
“Examples of limitations that the courts have described as merely indicating a field of use or technological environment in which to apply a judicial exception include: … vi. Limiting the abstract idea of collecting information, analyzing it, and displaying certain results of the collection and analysis to data related to the electric power grid, because limiting application of the abstract idea to power-grid monitoring is simply an attempt to limit the use of the abstract idea to a particular technological environment, Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1354, 119 USPQ2d 1739, 1742 (Fed. Cir. 2016)” (see MPEP 2106.05(h)).
Regarding the argument about the newly claimed details incorporating the appending of an MLP to a vision transformer encoder, such that the claims recite the construction of a particularly recited technological element for performing various operations, the examiner submits that, under the broadest reasonable interpretation in light of the specification, the claimed invention applies machine learning techniques (i.e., appending of an MLP to a vision transformer encoder, see also Tuncer reference) on a generic computer (i.e., claim 1 recites “the method being implemented by at least one processor”; see Fig. 1-3 and specification at [0039]-[0052], [0054]-[0075]) to analyze the collected data and obtain a result (see specification at [0010]-[0018]), which under the current guidance corresponds to mere instructions to apply an exception (see MPEP 2106.05(f)).
Applicant also argues (p. 16) that these newly claimed details are inextricably tied to and solve problems in current computer technology. Conventional systems use line plots as visual representations of time series, which misses crucial information required for understanding time-series, as described in paragraph [0082] of the instant specification. In contrast, these newly claimed features for using a vision transformer encoder appended with an MLP allow the system to use time-frequency spectrograms as visual representations of time-series data and leverage vision transformers to achieve simultaneous learning in both the time and frequency domains, enabling enhanced predictions across various domains and diverse datasets, as well as the ability to learn cross-modality information across time and frequency domains to enhance the forecasting process, as described in paragraphs [0084] and [0085]. Thus, these claimed features enhance the ability of a computer-based system to perform accurate forecasting.
These arguments are not persuasive.
The examiner submits that manipulating data (images) using particular machine learning algorithms does not improve the functioning of a computer or any other technology or technical field, but instead and as indicated above, appends mere instructions to apply the judicial exception in a generic computer (see MPEP 2106.05(a)). The examiner also submits that the prior art of record describes the use of spectrograms as well as the recited machine learning methods applied to images for forecasting purposes (see references of record).
In addition, the examiner submits that as indicated in the MPEP ““claiming the improved speed or efficiency inherent with applying the abstract idea on a computer” does not provide an inventive concept. Intellectual Ventures I LLC v. Capital One Bank (USA), 792 F.3d 1363, 1367, 115 USPQ2d 1636, 1639 (Fed. Cir. 2015)” (see MPEP 2106.05(f)).
Furthermore, Applicant argues (p. 16-17) that the above-described improvements made by the recited features integrate the claims into a practical application and the § 101 rejection should be withdrawn for reasons similar to those set forth by the Appeals Review Panel (ARP) in its decision on request for rehearing in Ex parte Desjardins et al. In that decision the panel held that “claims directed to an improvement in the functioning of a computer, an improvement to other technology or technical field are patent eligible.” In that case, the panel held that because the claimed system used less storage capacity and enabled a reduction in system complexity, the claims reflected sufficient improvement so as to be integrated into a practical application. See the ARP’s decision on request for rehearing in Ex parte Desjardins et al. (Appeal No. 2024-000567, dated September 26, 2025) at pages 8 and 9. Similarly, the claimed features recite appending an MLP component to vision transformer encoder for performing a series of generating and converting operations that improve the accuracy, efficiency, and performance of time-series forecasting technology.
These arguments are not persuasive.
The examiner submits that the claimed invention does not improve the functioning of a computer or any other technology or technical field in view of the Appeals Review Panel decision in Ex Parte Desjardins because, contrary to the ARP decision that found the claimed invention to improve the performance of the machine learning model itself (see Advance notice of change to the MPEP in light of Ex Parte Desjardins, p. 2), the claimed invention uses existing machine learning techniques, as evidenced by applicant’s disclosure (see specification at [0006]-[0009] and the prior art of record) to manipulate data and obtain a result.
Moreover, applicant argues (p. 17-18) that The Office Action rejects independent claims 1, 10, and 19 for being an abstract idea because they refer to applying mathematical concepts to manipulate data. But these conclusions do not apply to amended claims 1, 10, and 19 … These features for appending, converting to an image, generating an image-patch-sized time interval, converting to a token, generating a latent feature vector, and outputting a trading decision cannot be considered mathematical calculations. Thus, amended independent claims 1, 10, and 19 cannot be classified as an abstract idea.
These arguments are not persuasive.
