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 .
Information Disclosure Statement
The information disclosure statements (IDSs) submitted on December 15, 2023, and December 15, 2023 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Claim Objections
Claim 6 is objected to because of the following informalities: “a set of weight” should read “a set of weights”. Appropriate correction is required.
Claim 11 is objected to because of the following informalities: “if the clause length index is non-zero, one or more pairs of inclusion indices,” should read “if the clause length index is non-zero, encoding one or more pairs of inclusion indices,”. 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 1-11 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1,
Claim 1 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 1 is directed to an apparatus comprising one or more sensors, at least one processor, and at least one memory, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“classify the one or more sets of sensor data”
As drafted, under their broadest reasonable interpretations, cover mental processes, i.e., concepts performed in the human mind (including an observation, evaluation, judgement, opinion). The above limitations in the context of this claim correspond to mental processes, e.g., evaluation and judgement with assistance of pen and paper.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply an exception (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)).
The limitations:
“Apparatus comprising: one or more sensors; at least one processor; and at least one memory with storing instructions that, when executed by the at least one processor, cause the apparatus at to…”
“…using an encoded Tsetlin machine”
“wherein the encoded Tsetlin machine comprises a compressed representation of a trained Tsetlin machine, the compressed representation being based on a number of exclude decisions of the trained Tsetlin machine being greater than a number of include decisions of the trained Tsetlin machine”
As drafted, are additional elements that amount to no more than mere instructions to apply an exception for the abstract ideas. See MPEP 2106.05(f).
The limitations:
“collect one or more sets of sensor data using the one or more sensors”
As drafted, are additional elements that amount to no more than insignificant extra-solution activity. See MPEP 2106.05(g).
Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract ideas into a practical application, all of the additional elements are “mere instructions to apply” and “insignificant extra-solution activity”. Specifically, the collecting limitation recites the well-understood, routine, and conventional activity of receiving and transmitting data over a network. MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network). Mere instructions to apply and insignificant extra-solution activity cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 2,
Claim 2 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 2 is directed to an apparatus comprising one or more sensors, at least one processor, and at least one memory, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: See corresponding analysis of claim 1.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recited additional elements that do not apply the exception in a meaningful way (See MPEP 2106.05(e)).
The limitations:
“wherein the compressed representation comprises a sequence of N-bit blocks, each block encoding a repeating sequence of include and/or exclude decisions of the trained Tsetlin machine and comprising: a two-bit key, the two-bit key representing an include and/or exclude pattern of the trained Tsetlin machine; and an N-2 bit number, the number encoding a length of a repeating sequence of the two-bit key”
As drafted, is an additional element that does not apply an exception for the abstract ideas in a meaningful way. See MPEP 2106.05(e).
Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract ideas into a practical application, all of the additional elements do not apply the exception in a meaningful way. The claim is not patent eligible.
Regarding Claim 3,
Claim 3 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 3 is directed to an apparatus comprising one or more sensors, at least one processor, and at least one memory, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: See corresponding analysis of claim 2.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recited additional elements that do not apply the exception in a meaningful way (See MPEP 2106.05(e)).
The limitations:
“wherein the N-2 bit number encodes a number of repetitions of the 2-bit key before either: (i) the repeating 2-bit pattern changes or (ii) the maximum number of repetitions that is representable by the block is reached”
As drafted, is an additional element that does not apply an exception for the abstract ideas in a meaningful way. See MPEP 2106.05(e).
Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract ideas into a practical application, all of the additional elements do not apply the exception in a meaningful way. The claim is not patent eligible.
Regarding Claim 4,
Claim 4 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 4 is directed to an apparatus comprising one or more sensors, at least one processor, and at least one memory, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: See corresponding analysis of claim 1.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recited additional elements that do not apply the exception in a meaningful way (See MPEP 2106.05(e)).
The limitations:
“wherein the compressed representation comprises one or more blocks corresponding to the include decisions of a respective class of the trained Tsetlin machine, wherein a block comprises: a block-length index, the block-length index indicating a length of the block; and one or more sub-blocks, wherein a sub-blocks corresponds to a clause in the respective class represented by the block, the sub-block comprising: a clause length index indicating length of the sub-block; and if the clause length index is non-zero, one or more pairs of inclusion indices, each pair of inclusion indices comprising a feature index and a literal index identifying an include decision of the trained Tsetlin machine for the clause represented by the sub-block”
As drafted, is an additional element that does not apply an exception for the abstract ideas in a meaningful way. See MPEP 2106.05(e).
Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract ideas into a practical application, all of the additional elements do not apply the exception in a meaningful way. The claim is not patent eligible.
Regarding Claim 5,
Claim 5 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 5 is directed to an apparatus comprising one or more sensors, at least one processor, and at least one memory, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: See corresponding analysis of claim 4.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recited additional elements that do not apply the exception in a meaningful way (See MPEP 2106.05(e)).
The limitations:
“wherein each clause in a class of the trained Tsetlin machine is represented by a sub-block in a respective block of the encoded representation”
As drafted, is an additional element that does not apply an exception for the abstract ideas in a meaningful way. See MPEP 2106.05(e).
Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract ideas into a practical application, all of the additional elements do not apply the exception in a meaningful way. The claim is not patent eligible.
Regarding Claim 6,
Claim 6 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 6 is directed to an apparatus comprising one or more sensors, at least one processor, and at least one memory, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“estimate an amount of energy available for classification of the one or more sets of sensor data”
“determine a subset of clauses for use by the encoded Tsetlin machine for classifying the sensor data based on the estimated amount of energy and a set of weight, each weight associated with a respective clause in a set of clauses”
As drafted, under their broadest reasonable interpretations, cover mental processes, i.e., concepts performed in the human mind (including an observation, evaluation, judgement, opinion). The above limitations in the context of this claim correspond to mental processes, e.g., evaluation and judgement with assistance of pen and paper.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recited additional elements that are mere instructions to apply an exception (See MPEP 2106.05(f)).
The limitations:
“a power system for generating power from ambient energy”
As drafted, are additional elements that amount to no more than mere instructions to apply an exception for the abstract ideas. See MPEP 2106.05(f).
Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract ideas into a practical application, all of the additional elements are “mere instructions to apply”. Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 7,
Claim 7 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 7 is directed to an apparatus comprising one or more sensors, a power system for generating power from ambient energy, at least one processor, and at least one memory, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“estimate an amount of energy available for classification of the one or more sets of sensor data”
“determine a subset of clauses for use by a Tsetlin machine for classifying the sensor data based on the estimated amount of energy and an ordered list of clauses, the ordered list of clauses indicating a relative importance of each clause in the Tsetlin machine”
“classify the one or more sets of sensor data”
As drafted, under their broadest reasonable interpretations, cover mental processes, i.e., concepts performed in the human mind (including an observation, evaluation, judgement, opinion). The above limitations in the context of this claim correspond to mental processes, e.g., evaluation and judgement with assistance of pen and paper.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply an exception (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)).
The limitations:
“Apparatus comprising: one or more sensors; a power system for generating power from ambient energy; at least one processor; and at least one memory with storing instructions that, when executed by the at least one processor, cause the apparatus at least to…
“using the Tsetlin machine”
As drafted, are additional elements that amount to no more than mere instructions to apply an exception for the abstract ideas. See MPEP 2106.05(f).
The limitations:
“collect one or more sets of sensor data using the one or more sensors”
As drafted, are additional elements that amount to no more than insignificant extra-solution activity. See MPEP 2106.05(g).
Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract ideas into a practical application, all of the additional elements are “mere instructions to apply” and “insignificant extra-solution activity”. Specifically, the collecting limitation recites the well-understood, routine, and conventional activity of receiving and transmitting data over a network. MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network). Mere instructions to apply and insignificant extra-solution activity cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 8,
Claim 8 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 8 is directed to an apparatus comprising one or more sensors, a power system for generating power from ambient energy, at least one processor, and at least one memory, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“determine an off-time of the apparatus”
“compare the off-time to an off-threshold”
“if the off-time is greater than the off-threshold, drop one or more clauses from the Tsetlin machine based on a position of the one or more clauses on the ordered list”
“if the off-time is less than the off-threshold, add one or more previously-dropped clauses to the Tsetlin machine based on a position of the one or more clauses on the ordered list”
As drafted, under their broadest reasonable interpretations, cover mental processes, i.e., concepts performed in the human mind (including an observation, evaluation, judgement, opinion). The above limitations in the context of this claim correspond to mental processes, e.g., evaluation and judgement with assistance of pen and paper.
Step 2A Prong Two Analysis: See corresponding analysis of claim 7.
Step 2B Analysis: See corresponding analysis of claim 7.
Regarding Claim 9,
Claim 9 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 9 is directed to a method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“classifying …the training data to determine a set of test classifications”
“separating positive and negative clauses of the trained Tsetlin machine”
“generating a set of ranked positive clauses and a set of ranked negative clauses by ranking the positive clauses and negative clauses of the trained Tsetlin machine based on the set of test classifications”
“combining the set of ranked positive clauses and the set of ranked negative clauses to generate a combined ranked list of clauses, wherein the combined ranked list of clauses alternates between positive and negative clauses”
“encoding the combined ranked list of clauses using an encoding scheme based on a number of exclude decisions of the trained Tsetlin machine being greater than a number of include decisions of the trained Tsetlin machine”
As drafted, under their broadest reasonable interpretations, cover mental processes, i.e., concepts performed in the human mind (including an observation, evaluation, judgement, opinion). The above limitations in the context of this claim correspond to mental processes, e.g., evaluation and judgement with assistance of pen and paper.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recited additional elements that are mere instructions to apply an exception (See MPEP 2106.05(f)).
The limitations:
“training a Tsetlin machine on a set of training data, the set of training data comprising a plurality of sets of sensor data, each with a corresponding classification”
“…using the trained Tsetlin machine”
As drafted, are additional elements that amount to no more than mere instructions to apply an exception for the abstract ideas. See MPEP 2106.05(f).
Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract ideas into a practical application, all of the additional elements are “mere instructions to apply”. Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 10,
Claim 10 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 10 is directed to a method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“for each of a plurality of training examples in the training dataset, each training example comprising a set of sensor data and a ground truth classification: comparing the test classification of the training example to the ground truth classification of the training example”
“if the test classification of the training example matches the ground truth classification of the training example: increasing weights associated with positive clauses with a positive output in the class associated with the test classification”
“decreasing weights associated with negative clauses with a positive output in the class associated with the test classification”
“if the test classification of the training example does not match the ground truth classification of the training example: decreasing weights associated with positive clauses with a positive output in the class associated with the test classification”
“increasing weights associated with negative clauses with a positive output in the class associated with the test classification”
“ranking the positive clauses based on their respective weights”
“ranking the negative clauses based on their respective weights”
As drafted, under their broadest reasonable interpretations, cover mental processes, i.e., concepts performed in the human mind (including an observation, evaluation, judgement, opinion). The above limitations in the context of this claim correspond to mental processes, e.g., evaluation and judgement with assistance of pen and paper.
Step 2A Prong Two Analysis: See corresponding analysis of claim 9.
Step 2B Analysis: See corresponding analysis of claim 9.
Regarding Claim 11,
Claim 11 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 11 is directed to a method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“encoding include and exclude states of the Tsetlin machine into a plurality of blocks, each block comprising: a two-bit key, the two-bit key representing an include and/or exclude pattern of the trained Tsetlin machine; and an N-2 bit number, the number encoding a length of a repeating sequence of the two-bit key”
“encoding include and exclude states of the Tsetlin machine into a plurality of blocks, each block comprising: a class index, the class index indicating a length of the block; and one or more sub-blocks, wherein a sub-blocks corresponds to a clause in the respective class represented by the block, a sub-block comprising: a clause length index indicating length of the sub-block”
“if the clause length index is non-zero, one or more pairs of inclusion indices, each pair of inclusion indices comprising a feature index and a literal index identifying an include decision of the trained Tsetlin machine for the clause represented by the sub-block”
As drafted, under their broadest reasonable interpretations, cover mental processes, i.e., concepts performed in the human mind (including an observation, evaluation, judgement, opinion). The above limitations in the context of this claim correspond to mental processes, e.g., evaluation and judgement with assistance of pen and paper.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recited additional elements that do not apply the exception in a meaningful way (See MPEP 2106.05(e)).