The examiner submits that, based on the Office guidance, the subject matter eligibility rejection pointed to the specific claim limitations that recite (i.e., sets forth or describes) the judicial exception. For example, the examiner submits that according to the October 2019 Patent Eligibility Guidance Update:
“A claim that recites a mathematical calculation will be considered as falling within the “mathematical concepts” grouping. A mathematical calculation is a mathematical operation (such as multiplication) or an act of calculating using mathematical methods to determine a variable or number, e.g., performing an arithmetic operation such as exponentiation. There is no particular word or set of words that indicates a claim recites a mathematical calculation. That is, a claim does not have to recite the word “calculating” in order to be considered a mathematical calculation. For example, a step of “determining” a variable or number using mathematical methods or “performing” a mathematical operation may also be considered mathematical calculations when the broadest reasonable interpretation of the claim in light of the specification encompasses a mathematical calculation” (p. 4, section “iii. “Mathematical Calculations””, par. 1); and
“claims do recite a mental process when they contain limitations that can practically be performed in the human mind, including for example, observations, evaluations, judgments, and opinions. Examples of claims that recite mental processes include: a claim to “collecting information, analyzing it, and displaying certain results of the collection and analysis,” where the data analysis steps are recited at a high level of generality such that they could practically be performed in the human mind, Electric Power Group, LLC v. Alstom, S.A. (p. 7-8).
Based on these guidelines, the examiner submits that the rejection indicated that the claimed invention recites limitations that fall under the “Mental Processes” and “Mathematical Concepts” groupings of Abstract Ideas (see rejection below).
Claim Objections
Claim 4 is objected to because of the following informalities:
Claim language should read “The method of claim 2, wherein the time-frequency spectrogram comprises rows and columns, the rows corresponding to varying wavelet frequencies and the columns corresponding to [[the]] time” in order to provide appropriate antecedence basis (e.g., the language “time” was removed from the independent claim in the current response).
Appropriate correction is required.
Claim 13 is objected to because of the following informalities:
Claim language should read “The computing apparatus of claim 11, wherein the time- frequency spectrogram comprises rows and columns, the rows corresponding to varying wavelet frequencies and the columns corresponding to [[the]] time” in order to provide appropriate antecedence basis (e.g., the language “time” was removed from the independent claim in the current response).
Appropriate correction is required.
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.
Claims 1-4, 6-7, 10-13, 15-16 and 19-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception without significantly more.
Regarding claim 1, the examiner submits that under Step 1 of the 2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence (see also 2019 Revised Patent Subject Matter Eligibility Guidance) for evaluating claims for eligibility under 35 U.S.C. 101, the claim is to a process, which is one of the statutory categories of invention.
Continuing with the analysis, under Step 2A - Prong One of the test (see italic text for abstract idea):
the limitation “converting, by the at least one processor, the numeric time-series data into an image, wherein the image is a time-frequency spectrogram” is a process that, under its broadest reasonable interpretation in light of the specification, covers performance of the limitation using mathematical concepts to transform data (e.g., transforming numeric time-series data into an image using mathematical concepts such as wavelet transform; see claim 2; see also specification at [0072], [0077], [0093]-[0094]). Except for the recitation of the extra-solution activities (e.g., source/type of data being evaluated), the particular technological environment or field of use, and the generic computer elements (i.e., at least one processor), the limitation in the context of the claim mainly refers to applying mathematical concepts to manipulate data.
the limitation “generating, by the at least one processor via the vision transformer encoder with the MLP, image-patch-sized time intervals by dividing the image into non-overlapping patches of equal size” is a process that, under its broadest reasonable interpretation in light of the specification, covers performance of the limitation using mental processes and/or mathematical concepts to manipulate data (e.g., dividing an image; see specification at [0079], [0081], [0096]). Except for the recitation of the extra-solution activities (e.g., source/type of data being evaluated), the particular technological environment or field of use, and the generic computer elements/implementation (i.e., at least one processor, vision transformer encoder with the MLP), the limitation in the context of the claim mainly refers to applying mental evaluations and/or mathematical concepts to segment an image.
the limitation “converting, by the at least one processor via the vision transformer encoder with the MLP, the non-overlapping patches to tokens by linear projection” is a process that, under its broadest reasonable interpretation in light of the specification, covers performance of the limitation using mathematical concepts to transform data (e.g., transforming a divided image based on mathematical transformations; see specification at [0079], [0081], [0096]). Except for the recitation of the extra-solution activities (e.g., source/type of data being evaluated), the particular technological environment or field of use, and the generic computer elements/implementation (i.e., at least one processor, vision transformer encoder with the MLP), the limitation in the context of the claim mainly refers to applying mathematical concepts to manipulate data.
the limitation “generating, by the at least one processor via the vision transformer encoder with the MLP, latent feature vectors by adding one-dimensional position embeddings to the tokens” is a process that, under its broadest reasonable interpretation in light of the specification, covers performance of the limitation using mathematical concepts to transform data (e.g., augmenting information to data; see specification at [0079], [0096]). Except for the recitation of the extra-solution activities (e.g., source/type of data being evaluated), the particular technological environment or field of use, and the generic computer elements/implementation (i.e., at least one processor, vision transformer encoder with the MLP), the limitation in the context of the claim mainly refers to applying mathematical concepts to manipulate data.
the limitation “forecasting, by the at least one processor based on a result of the analyzing of the image generating of the latent feature vectors, at least one future time-series data point forecasted price chart that relates to the at least one financial instrument” is a process that, under its broadest reasonable interpretation in light of the specification, covers performance of the limitation using mathematical concepts to evaluate data to obtain additional data (e.g., analyze data for prediction purposes; see specification at [0003]-[0008], [0080]-[0081]). Except for the recitation of the extra-solution activities (e.g., source/type of data being evaluated), the particular technological environment or field of use, and the generic computer elements (i.e., at least one processor), the limitation in the context of the claim mainly refers to applying mathematical concepts to manipulate data for obtaining additional data.
Therefore, the claim recites a judicial exception under Step 2A - Prong One of the test.
Furthermore, under Step 2A - Prong Two of the test, this judicial exception is not integrated into a practical application when considering the claim as a whole. In particular, the additional elements recited in the claim (see non-italic text for additional elements):
“A method for using image analysis for time-series forecasting of financial instruments, the method being implemented by at least one processor” generally links the use of the judicial exception to a particular technological environment or field of use (see MPEP 2106.05(h)) while 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 (i.e., at least one processor) as a tool to perform an abstract idea (see MPEP 2106.05(f));
“receiving, by the at least one processor, numeric time-series data that includes daily prices for at least one financial instrument over a predetermined period of time” adds extra-solution activities (e.g., mere data gathering, source/type of data to be manipulated) (see MPEP 2106.05(g)) while 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 (i.e., at least one processor) as a tool to perform an abstract idea (see MPEP 2106.05(f));
“appending, by the at least one processor, a multi-layer perceptron (MLP) to a vision transformer encoder” adds 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 (i.e., at least one processor, a multi-layer perceptron (MLP) to a vision transformer encoder) as a tool to perform an abstract idea (see MPEP 2106.05(f)); and
“outputting, by the at least one processor, a trading decision based on a result of the forecasting” adds extra-solution activities (e.g., mere data output, type of data to be output) (see MPEP 2106.05(g)) while 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 (i.e., at least one processor) as a tool to perform an abstract idea (see MPEP 2106.05(f)).
Accordingly, these additional elements, when considered individually and in combination, do not integrate the judicial exception into a practical application because they do not impose any meaningful limits on practicing the abstract idea when considering the claim as a whole. The claim is directed to a judicial exception under Step 2A of the test.
Additionally, under Step 2B of the test, the claim, when considered as a whole, does not include additional elements that, when considered individually and in combination, are sufficient to amount to significantly more than the judicial exception because the additional elements:
generally link the use of the judicial exception to a particular technological environment or field of use (i.e., using image analysis for time-series forecasting of financial instruments), which as indicated in the MPEP: “As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible “simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use.” Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application” (see MPEP 2106.05(h));
recite extra-solution activities at a high level of generality (i.e., mere data gathering/outputting by selecting a particular data source/type to be manipulated) which as indicated in the MPEP: “Another consideration when determining whether a claim integrates the judicial exception into a practical application in Step 2A Prong Two or recites significantly more in Step 2B is whether the additional elements add more than insignificant extra-solution activity to the judicial exception. The term “extra-solution activity” can be understood as activities incidental to the primary process or product that are merely a nominal or tangential addition to the claim. Extra-solution activity includes both pre-solution and post-solution activity” (see MPEP 2106.05(g)); and
append generic computer components (i.e., at least one processor) used to facilitate the application of the abstract idea (i.e., mere computer implementation using a multi-layer perceptron (MLP) to a vision transformer encoder), which as indicated in the MPEP: “Use of a computer or other machinery in its ordinary capacity for economic or other tasks (e.g., to receive, store, or transmit data) or simply adding a general purpose computer or computer components after the fact to an abstract idea (e.g., a fundamental economic practice or mathematical equation) does not provide significantly more” (see MPEP 2106.05(f)).
The claim, when considered as a whole, does not provide significantly more under Step 2B of the test.
Based on the analysis, the claim is not patent eligible.
Similarly, independent claims 10 and 19 are directed to a judicial exception (abstract idea) without significantly more as explained above with regards to claim 1.
With regards to the dependent claims they are also directed to the non-statutory subject matter because:
they just extend the abstract idea of the independent claims by additional limitations (Claims 2-3, 6-7, 11-12, 15-16 and 20), that under the broadest reasonable interpretation in light of the specification, cover performance of the limitations using mathematical concepts, and
the additional elements recited in the dependent claims, when considered individually and in combination, refer to extra-solution activities (e.g., mere data gathering using a data type or source), generic computer components and/or field of use (Claims 4 and 13), which as indicated in the Office’s guidance does not integrate the judicial exception into a practical application (Step 2A – Prong Two) and/or does not provide significantly more (Step 2B) when considering the claimed invention as a whole.