The limitations:
“wherein the encoding scheme comprises a first encoding scheme or a second encoding scheme”
As drafted, is an additional element that does not apply an exception for the abstract ideas in a meaningful way. See MPEP 2106.05(e).
Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract ideas into a practical application, all of the additional elements do not apply the exception in a meaningful way. The claim is not patent eligible.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim 1 is rejected under 35 U.S.C. 103 as being unpatentable over Phoulady et al. (The Weighted Tsetlin Machine: Compressed Representations with Weighted Clauses) (“Phoulady”) in view of Lei et al (Low-Power Audio Keyword Spotting Using Tsetlin Machines) (“Lei”).
Regarding claim 1, Phoulady teaches an apparatus comprising: one or more sensors (Phoulady Section 4.1 MNIST Handwritten Digit Recognition “The MNIST dataset contains images of handwritten digits, 60,000 training and 10,000 test examples. Each example is a labelled 28 × 28-pixel grayscale image” Phoulady provides a dataset of images and while it would be obvious that a sensor is used, the reference does not explicitly teach a sensor.); at least one processor; and at least one memory with storing instructions that, when executed by the at least one processor, cause the apparatus at to at least (Phoulady Section 1 Introduction “Our intent is to significantly reduce memory usage and computation time by using the clause weights to compress the pattern representation of the TM.” Phoulady provides memory and computation corresponding to computing hardware.): collect one or more sets of sensor data using the one or more sensors (Phoulady Section 4.1 MNIST Handwritten Digit Recognition “The MNIST dataset contains images of handwritten digits, 60,000 training and 10,000 test examples. Each example is a labelled 28 × 28-pixel grayscale image The pixels are represented by integers between 0 (black) and 255 (white), and each label is an integer from 0 to 9. We binarize the input pixels to get 768 bits of data.” Phoulady provides an image dataset corresponding to collected sensor data.); classify the one or more sets of sensor data using an encoded Tsetlin machine (Phoulady Section 2.2 Tsetlin Machine Structure and Inference “A binary classifier over o binary features can be regarded as a Boolean function—each of the 2 opossible combinations of the o binary inputs belongs to one of the classes 0 or 1.”; Section 4.1 MNIST Handwritten Digit Recognition “Each example is a labelled 28 × 28-pixel grayscale image. The pixels are represented by integers between 0 (black) and 255 (white), and each label is an integer from 0 to 9. We binarize the input pixels to get 768 bits of data. A value less than 77 = ⌈.3 · 255⌉ is converted to 0, and a value greater than or equal to 77 is converted to 1.” Phoulady provides the Tsetlin machine classifying the image dataset and classifying the pixel values as 0 or 1 based on a threshold value of 77.), wherein the encoded Tsetlin machine comprises a compressed representation of a trained Tsetlin machine (Phoulady Section 1 Introduction “Our intent is to significantly reduce memory usage and computation time by using the clause weights to compress the pattern representation of the TM.” Phoulady provides a compressed representation of a trained Tsetlin Machine.), the compressed representation being based on a number of exclude decisions of the trained Tsetlin machine being greater than a number of include decisions of the trained Tsetlin machine (Phoulady Section 2.3 Tsetlin Machine Learning “Note that the probability ps is a hyperparameter that controls the sparsity of the clauses produced. In brief, the sparsity of clauses increases with ps, simply because a higher ps produces more exclude actions relative to include actions.” Phoulady provides the number of exclude decisions greater than the number of include decisions for the compressed representation.).
Phoulady fails to explicitly teach a sensor, as discussed above.
However, Lei teaches one or more sensors (Lei Section 4.7. KWS-TM on the Embedded System “Figure 16 shows the system diagram of our MFCC-TM pipeline on the ARM Cortex M7 microcontroller. The design is similar with the Python pipeline (Figure 4) but it is configured for inference only and uses a microphone [sensor] followed by an analog-to-digital (ADC) converter for audio pre-processing” Lei teaches a microphone corresponding to a sensor, as shown in Figure 16.) …collect one or more sets of sensor data using the one or more sensors (Lei Section 3.1 Audio Feature Extraction Using MFCC “Audio data streams are always subject to redundancies in the channel that formalize as nonvocal noise, background noise and silence [20,21]. Therefore, the challenge becomes identification and extraction of the desired linguistic content (the keyword) and maximally discarding everything else. To achieve this we must consider transformation and filtering techniques that can amplify the characteristics of the speech signals against the background information. This is often done through the generation of MFCCs as seen in the signal processing flow in Figure 3.” Lei teaches collecting audio streams using a microphone, corresponding to collecting sensor data.)
Phoulady and Lei are both considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to Tsetlin Machines. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady with the above teachings of Lei. Doing so would allow for an increase in the energy frugality of the whole system and a transition toward low-power hardware accelerators of the pipeline to tackle real-time applications (Lei Section 6 “Through these design considerations we are able to increase the energy frugality of the whole system and transition toward low-power hardware accelerators of the pipeline to tackle real-time applications.”).
Claims 4-5 are rejected under 35 U.S.C. 103 as being unpatentable over Phoulady et al. (The Weighted Tsetlin Machine: Compressed Representations with Weighted Clauses) (“Phoulady”) in view of Lei et al. (Low-Power Audio Keyword Spotting Using Tsetlin Machines) (“Lei”) in further view of Bakar et al. (Logic-based Intelligence for Batteryless Sensors) (“Bakar 1”).
Regarding claim 4, Phoulady in view of Lei teaches the apparatus of claim 1, as discussed above in the rejection of claim 1, but fails to teach wherein the compressed representation comprises one or more blocks corresponding to the include decisions of a respective class of the trained Tsetlin machine, wherein a block comprises: a block-length index, the block-length index indicating a length of the block; and one or more sub-blocks, wherein a sub-blocks corresponds to a clause in the respective class represented by the block, the sub-block comprising: a clause length index indicating length of the sub-block; and if the clause length index is non-zero, one or more pairs of inclusion indices, each pair of inclusion indices comprising a feature index and a literal index identifying an include decision of the trained Tsetlin machine for the clause represented by the sub-block.
However, Bakar 1 teaches wherein the compressed representation comprises one or more blocks corresponding to the include decisions of a respective class of the trained Tsetlin machine (Bakar Section 3.1 Encoding TM Models “Figure 4 shows the full pipeline for the embedded RLE-TM we propose [the compressed representation]. Through discretization of the actual TA states followed by tallying the runs of 1s and 0s we are able to significantly reduce the model’s memory requirements. Notice that this is a lossless compression since the original include/exclude sequence can always be recovered, resulting in zero accuracy loss at inference time. The Extracted TAs blocks [one or more blocks] show the TA states post-training, plotted in Figure 4 to visually demonstrate the ratio of include to exclude decisions. [include decisions]” Bakar provides the RLE corresponding to the compressed representation, which comprises blocks corresponding to include decisions of the trained TM, as shown in Figure 4.), wherein a block comprises: a block-length index, the block-length index indicating a length of the block (Bakar Section 3.1 Encoding TM Models “Hence we propose to use Run Length Encoding (RLE) to compress the TAs after training. With this approach, long sequences of the same value (e.g., 0) are replaced by its count [the block-length index indicating a length of the block]. The ratio of exclude to include TA decisions can be attributed to the granularity of the boolean inputs …In Figure 4 the TM shows each TA has 200 states and there are 45240 TAs altogether, only 84 TAs are include decisions for an example Key Word Spot ting (KWS) application. The substantial imbalance between the include and exclude decisions allow for very large runs of excludes separated by one include enabling significant compression ratios to be achieved through RLE encoding.”); and one or more sub-blocks, wherein a sub-blocks corresponds to a clause in the respective class represented by the block (Bakar Section 2.2 “Full TM Model. Figure 3 shows the architecture of the TM for training. The clauses are grouped together for each class with an equal number of clauses per class.” Bakar provides groups of clauses for each class, corresponding to a clause in a respective class.), the sub-block comprising: a clause length index indicating length of the sub-block (Bakar Section 4.2 Memory Usage “For TM, we use 50, 400 and 100 clauses for MNIST, CIFAR and KWS, respectively, as they offer comparable accuracy to the BNNs.” Bakar provides number of clauses in the grouped clauses, corresponding to the clause length index.); and if the clause length index is non-zero, one or more pairs of inclusion indices, each pair of inclusion indices comprising a feature index and a literal index identifying an include decision of the trained Tsetlin machine for the clause represented by the sub-block (Bakar Section 2.2 Tsetlin Machines “Figure 2 demonstrates the data preparation pipeline on the left side. The raw features (integer or floating point values) are booleanized into Boolean Features [a feature index] (1 or 0 values)… For simplicity, in our example, we chose one bit representation for each raw feature, but this is a design choice that should be made depending on the application. Boolean features are then converted to Boolean Literals [a literal index]… The clause relates the input data (Boolean Literal) to its respective learning element (TA). As shown through Figure 2, the include/exclude decision of the TA is combined with its Boolean Literal through a simple logic circuit. The output of a clause is a single bit [an include decision of the trained Tsetlin machine for the clause]. The number of clauses is a parameter the user will configure much like the number of filters or layers in a Neural Network. Typically, higher number clauses [the clause length index is non-zero] result in better accuracy as there is greater likelihood of the TM ending the right propositions.”).
Phoulady, Lei, and Bakar 1 are all considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to Tsetlin Machines. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady in view of Lei with the above teachings of Bakar 1. Doing so would reduce memory requirements and improve latency (Bakar 1 Section 3.1 “When considering intermittently-powered systems, this com pression approach not only reduces the memory requirement of a TM-based classification task but improves also its latency.”).
Regarding claim 5, Phoulady in view of Lei in further view of Bakar 1 teaches the apparatus of claim 4, wherein each clause in a class of the trained Tsetlin machine is represented by a sub-block in a respective block of the encoded representation (Bakar Section 2.2 Tsetlin Machines “Architectural Building Blocks. The right-hand side of Figure 2 shows two of the fundamental elements of the TM: the Tsetlin Automata and the Clause… Figure 3 shows the architecture of the TM for training. The clauses are grouped together [sub-blocks] for each class with an equal number of clauses per class [a clause in the respective class]. The one bit clause outputs are multiplied with a positive or negative polarity (⇥1 or ⇥−1) and summed for each class. The polarity allows each clause to learn both supporting and opposing propositions for their respective class.”; Section 3.1 Encoding TM Models “The Extracted TAs blocks [a respective block of the encoded representation] show the TA states post-training, plotted in Figure 4 to visually demonstrate the ratio of include to exclude decisions.”).
Phoulady, Lei, and Bakar 1 are all considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to Tsetlin Machines. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady in view of Lei in further view of Bakar with the above teachings of Bakar. Doing so would reduce memory requirements and improve latency (Bakar 1 Section 3.1 “When considering intermittently-powered systems, this compression approach not only reduces the memory requirement of a TM-based classification task but improves also its latency.”).
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Phoulady et al. (The Weighted Tsetlin Machine: Compressed Representations with Weighted Clauses) (“Phoulady”) in view of Lei et al. (Low-Power Audio Keyword Spotting Using Tsetlin Machines) (“Lei”) in further view of Bakar et al. (REHASH: A Flexible, Developer Focused, Heuristic Adaptation Platform for Intermittently Powered Computing) (“Bakar 2”).