Subject Matter Not Rejected Over Prior Art
Claims 1-4, 6-7, 10-13, 15-16 and 19-20 are distinguished over the prior art of record for the following reasons:
Regarding claim 1.
Tuncer (Tuna Tuncer, Uygar Kaya, Emre Sefer, Onur Alacam, and Tugcan Hoser.
2022. Asset Price and Direction Prediction via Deep 2D Transformer and
Convolutional Neural Networks. In 3rd ACM International Conference on AI
in Finance (ICAIF ’22), November 2–4, 2022, New York, NY, USA. ACM, New
York, NY, USA, 8 pages. https://doi.org/10.1145/3533271.3561738) discloses:
A method for using image analysis for time-series forecasting of financial instruments (Abstract: a Deep Attention-based Price Prediction (DAPP) approach is employed to predict asset price using images (see p. 80, col. 1, par. 3)), the method being implemented by at least one processor (Abstract: computer methodology implies the use of at least one processor), the method comprising:
receiving, by the at least one processor, numeric time-series data that includes daily prices for at least one financial instrument over a predetermined period of time (p. 81, section “3.1 Dataset Preparation”: data for 9 Exchange-Traded Funds (ETFs) over a 20 year period was collected (see also p. 81-82, section “3.2 Labeling” regarding daily closing prices, and p. 83, section “4.1 Data Preparation” regarding daily intervals));
appending, by the at least one processor, a multi-layer perceptron (MLP) to a vision transformer encoder (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: DAPP applies a vision transformer (VT), the transformer encoder being made up of multi-head self-attention and MLP blocks, with a single hidden layer MLP being connected to the end of the transformer encoder);
converting, by the at least one processor, the numeric time-series data into an image (p. 82, section “3.3 Technical Indicators to Images”: the time-series datasets are converted into 2D images);
generating, by the at least one processor via the vision transformer encoder with the MLP, image-patch-sized time intervals by dividing the image into non-overlapping patches of equal size (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the transformer divides the image into patches, which examiner interprets to be non-overlapping patches of equal size (see details about DAPP reshaping image into an N number of patches));
converting, by the at least one processor via the vision transformer encoder with the MLP, the non-overlapping patches to tokens by linear projection (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the transformer flattens the patches with a trainable linear projection (flattened patches are interpreted as tokens in light of the specification at [0015]));
generating, by the at least one processor via the vision transformer encoder with the MLP, latent feature vectors by adding one-dimensional position embeddings to the tokens (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the transformer applies positional embedding on top of the flattened patches (generating latent feature vectors in light of the specification at [0015]));
forecasting, by the at least one processor based on a result of the generating of the latent feature vectors, at least one future time-series data point forecasted price chart that relates to the at least one financial instrument (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the embedded patches are fed into the transformer encoder structure and prediction labels regarding buy, sell or hold are made (see p. 81-82, section “3.2 Labeling” and p. 83, section “4.3 Performance Evaluation”)); and
outputting, by the at least one processor, a trading decision based on a result of the forecasting (p. 83, section “4.3 Performance Evaluation”: based on the prediction labels, decisions such as buying, selling or holding are made, which implies these decisions being output).
Sasal (Sasal, L, et al., “W-Transformers: A Wavelet-based Transformer Framework for Univariate Time Series Forecasting”, 21st IEEE International Conference on Machine Learning and Applications (ICMLA), 2022) discloses/teaches:
A method for using analysis for time-series forecasting of financial instruments (Abstract: time series forecasting based on wavelet-based transformer encoder is presented (see p. 1, col. 1, section “I. Introduction” regarding time-series forecasting applied to fields such as stock price prediction; see also Fig. 2)), the method being implemented by at least one processor (p. 5, col. 1, par. 1: analysis is performed on a GPU or CPU), the method comprising:
receiving, by the at least one processor, numeric time-series data that includes daily prices for at least one financial instrument over a predetermined period of time (Fig. 1; Fig. 2, item “Time Series (length = N)”; p. 4, col. 2, section “A. Data”: datasets are received for analysis (see p. 1, col. 1, section “I. Introduction” regarding time-series forecasting applied to fields such as stock price prediction));
appending, by the at least one processor, a transformer encoder (Fig. 2, item “Transformer Models”; p. 3, section “III. Proposed Method: W-Transformers”: captured information is passed into several independent Transformers for prediction (see also p. 2, col. 1, last par.));
converting, by the at least one processor, the numeric time-series data (Fig. 1; Fig. 2, item ‘MODWT’; p. 3, col. 1, section “III. Proposed Method: W-Transformers”, par. 1: a Maximal Overlap Discrete Wavelet Transform (MODWT) is applied to the time-series data to capture more explicit information about the time series (see also p. 2, col. 1, par. 2-3; p. 2, col. 2, section “B. MODWT algorithm”));
forecasting, by the at least one processor, at least one forecasted price chart that relates to the at least one financial instrument (Fig. 2, item “Output of decompositions”; p. 3, section “III. Proposed Method: W-Transformers”: once processed, the model is ready for forecasting (see p. 1, col. 1, section “I. Introduction” regarding time-series forecasting applied to fields such as stock price prediction)); and
outputting, by the at least one processor, a trading decision based on a result of the forecasting (p. 1, col. 1, section “I. Introduction”: time-series forecasting is applied to fields such as stock price prediction for formulating effective business strategies).