Regarding claim 6, Phoulady in view of Lei teaches …wherein the instructions further cause the apparatus to: estimate an amount of energy available for classification of the one or more sets of sensor data (Phoulady Section 4.1 MNIST Handwritten Digit Recognition “However, replacing the Bernoulli process sampling with our binomial distribution based sampling approach (Algorithm 1), we are able to speed up the random feedback generation by a factor of 7, significantly improving the overall learning speed of the TM. Weighted Clauses with WTM We now turn to evaluating the effect of weighted clauses. In our first experiment we used a 10-class TM with 2,000 clauses per class as baseline. The accuracy obtained by this configuration was parred by a WTM with merely 500 clauses. This reduced memory usage 4 times, while increasing execution time by a factor of 4” Phoulady provides memory usage and execution time evaluations corresponding to an estimation of energy available for classification of the image (Sensor) data.); and determine a subset of clauses for use by the encoded Tsetlin machine for classifying the sensor data based on the estimated amount of energy and a set of weight (Phoulady Section 4.1 MNIST Handwritten Digit Recognition “Feedback Generation Speedup Employing 2,000 class clauses on MNIST produces 12.6 millions Type I feedback calls in the first epoch . In turn, each call samples n = 2·28· 28 = 1568 values from a Bernoulli process with success probability ps = .1, to assign feedback to the individual Tsetlin automata. This is the most time consuming part of TM learning. Whereas recognizing the 60,000 images takes 23.5 seconds, feedback generation using standard Bernoulli process sampling takes 61.2 seconds… Weighted Clauses with WTM We now turn to evaluating the effect of weighted clauses. In our first experiment we used a 10-class TM with 2,000 clauses per class as baseline” Phoulady provides determining to use 2,000 clauses per class, corresponding to the subset of clauses for classification based on energy estimation and including weighted clauses.), each weight associated with a respective clause in a set of clauses (Phoulady Section 3.4 WTM Learning “Weight Updating. Adjusting the Tsetlin automata’s states in the WTM is performed in the same way as for the TM. However, we introduce a novel approach for updating the weights. This updating is governed by Type I and Type II feedback. Note that, below, we only update the weights of those clauses that evaluate to 1 for the current input x.” Phoulady provides the weights associated with respective clauses.).
Phoulady in view of Lei fails to teach a power system for generating power from ambient energy.
However, Bakar 2 teaches a power system for generating power from ambient energy (Bakar 2 Section 4.1 Energy Harvesting and Storage Model “An energy harvester’s power output is not constant, but depends on the load on the harvester. This relationship between the harvester current and the harvester voltage +, is captured by an IV curve which represents the ⇢ state of possible power outputs at a point in time. Solar panel IV curves are presented in datasheets and have a distinct shape with a "knee" where the most power can be harvested.” Bakar 2 provides generating power from solar panels, wherein the solar panels generate power from ambient energy (i.e., sunlight).)
Phoulady, Lei and Bakar 2 are all considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to sensor based classification. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady in view of Lei with the above teachings of Bakar 2. Doing so would allow for a task-based adaptive runtime system that are flexible, highly tunable to application requirements, hardware independent, and that enables computation despite scarce energy (Bakar 2 Section 1.2 “These contributions make a task-based adaptive runtime system that i) is flexible and highly tunable to application requirements, ii) is hardware-independent since heuristic signals are easily gathered on commodity platforms, and iii) enables computation despite scarce energy.”).
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Phoulady et al. (The Weighted Tsetlin Machine: Compressed Representations with Weighted Clauses) (“Phoulady”) in view of Bakar et al. (REHASH: A Flexible, Developer Focused, Heuristic Adaptation Platform for Intermittently Powered Computing) (“Bakar 2”) in further view of Rahman et al. (MILEAGE: An Automated Optimal Clause Search Paradigm for Tsetlin Machines) (“Rahman”) in further view of Lei et al. (Low-Power Audio Keyword Spotting Using Tsetlin Machines) (“Lei”).
Regarding claim 7, Phoulady teaches an apparatus comprising: one or more sensors (Phoulady Section 4.1 MNIST Handwritten Digit Recognition “The MNIST dataset contains images of handwritten digits, 60,000 training and 10,000 test examples. Each example is a labelled 28 × 28-pixel grayscale image” Phoulady provides a dataset of images and while it would be obvious to have a sensor, the reference does not explicitly teach a sensor.); …at least one processor; and at least one memory with storing instructions that, when executed by the at least one processor, cause the apparatus at least to (Phoulady Section 1 Introduction “Our intent is to significantly reduce memory usage and computation time by using the clause weights to compress the pattern representation of the TM.” Phoulady provides memory and computation corresponding to computing hardware.): collect one or more sets of sensor data using the one or more sensors (Phoulady Section 4.1 MNIST Handwritten Digit Recognition “The MNIST dataset contains images of handwritten digits, 60,000 training and 10,000 test examples. Each example is a labelled 28 × 28-pixel grayscale image The pixels are represented by integers between 0 (black) and 255 (white), and each label is an integer from 0 to 9. We binarize the input pixels to get 768 bits of data.” Phoulady provides an image dataset corresponding to collected sensor data.); estimate an amount of energy available for classification of the one or more sets of sensor data (Phoulady Section 4.1 MNIST Handwritten Digit Recognition “However, replacing the Bernoulli process sampling with our binomial distribution based sampling approach (Algorithm 1), we are able to speed up the random feedback generation by a factor of 7, significantly improving the overall learning speed of the TM. Weighted Clauses with WTM We now turn to evaluating the effect of weighted clauses. In our first experiment we used a 10-class TM with 2,000 clauses per class as baseline. The accuracy obtained by this configuration was parred by a WTM with merely 500 clauses. This reduced memory usage 4 times, while increasing execution time by a factor of 4” Phoulady provides memory usage and execution time evaluations corresponding to an estimation of energy available for classification of the sensor data.); …and classify the one or more sets of sensor data using the Tsetlin machine (Phoulady Section 2.2 Tsetlin Machine Structure and Inference “A binary classifier over o binary features can be regarded as a Boolean function—each of the 2 opossible combinations of the o binary inputs belongs to one of the classes 0 or 1.”; Section 4.1 MNIST Handwritten Digit Recognition “Each example is a labelled 28 × 28-pixel grayscale image. The pixels are represented by integers between 0 (black) and 255 (white), and each label is an integer from 0 to 9. We binarize the input pixels to get 768 bits of data. A value less than 77 = ⌈.3 · 255⌉ is converted to 0, and a value greater than or equal to 77 is converted to 1.” Phoulady provides the Tsetlin machine classifying the image dataset and classifying the pixel values as 0 or 1 based on a threshold value of 77.).
Phoulady fails to teach explicitly teach one or more sensors; a power system for generating power from ambient energy …determine a subset of clauses for use by a Tsetlin machine for classifying the sensor data based on the estimated amount of energy and an ordered list of clauses, the ordered list of clauses indicating a relative importance of each clause in the Tsetlin machine.
However, Bakar 2 teaches a power system for generating power from ambient energy (Bakar 2 Section 4.1 “An energy harvester’s power output is not constant, but depends on the load on the harvester. This relationship between the harvester current and the harvester voltage +, is captured by an IV curve which represents the ⇢ state of possible power outputs at a point in time. Solar panel IV curves are presented in datasheets and have a distinct shape with a "knee" where the most power can be harvested.” Bakar 2 provides generating power from solar panels, wherein the solar panels generate power from ambient energy (i.e., sunlight).);
Phoulady and Bakar 2 are both considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to sensor classification concerning energy usage. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady with the above teachings of Bakar 2. Doing so would allow for a task-based adaptive runtime system that are flexible, highly tunable to application requirements, hardware independent, and that enables computation despite scarce energy (Bakar 2 Section 1.2 “These contributions make a task-based adaptive runtime system that i) is flexible and highly tunable to application requirements, ii) is hardware-independent since heuristic signals are easily gathered on commodity platforms, and iii) enables computation despite scarce energy.”).
Further, Rahman teaches determine a subset of clauses for use by a Tsetlin machine for classifying the sensor data based on the estimated amount of energy and an ordered list of clauses (Rahman Section 1 “The recently proposed Tsetlin Machine (TM) [5] offers a much simpler logic rather than arithmetic alternative to NN and has opened new possibilities for energy efficient logic based learning [estimated amount of energy]”; Section 2 “After the initial training, the clause weights are evaluated by performing the inference routine on the training datapoints. MILEAGE maintains a list of useful clauses and measures/ updates their weights at the end of the training cycles… If the net contribution of a clause towards correct classification is zero, it is discarded and the useful clause list is updated [ordered list of clauses]. Then, the inference routine is performed and clause weights are updated again. If the new inference accuracy is better than the previous best inference accuracy then the pruning routine is performed, again, otherwise the TM continues with normal training procedure.”), the ordered list of clauses indicating a relative importance of each clause in the Tsetlin machine (Rahman Section 1 “Each class maintains weights corresponding to the clauses that determine its importance [relative importance of each clause] and contribution towards a correct classification.”; Section 2 “After the initial training, the clause weights are evaluated by performing the inference routine on the training datapoints. MILEAGE maintains a list of useful clauses and measures/ updates their weights at the end of the training cycles…”);
Phoulady, Bakar 2, and Rahman are all considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to Tsetlin Machines. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady on view of Bakar 2 with the above teachings of Rahman. Doing so would allow for the optimal number of clauses that will be sufficient to solve a classification problem with competitive accuracy (Rahman Section 1 “This paper proposes an autoMated optImal cLause sEarch pAradiGm for standard tsEtlin machines (MILEAGE) which aims to search for the optimal number of clauses that will be sufficient to solve a classification problem with competitive accuracy.”).
Further, Lei teaches one or more sensors (Lei Section 4.7. KWS-TM on the Embedded System “Figure 16 shows the system diagram of our MFCC-TM pipeline on the ARM Cortex M7 microcontroller. The design is similar with the Python pipeline (Figure 4) but it is configured for inference only and uses a microphone [sensor] followed by an analog-to-digital (ADC) converter for audio pre-processing” Lei teaches a microphone corresponding to a sensor, as shown in Figure 16.) …collect one or more sets of sensor data using the one or more sensors…(Lei Section 3.1 Audio Feature Extraction Using MFCC “Audio data streams are always subject to redundancies in the channel that formalize as nonvocal noise, background noise and silence [20,21]. Therefore, the challenge becomes identification and extraction of the desired linguistic content (the keyword) and maximally discarding everything else. To achieve this we must consider transformation and filtering techniques that can amplify the characteristics of the speech signals against the background information. This is often done through the generation of MFCCs as seen in the signal processing flow in Figure 3.” Lei teaches collecting audio streams using a microphone, corresponding to collecting sensor data.)
Phoulady, Bakar 2, Rahman and Lei are all considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to Tsetlin Machines. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady in view of Bakar 2 in further view of Rahman with the above teachings of Lei. Doing so would allow for an increase in the energy frugality of the whole system and a transition toward low-power hardware accelerators of the pipeline to tackle real-time applications (Lei Section 6 “Through these design considerations we are able to increase the energy frugality of the whole system and transition toward low-power hardware accelerators of the pipeline to tackle real-time applications.”).
Claims 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Phoulady et al. (The Weighted Tsetlin Machine: Compressed Representations with Weighted Clauses) (“Phoulady”) in view of Rahman et al. (MILEAGE: An Automated Optimal Clause Search Paradigm for Tsetlin Machines) (“Rahman”) in further view of Bakar et al. (Logic-based Intelligence for Batteryless Sensors) (“Bakar”) in further view of Lei et al. (Low-Power Audio Keyword Spotting Using Tsetlin Machines) (“Lei”).
Regarding claim 9, Phoulady teaches a method comprising: training a Tsetlin machine on a set of training data, the set of training data comprising a plurality of sets of sensor data, (Phoulady Section 4.1 MNIST Handwritten Digit Recognition “After 130 epochs, the WTM had a steady training accuracy of 100%, whereas the TM peaked at 99.86%.The WTM also had a higher testing accuracy, improving peak accuracyfrom98.27%to98.63%.Table 3 contains test accuracy statistics of the WTM, collected from the last 50 of 300 epochs of single-run training (multiple runs behave similarly).” Phoulady provides training Tsetlin Machines with image data (Sensor data), and while it would be obvious to have a sensor, the reference does not explicitly teach a sensor.), each with a corresponding classification (Phoulady Section 4.1 MNIST Handwritten Digit Recognition “The MNIST dataset contains images of handwritten digits, 60,000 training and 10,000 test examples. Each example is a labelled 28 × 28-pixel grayscale image” Phoulady provides labeled training data (images).); classifying, using the trained Tsetlin machine, the training data to determine a set of test classifications (Phoulady Section 4.1 MNIST Handwritten Digit Recognition “Employing 2,000 class clauses on MNIST produces 12.6 millions Type I feedback calls in the first epoch . In turn, each call samples n = 2·28· 28 = 1568 values from a Bernoulli process with success probability ps = .1, to assign feedback to the individual Tsetlin automata. This is the most time consuming part of TM learning. Whereas recognizing the 60,000 images takes 23.5 seconds, feedback generation using standard Bernoulli process sampling takes 61.2 seconds. This is about 2/3 of the training time in the first epoch.” Phoulady provides image recognition during training corresponding to classifying training data to determine the test classifications.); separating positive and negative clauses of the trained Tsetlin machine (Phoulady Section 2.2 Tsetlin Machine Structure and Inference “Further, the TM groups the clauses into positive ones C+ 1 ,C+ 2 ,...,C+cP and negative ones C− 1 ,C− 2 , ...,C− cN” Phoulady provides grouping clauses into positive and negative respective groups, corresponding to separating positive and negative clauses.).