Regarding “wherein the image is a time-frequency spectrogram”, Zhao (Y. Zhao, Y. Shen, Y. Zhu and J. Yao, “Forecasting Wavelet Transformed Time Series with Attentive Neural Networks,” 2018 IEEE International Conference on Data Mining (ICDM), Singapore, 2018, pp. 1452-1457, doi: 10.1109/ICDM.2018.00201) discloses:
“The objective of this work is to exploit wavelet transformed time series towards accurate forecasting of future values. Figure 1 provides an overview of our proposed solution, which consists of two major steps: preprocessing and modeling. In the preprocessing part, we take the original univariate time series (i.e., X = {x1, · · · , xT }) as input and transform it into a scalogram as output. We adopt the Morlet wavelet transform which is known to have the optimal joint time-frequency concentration among the existing continuous wavelet transform methods … In the modeling part, we develop a novel attentive neural network for wavelet transformed time series forecasting. As the time-frequency scalogram is analogous to an image, we leverage convolutional operations [12] to extract localized time-frequency features from the scalogram” (p. 1453, col. 2, section “A. Overview”: time series data is transformed into a scalogram using wavelet transform for forecasting future values based on CNN, LSTM and MLP analysis to capture data trends (see also Fig. 1 and 2 and p. 1454, section “B. Model”)).
Du (Du, B., et al., “IMAGE PROCESSING TOOLS FOR FINANCIAL TIME SERIES
CLASSIFICATION”, 2020) discloses:
“The application of deep learning to time series forecasting is one of the major challenges in present machine learning. We propose a novel methodology that combines machine learning and image processing methods to define and predict market states with intraday financial data. A wavelet transform is applied to the log-return of stock prices for both image extraction and denoising. A convolutional neural network then extracts patterns from denoised wavelet images to classify daily time series” (Abstract: time forecasting for market states is accomplished by using wavelet transforms and a convolutional neural network (see Fig. 1)).
The closest prior art of record, taken individually or in combination, fail to teach or suggest using a time-frequency spectrogram resulting from the numeric time-series data with a vision transformer encoder with the MLP for time-series forecasting of financial instruments, in combination with all other limitations within the claim, as claimed and defined by the applicant (the examiner submits that while Tuncer discloses the use of images with a vision transformer encoder with the MLP for time-series forecasting of financial instruments, Tuncer does not disclose these images being spectrograms; and while other references describe the use of spectrograms, these references rely on CNN or other machine learning techniques (see also Tuncer, section “2. Related Work” regarding CNNs not mainly used for time-series datasets)).
Regarding claim 10.
Tuncer (Tuna Tuncer, Uygar Kaya, Emre Sefer, Onur Alacam, and Tugcan Hoser.
2022. Asset Price and Direction Prediction via Deep 2D Transformer and
Convolutional Neural Networks. In 3rd ACM International Conference on AI
in Finance (ICAIF ’22), November 2–4, 2022, New York, NY, USA. ACM, New
York, NY, USA, 8 pages. https://doi.org/10.1145/3533271.3561738) discloses:
A computing apparatus for using image analysis for time-series forecasting of financial instruments (Abstract: a Deep Attention-based Price Prediction (DAPP) approach is employed to predict asset price using images (see p. 80, col. 1, par. 3); computer methodology implies the use of a computer), the computing apparatus comprising:
a processor (Abstract: computer methodology implies the use of at least one processor);
a memory (Abstract: computer methodology implies the use of memory); and
a communication interface coupled to each of the processor and the memory (Abstract: computer methodology implies the use of communication interfaces to retrieve data from memory for processing by the at least one processor), wherein the processor is configured to:
receive, via the communication interface, numeric time-series data that includes daily prices for at least one financial instrument over a predetermined period of time (p. 81, section “3.1 Dataset Preparation”: data for 9 Exchange-Traded Funds (ETFs) over a 20 year period was collected (see also p. 81-82, section “3.2 Labeling” regarding daily closing prices, and p. 83, section “4.1 Data Preparation” regarding daily intervals));
append a multi-layer perceptron (MLP) to a vision transformer encoder (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: DAPP applies a vision transformer (VT), the transformer encoder being made up of multi-head self-attention and MLP blocks, with a single hidden layer MLP being connected to the end of the transformer encoder);
convert the numeric time-series data into an image (p. 82, section “3.3 Technical Indicators to Images”: the time-series datasets are converted into 2D images);
generate, via the vision transformer encoder with the MLP, image-patch-sized time intervals by dividing the image into non-overlapping patches of equal size (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the transformer divides the image into patches, which examiner interprets to be non-overlapping patches of equal size (see details about DAPP reshaping image into an N number of patches));
convert, via the vision transformer encoder with the MLP, the non-overlapping patches to tokens by linear projection (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the transformer flattens the patches with a trainable linear projection (flattened patches are interpreted as tokens in light of the specification at [0015]));
generate, via the vision transformer encoder with the MLP, latent feature vectors by adding one-dimensional position embeddings to the tokens (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the transformer applies positional embedding on top of the flattened patches (generating latent feature vectors in light of the specification at [0015]));
forecast, based on a result of the generating of the latent feature vectors, at least one forecasted price chart that relates to the at least one financial instrument (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the embedded patches are fed into the transformer encoder structure and prediction labels regarding buy, sell or hold are made (see p. 81-82, section “3.2 Labeling” and p. 83, section “4.3 Performance Evaluation”)); and
output a trading decision based on a result of the forecasting (p. 83, section “4.3 Performance Evaluation”: based on the prediction labels, decisions such as buying, selling or holding are made, which implies these decisions being output).