Phoulady fails to explicitly teach a plurality of sets of sensor data... generating a set of ranked positive clauses and a set of ranked negative clauses by ranking the positive clauses and negative clauses of the trained Tsetlin machine based on the set of test classifications; combining the set of ranked positive clauses and the set of ranked negative clauses to generate a combined ranked list of clauses, wherein the combined ranked list of clauses alternates between positive and negative clauses; and encoding the combined ranked list of clauses using an encoding scheme based on a number of exclude decisions of the trained Tsetlin machine being greater than a number of include decisions of the trained Tsetlin machine.
However, Rahman teaches generating a set of ranked positive clauses and a set of ranked negative clauses by ranking the positive clauses and negative clauses of the trained Tsetlin machine based on the set of test classifications (Rahman Section I Introduction “Each class maintains weights corresponding to the clauses that determine its importance and contribution towards a correct classification. …MILEAGE prunes the given model, often initialized with high number of clauses, during training to generate a compressed model with similar or better accuracy. It evaluates the clause weights depending on their contribution towards a correct classification and removes the clauses with no impact on the classification outcomes over the training sample space.”; Section II Proposed Approach “For a correct classification, the clause weight for the positive polarity clauses are incremented and vice versa for the negative polarity clauses.” Rahman teaches weight values that indicate the importance of a specific clause, wherein both positive and negative polarity clauses are included, corresponding to ranking the positive and negative clauses based on test classifications (weights).); combining the set of ranked positive clauses and the set of ranked negative clauses to generate a combined ranked list of clauses wherein the combined ranked list of clauses alternates between positive and negative clauses (Rahman Listing 1 Update Clause Weights function; Section II Proposed Approach “After the initial training, the clause weights are evaluated by performing the inference routine on the training datapoints. MILEAGE maintains a list of useful clauses [combined ranked list of clauses] and measures/ updates their weights at the end of the training cycles… If the net contribution of a clause towards correct classification is zero, it is discarded and the useful clause list is updated… For a correct classification, the clause weight for the positive polarity clauses are incremented and vice versa for the negative polarity clauses. [alternating between positive and negative]” Rahman provides creating a list of useful weighted (ranked) positive and negative clauses in accordance with Listing 1, which alternates between incrementing/decrementing positive/negative clauses based on conditional statements.);
Phoulady and Rahman are both considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to Tsetlin Machines. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady with the above teachings of Rahman. Doing so would allow for the optimal number of clauses that will be sufficient to solve a classification problem with competitive accuracy (Rahman Section 1 “This paper proposes an autoMated optImal cLause sEarch pAradiGm for standard tsEtlin machines (MILEAGE) which aims to search for the optimal number of clauses that will be sufficient to solve a classification problem with competitive accuracy.”)
Further Bakar teaches encoding the combined ranked list of clauses using an encoding scheme based on a number of exclude decisions of the trained Tsetlin machine being greater than a number of include decisions of the trained Tsetlin machine (Bakar Section 2.2 “The output of a clause is a single bit. The number of clauses is a parameter the user will configure much like the number of filters or layers in a Neural Network. Typically, higher number clauses [combined ranked list of clauses] result in better accuracy as there is greater likelihood of the TM ending the right propositions.”; Section 3.1 Encoding TM Models “However, we notice two intrinsic properties of the TM: 1) there is no need to store the actual value of each TA state but just their binary include/exclude decision, and 2) typically the number of exclude decisions far outnumber that of the include decisions [a number of exclude decisions greater than a number of include decisions]… Hence we propose to use Run Length Encoding (RLE) [encoding] to compress the TAs after training. With this approach, long sequences of the same value (e.g., 0) are replaced by its count.” Section 4.2 Memory usage “This means the compression ratio gets better when the number of clauses increases.” Bakar provides encoding a number of exclude decisions that are greater than a number of include decisions for a clause to generate a compressed representation, wherein the number of clauses of the TM is being interpretated as a combined ranked list of clauses.).
Phoulady, Rahman and Bakar are all considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to Tsetlin Machines. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady in further view of Rahman with the above teachings of Bakar. Doing so would reduce memory requirements and improve latency (Bakar Section 3.1 “When considering intermittently-powered systems, this compression approach not only reduces the memory requirement of a TM-based classification task but improves also its latency.”).
Further, Lei teaches …a plurality of sets of sensor data (Lei Section 3.1 Audio Feature Extraction Using MFCC “Audio data streams are always subject to redundancies in the channel that formalize as nonvocal noise, background noise and silence [20,21]. Therefore, the challenge becomes identification and extraction of the desired linguistic content (the keyword) and maximally discarding everything else. To achieve this we must consider transformation and filtering techniques that can amplify the characteristics of the speech signals against the background information. This is often done through the generation of MFCCs as seen in the signal processing flow in Figure 3.” Lei teaches collecting audio streams using a microphone, corresponding to collecting sensor data.)
Phoulady, Rahman, Bakar, and Lei are all considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to Tsetlin Machines. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady in view of Rahman in further view of Bakar with the above teachings of Lei. Doing so would allow for an increase in the energy frugality of the whole system and a transition toward low-power hardware accelerators of the pipeline to tackle real-time applications (Lei Section 6 “Through these design considerations we are able to increase the energy frugality of the whole system and transition toward low-power hardware accelerators of the pipeline to tackle real-time applications.”).