Sasal (Sasal, L, et al., “W-Transformers: A Wavelet-based Transformer Framework for Univariate Time Series Forecasting”, 21st IEEE International Conference on Machine Learning and Applications (ICMLA), 2022) discloses/teaches:
A computing apparatus (p. 5, col. 1, par. 1: analysis is performed on a GPU or CPU) for using analysis for time-series forecasting of financial instruments (Abstract: time series forecasting based on wavelet-based transformer encoder is presented (see p. 1, col. 1, section “I. Introduction” regarding time-series forecasting applied to fields such as stock price prediction; see also Fig. 2)), the computing apparatus comprising:
a processor (p. 5, col. 1, par. 1: analysis is performed on a GPU or CPU);
a memory (p. 5, col. 1, par. 1: analysis is performed on a GPU or CPU; examiner interprets these components to include memory for storing data and results during analysis); and
a communication interface coupled to each of the processor and the memory (p. 5, col. 1, par. 1: analysis is performed on a GPU or CPU; examiner interprets these components to communicate with memory for storing data and results during analysis), wherein the processor is configured to:
receive, via the communication interface, numeric time-series data that includes daily prices for at least one financial instrument over a predetermined period of time (Fig. 1; Fig. 2, item “Time Series (length = N)”; p. 4, col. 2, section “A. Data”: datasets are received for analysis (see p. 1, col. 1, section “I. Introduction” regarding time-series forecasting applied to fields such as stock price prediction));
append a transformer encoder (Fig. 2, item “Transformer Models”; p. 3, section “III. Proposed Method: W-Transformers”: captured information is passed into several independent Transformers for prediction (see also p. 2, col. 1, last par.));
convert the numeric time-series data (Fig. 1; Fig. 2, item ‘MODWT’; p. 3, col. 1, section “III. Proposed Method: W-Transformers”, par. 1: a Maximal Overlap Discrete Wavelet Transform (MODWT) is applied to the time-series data to capture more explicit information about the time series (see also p. 2, col. 1, par. 2-3; p. 2, col. 2, section “B. MODWT algorithm”));
forecast at least one forecasted price chart that relates to the at least one financial instrument (Fig. 2, item “Output of decompositions”; p. 3, section “III. Proposed Method: W-Transformers”: once processed, the model is ready for forecasting (see p. 1, col. 1, section “I. Introduction” regarding time-series forecasting applied to fields such as stock price prediction)); and
output a trading decision based on a result of the forecasting (p. 1, col. 1, section “I. Introduction”: time-series forecasting is applied to fields such as stock price prediction for formulating effective business strategies).
Regarding “wherein the image is a time- frequency spectrogram”, Zhao (Y. Zhao, Y. Shen, Y. Zhu and J. Yao, “Forecasting Wavelet Transformed Time Series with Attentive Neural Networks,” 2018 IEEE International Conference on Data Mining (ICDM), Singapore, 2018, pp. 1452-1457, doi: 10.1109/ICDM.2018.00201) discloses:
“The objective of this work is to exploit wavelet transformed time series towards accurate forecasting of future values. Figure 1 provides an overview of our proposed solution, which consists of two major steps: preprocessing and modeling. In the preprocessing part, we take the original univariate time series (i.e., X = {x1, · · · , xT }) as input and transform it into a scalogram as output. We adopt the Morlet wavelet transform which is known to have the optimal joint time-frequency concentration among the existing continuous wavelet transform methods … In the modeling part, we develop a novel attentive neural network for wavelet transformed time series forecasting. As the time-frequency scalogram is analogous to an image, we leverage convolutional operations [12] to extract localized time-frequency features from the scalogram” (p. 1453, col. 2, section “A. Overview”: time series data is transformed into a scalogram using wavelet transform for forecasting future values based on CNN, LSTM and MLP analysis to capture data trends (see also Fig. 1 and 2 and p. 1454, section “B. Model”)).