Regarding claim 10, Phoulady in view of Rahman in further view of Bakar and Lei teaches the method of claim 9, as discussed above in the rejection of claim 9.
Phoulady further teaches…decreasing weights associated with negative clauses with a positive output in the class associated with the test classification (Phoulady Section 2.3 Tsetlin Machine Learning “Accordingly, if the signed sum of the clauses is negative, we increase it by giving some of the positive clauses Type I feedback and some of the negative clauses Type II feedback.”; Section 3.4 WTM Learning “For Type II feedback, on the other hand, the weights are decreased instead. This is to diminish the impact of the associated clauses, thus com bating false positives.” Phoulady provides decreasing weights associated with negative clauses through Type II Feedback on the negative clauses, which decreases the weights.).
Further Rahman teaches…wherein ranking the positive clauses and negative clauses of the trained Tsetlin machine based on the set of test classifications (Rahman Section II “After the initial training, the clause weights are evaluated by performing the inference routine on the training datapoints. MILEAGE maintains a list of useful clauses and measures/ updates their weights at the end of the training cycles.” Rahman provides a useful list of clauses ranked by weight.) further comprises, for each of a plurality of training examples in the training dataset, each training example comprising a set of sensor data and a ground truth classification (Rahman Section II “The updating method for the clause weights is presented in Listing 1. The update_clause_weights() function is called for every class in the TM. If the guessed_class produced by the TM through the inference for a datapoint is the same class, i.e. the (class_identity), for which this function is called, then for this particular class the clause weights are updated. The update itself is for clauses where the output was 1 for the classification in question. For a correct classification, the clause weight for the positive polarity clauses are incremented and vice versa for the negative polarity clauses.” Rahman provides the algorithm in Listing 1, which compares training examples to correct classification, as noted by the if statement in Listing 1.): comparing the test classification of the training example to the ground truth classification of the training example (Rahman Section II; Listing 1 “if(class_identity == guessed_class){ for(int i =0;i < clauses.size(); i++){ if(clause_output[i]==1&& (guessed_class == correct_class))” Rahman provides comparing the test classification of the training example to the ground truth classification (correct_class) using the if statement in Listing 1.); if the test classification of the training example matches the ground truth classification of the training example: increasing weights associated with positive clauses with a positive output in the class associated with the test classification (Rahman Section II; Listing 1 “if(class_identity == guessed_class){ for(int i =0;i < clauses.size(); i++){ if(clause_output[i]==1&& (guessed_class == correct_class)) { // Guessed correct, // increment the +ve clauses if(i%2 == 0){ clause_weight[i] += 1; } } else if(clause_output[i]==1 && (guess_class != correct_class)){ // Guessed incorrect, increment the //-ve clauses, decrement +ve clauses” Section II “If the guessed_class produced by the TM through the inference for a datapoint is the same class, i.e. the (class_identity), for which this function is called, then for this particular class the clause weights are updated. The update itself is for clauses where the output was 1 for the classification in question. For a correct classification, the clause weight for the positive polarity clauses are incremented and vice versa for the negative polarity clauses.” Rahman teaches increasing weights associated with positive clauses.); …and if the test classification of the training example does not match the ground truth classification of the training example: decreasing weights associated with positive clauses with a positive output in the class associated with the test classification (Rahman Section II; Listing 1 “if(class_identity == guessed_class){ for(int i =0;i < clauses.size(); i++){ if(clause_output[i]==1&& (guessed_class == correct_class)) { // Guessed correct, // increment the +ve clauses if(i%2 == 0){ clause_weight[i] += 1; } } else if(clause_output[i]==1 && (guess_class != correct_class)){ // Guessed incorrect, increment the //-ve clauses, decrement +ve clauses”; Section II “If the guessed_class produced by the TM through the inference for a datapoint is the same class, i.e. the (class_identity), for which this function is called, then for this particular class the clause weights are updated. The update itself is for clauses where the output was 1 for the classification in question. For a correct classification, the clause weight for the positive polarity clauses are incremented and vice versa for the negative polarity clauses.” Rahman provides an else if statement for when the training example does not match a correct classification.); and increasing weights associated with negative clauses with a positive output in the class associated with the test classification (Rahman Section II; Listing 1 “if(class_identity == guessed_class){ for(int i =0;i < clauses.size(); i++){ if(clause_output[i]==1&& (guessed_class == correct_class)) { // Guessed correct, // increment the +ve clauses if(i%2 == 0){ clause_weight[i] += 1; } } else if(clause_output[i]==1 && (guess_class != correct_class)){ // Guessed incorrect, increment the //-ve clauses, decrement +ve clauses” Rahman teaches increasing weights associated with negative clauses in Listing 1.); ranking the positive clauses based on their respective weights and ranking the negative clauses based on their respective weights (Rahman Section II “After the initial training, the clause weights are evaluated by performing the inference routine on the training datapoints. MILEAGE maintains a list of useful clauses and measures/ updates their weights at the end of the training cycles… For a correct classification, the clause weight for the positive polarity clauses are incremented and vice versa for the negative polarity clauses. The system also decrements positive clauses that contribute to a negative classification. However, negative polarity clauses produce a 1 for a correct classification are not decremented.” Rahman provides a useful list of clauses ranked by weight including positive and negative clauses.).
Phoulady, Rahman, Bakar and Lei are all considered to be analogous to the claimed invention because they are in the same field of artificial intelligence and more specifically applied to Tsetlin Machines. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Phoulady in view of Rahman in further view of Bakar and Lei with the above teachings of Rahman. Doing so would allow for the optimal number of clauses that will be sufficient to solve a classification problem with competitive accuracy (Rahman Section 1 “This paper proposes an autoMated optImal cLause sEarch pAradiGm for standard tsEtlin machines (MILEAGE) which aims to search for the optimal number of clauses that will be sufficient to solve a classification problem with competitive accuracy.”)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KURT NICHOLAS PRESSLY whose telephone number is (703)756-4639. The examiner can normally be reached M-F 8-4.
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, Kamran Afshar can be reached at (571) 272-7796. 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.
/KURT NICHOLAS PRESSLY/Examiner, Art Unit 2125
/KAMRAN AFSHAR/Supervisory Patent Examiner, Art Unit 2125
Prosecution Insights