Du (Du, B., et al., “IMAGE PROCESSING TOOLS FOR FINANCIAL TIME SERIES
CLASSIFICATION”, 2020) discloses:
“The application of deep learning to time series forecasting is one of the major challenges in present machine learning. We propose a novel methodology that combines machine learning and image processing methods to define and predict market states with intraday financial data. A wavelet transform is applied to the log-return of stock prices for both image extraction and denoising. A convolutional neural network then extracts patterns from denoised wavelet images to classify daily time series” (Abstract: time forecasting for market states is accomplished by using wavelet transforms and a convolutional neural network (see Fig. 1)).
The closest prior art of record, taken individually or in combination, fail to teach or suggest using a time-frequency spectrogram resulting from the numeric time-series data with a vision transformer encoder with the MLP for time-series forecasting of financial instruments, in combination with all other limitations within the claim, as claimed and defined by the applicant (the examiner submits that while Tuncer discloses the use of images with a vision transformer encoder with the MLP for time-series forecasting of financial instruments, Tuncer does not disclose these images being spectrograms; and while other references describe the use of spectrograms, these references rely on CNN or other machine learning techniques (see also Tuncer, section “2. Related Work” regarding CNNs not mainly used for time-series datasets)).
Regarding claim 19.
Tuncer (Tuna Tuncer, Uygar Kaya, Emre Sefer, Onur Alacam, and Tugcan Hoser.
2022. Asset Price and Direction Prediction via Deep 2D Transformer and
Convolutional Neural Networks. In 3rd ACM International Conference on AI
in Finance (ICAIF ’22), November 2–4, 2022, New York, NY, USA. ACM, New
York, NY, USA, 8 pages. https://doi.org/10.1145/3533271.3561738) discloses:
A non-transitory computer readable storage medium storing instructions for using image analysis for time-series forecasting of financial instruments (Abstract: a Deep Attention-based Price Prediction (DAPP) approach is employed to predict asset price using images (see p. 80, col. 1, par. 3)), the non-transitory computer readable storage medium comprising executable code, which when executed by a processor (Abstract: computer methodology implies the use of a processor and memory), cause the processor to:
receive numeric time-series data that includes daily prices for at least one financial instrument over a predetermined period of time (p. 81, section “3.1 Dataset Preparation”: data for 9 Exchange-Traded Funds (ETFs) over a 20 year period was collected (see also p. 81-82, section “3.2 Labeling” regarding daily closing prices, and p. 83, section “4.1 Data Preparation” regarding daily intervals));
append a multi-layer perceptron (MLP) to a vision transformer encoder (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: DAPP applies a vision transformer (VT), the transformer encoder being made up of multi-head self-attention and MLP blocks, with a single hidden layer MLP being connected to the end of the transformer encoder);
convert the numeric time-series data into an image (p. 82, section “3.3 Technical Indicators to Images”: the time-series datasets are converted into 2D images);
generate, via the vision transformer encoder with the MLP, image-patch-sized time intervals by dividing the image into non-overlapping patches of equal size (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the transformer divides the image into patches, which examiner interprets to be non-overlapping patches of equal size (see details about DAPP reshaping image into an N number of patches));
convert, via the vision transformer encoder with the MLP, the non-overlapping patches to tokens by linear projection (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the transformer flattens the patches with a trainable linear projection (flattened patches are interpreted as tokens in light of the specification at [0015]));
generate, via the vision transformer encoder with the MLP, latent feature vectors by adding one-dimensional position embeddings to the tokens (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the transformer applies positional embedding on top of the flattened patches (generating latent feature vectors in light of the specification at [0015]));
forecast, based on a result of the analyzing of the image generating of the latent feature vectors, at least one future time-series data point forecasted price chart that relates to the at least one financial instrument (p. 82, section “3.4 Deep Attention-based Price Prediction DAPP”: the embedded patches are fed into the transformer encoder structure and prediction labels regarding buy, sell or hold are made (see p. 81-82, section “3.2 Labeling” and p. 83, section “4.3 Performance Evaluation”)); and
output a trading decision based on a result of the forecasting (p. 83, section “4.3 Performance Evaluation”: based on the prediction labels, decisions such as buying, selling or holding are made, which implies these decisions being output).
Sasal (Sasal, L, et al., “W-Transformers: A Wavelet-based Transformer Framework for Univariate Time Series Forecasting”, 21st IEEE International Conference on Machine Learning and Applications (ICMLA), 2022) discloses/teaches:
A non-transitory computer readable storage medium storing instructions (p. 5, col. 1, par. 1: analysis is performed on a GPU or CPU; examiner interprets these components to include memory for storing data and results during analysis as well as instructions for performing the analysis) for using analysis for time-series forecasting of financial instruments (Abstract: time series forecasting based on wavelet-based transformer encoder is presented (see p. 1, col. 1, section “I. Introduction” regarding time-series forecasting applied to fields such as stock price prediction; see also Fig. 2)), the non-transitory computer readable storage medium comprising executable code, which when executed by a processor (p. 5, col. 1, par. 1: analysis is performed on a GPU or CPU), cause the processor to:
receive numeric time-series data that includes daily prices for at least one financial instrument over a predetermined period of time (Fig. 1; Fig. 2, item “Time Series (length = N)”; p. 4, col. 2, section “A. Data”: datasets are received for analysis (see p. 1, col. 1, section “I. Introduction” regarding time-series forecasting applied to fields such as stock price prediction));
append a transformer encoder (Fig. 2, item “Transformer Models”; p. 3, section “III. Proposed Method: W-Transformers”: captured information is passed into several independent Transformers for prediction (see also p. 2, col. 1, last par.));
convert the numeric time-series data (Fig. 1; Fig. 2, item ‘MODWT’; p. 3, col. 1, section “III. Proposed Method: W-Transformers”, par. 1: a Maximal Overlap Discrete Wavelet Transform (MODWT) is applied to the time-series data to capture more explicit information about the time series (see also p. 2, col. 1, par. 2-3; p. 2, col. 2, section “B. MODWT algorithm”));
forecast at least one forecasted price chart that relates to the at least one financial instrument (Fig. 2, item “Output of decompositions”; p. 3, section “III. Proposed Method: W-Transformers”: once processed, the model is ready for forecasting (see p. 1, col. 1, section “I. Introduction” regarding time-series forecasting applied to fields such as stock price prediction)); and
output a trading decision based on a result of the forecasting (p. 1, col. 1, section “I. Introduction”: time-series forecasting is applied to fields such as stock price prediction for formulating effective business strategies).
Regarding “wherein the image is a time- frequency spectrogram”, Zhao (Y. Zhao, Y. Shen, Y. Zhu and J. Yao, “Forecasting Wavelet Transformed Time Series with Attentive Neural Networks,” 2018 IEEE International Conference on Data Mining (ICDM), Singapore, 2018, pp. 1452-1457, doi: 10.1109/ICDM.2018.00201) discloses:
“The objective of this work is to exploit wavelet transformed time series towards accurate forecasting of future values. Figure 1 provides an overview of our proposed solution, which consists of two major steps: preprocessing and modeling. In the preprocessing part, we take the original univariate time series (i.e., X = {x1, · · · , xT }) as input and transform it into a scalogram as output. We adopt the Morlet wavelet transform which is known to have the optimal joint time-frequency concentration among the existing continuous wavelet transform methods … In the modeling part, we develop a novel attentive neural network for wavelet transformed time series forecasting. As the time-frequency scalogram is analogous to an image, we leverage convolutional operations [12] to extract localized time-frequency features from the scalogram” (p. 1453, col. 2, section “A. Overview”: time series data is transformed into a scalogram using wavelet transform for forecasting future values based on CNN, LSTM and MLP analysis to capture data trends (see also Fig. 1 and 2 and p. 1454, section “B. Model”)).
Du (Du, B., et al., “IMAGE PROCESSING TOOLS FOR FINANCIAL TIME SERIES
CLASSIFICATION”, 2020) discloses:
“The application of deep learning to time series forecasting is one of the major challenges in present machine learning. We propose a novel methodology that combines machine learning and image processing methods to define and predict market states with intraday financial data. A wavelet transform is applied to the log-return of stock prices for both image extraction and denoising. A convolutional neural network then extracts patterns from denoised wavelet images to classify daily time series” (Abstract: time forecasting for market states is accomplished by using wavelet transforms and a convolutional neural network (see Fig. 1)).
The closest prior art of record, taken individually or in combination, fail to teach or suggest using a time-frequency spectrogram resulting from the numeric time-series data with a vision transformer encoder with the MLP for time-series forecasting of financial instruments, in combination with all other limitations within the claim, as claimed and defined by the applicant (the examiner submits that while Tuncer discloses the use of images with a vision transformer encoder with the MLP for time-series forecasting of financial instruments, Tuncer does not disclose these images being spectrograms; and while other references describe the use of spectrograms, these references rely on CNN or other machine learning techniques (see also Tuncer, section “2. Related Work” regarding CNNs not mainly used for time-series datasets)).
Regarding claims 2-4, 6-7, 11-13, 15-16 and 20.
They are also distinguished over the prior art of record due to their dependency.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
LU; Hanlin et al., US 20250111559 A1¸ MEDIA DATA PROCESSING
Reference discloses using transformers to process spectrogram-based features.
NGUYEN; Nam H. et al., US 20240362458 A1¸ TIME SERIES FORECASTING
Reference discloses time series forecasting using transformers.
Tolstikhin; Ilya et al., US 20220375211 A1, MULTI-LAYER PERCEPTRON-BASED COMPUTER VISION NEURAL NETWORKS
Reference discloses vision transformers being used for image analysis.
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 LINA CORDERO whose telephone number is (571)272-9969. The examiner can normally be reached 9:30 am - 6:00 pm.
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, ANDREW SCHECHTER can be reached at 571-272-2302. 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.
/LINA CORDERO/Primary Examiner, Art Unit 2857
Prosecution Insights