DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 11/03/2025 has been entered.
Status of Claims
The present application is being examined under the claims filed 11/03/2025.
Claims 1-5, 8-9, and 13 are pending.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 07/07/2025 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
The information disclosure statement filed 11/03/2025 fails to comply with 37 CFR 1.98(a)(3)(i) because it does not include a concise explanation of the relevance, as it is presently understood by the individual designated in 37 CFR 1.56(c) most knowledgeable about the content of the information, of each reference listed that is not in the English language. It has been placed in the application file, but the information referred to therein has not been considered. Particularly, NPL documents 2 and 9 have not been considered.
Response to Amendment
This Office Action is in response to Applicant’s communication filed 11/03/2025 in response to office action mailed 07/08/2025. The Applicant’s remarks and any amendments to the claims or specification have been considered with the results that follow.
Response to Arguments
Regarding Objections and informalities
In Remarks page 5, Argument 1
(Examiner summarizes Applicant’s argument) Applicant argues that claim amendments obviate objections to the claims.
Examiner’s response to Argument 1
Examiner agrees that the previous claim objections have been obviated by Applicant’s amendments. However, a new objection is issued in light of the new amendments (see below).
Regarding 35 U.S.C. 112(b) Rejections
In Remarks page 5, Argument 2
(Examiner summarizes Applicant’s argument) Applicant argues that the claim amendments obviate rejections under 35 U.S.C. 112(b).
Examiner’s response to Argument 2
Examiner agrees that the previous 112(b) rejections have been obviated by Applicant’s amendments. However, a new rejection is necessitated by applicant’s amendments.
Regarding 35 U.S.C. 101
In Remarks page 5-13, Argument 3
(Examiner summarizes Applicant’s arguments) Applicant makes various arguments that the claims are eligible under 35 U.S.C. 101.
Examiner’s response to Argument 3
While the claims still appear to recite mental process limitations, Applicant’s arguments regarding steps 2A prong 2 and 2B are convincing and thus the claims are subject-matter eligible. 35 U.S.C. 101 rejections are withdrawn accordingly.
Regarding 35 U.S.C. 103
In Remarks page 13, Argument 4
(Examiner summarizes Applicant’s arguments) Applicant argues that Hettinger does not teach “in response to detecting, by the timer, that the first time period has elapsed, setting the second artificial intelligence unit as the currently dominant artificial intelligence entity” and instead describes training layers of a machine learning model one at a time. Applicant further argues that Yan, and particularly Pfeil, nor any of the other cited references do not cure these deficiencies.
Examiner’s response to Argument 4
While Yan and Pfeil do not appear to teach the limitation, examiner maintains that Hettinger does teach using a timer to set the second artificial intelligence unit as the currently dominant artificial intelligence unit. Indeed, Hettinger teaches training a machine learning one layer at a time. In the rejection, the first iteration including the input layer, the first hidden layer, and the output layer is mapped to the first artificial intelligence unit. This ML model is trained for 100 epochs (on a 100-epoch timer), after which a new model is trained – a larger model containing the input layer, the first hidden layer (frozen), a new second hidden layer, and an output layer (mapped to the second artificial intelligence unit). After the 100-epoch timer is elapsed, the currently dominant model which produces outputs for the system shifts from the one hidden layer model to the two hidden layer model.
Moreover, examiner provides a new reference to teach controlling an actuator (see rejections under 35 U.S.C. 103 below for a complete analysis). Therefore, the claims are not patentable under 35 U.S.C. 103.
Claim Objections
Regarding Claim 1
Claim 1 and its dependents are objected to because of the following informalities: “setting the first artificial intelligence unit is set as a currently dominant artificial intelligence entity” should read “setting the first artificial intelligence unit . Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-5, 8-9, and 13 are rejected under 35 U.S.C. 112(b) as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor regards as the invention.
Regarding Claim 1
Claim 1 recites the limitation “wherein the input value comprises measured value detected by the sensor and the second artificial intelligence unit is larger than the first artificial intelligence unit”. It is unclear what aspect of the second artificial intelligence unit is larger than the first artificial intelligence unit. The specification recites:
(page 25) The second neural network 520 may then have a significantly larger and/or more complex classification system. For example, this memory 522 or the underlying classification may also be hierarchically structured in multiple levels 524, as shown in Figure 5. The total number m of classes K1, K2,...., Km of the second network 520 may be very large, in particular significantly larger than the number n of classes used by the first neural network 510.
(page 26) The hierarchical and, compared to the first network, significantly larger memory of the second network then allows a precise analysis of the input values, in the example mentioned a detailed classification into the class "dog", the respective breed, behavioral characteristics that indicate danger or a harmless situation, and others."
It is not clear whether, for example, an artificial intelligence unit with a smaller memory and a larger hierarchical structure would be larger than an artificial intelligence unit with a larger memory and a smaller hierarchical structure.
Regarding Dependent Claims
Claims 2-5, 8-9, and 13 are dependent upon claim 1, and are therefore similarly rejected for including the deficiencies of claim 1.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-5 are rejected under 35 U.S.C. 103 as being unpatentable over NPL reference Yan et al. “HD-CNN: Hierarchical Deep Convolutional Neural Networks for Large Scale Visual Recognition” herein referred to as Yan in view of NPL reference Hettinger et al. “Forward Thinking: Building and Training Neural Networks One Layer at a Time” herein referred to as Hettinger and Perumallapalli “Autonomous Vehicles: Real-Time AI for Safer Transportation Networks” herein referred to as Perumallapalli.
Regarding Claim 1
Yan teaches:
A method comprising: […], by a controller of a machine comprising a first artificial intelligence unit, a second artificial intelligence unit, a sensor
(page 2741 column 2 last line) “performs end-to-end classification, as illustrated in Fig 1(b). It mainly comprises four parts: (i) shared layers, (ii) a single component B to handle coarse categories[*Examiner notes: mapped to first AI unit], (iii) multiple components {Fk}Kk=1 (one for each group) for fine classification[*Examiner notes: mapped to second AI unit], and (iv) a single probabilistic averaging layer.”
PNG
media_image1.png
246
955
media_image1.png
Greyscale
inputting, by […] an input value to the first artificial intelligence unit and the second artificial intelligence unit, wherein the input value comprises measured value detected by the sensor
(page 2741 column 1 paragraph 2) “Shared layers (left of Fig 1 (b)) receive raw image pixels as input[*Examiner notes: inputting input values to at least a first AI unit and a second AI unit] and extract low-level features. The configuration of shared layers is set to be the same as the preceding layers in the building block CNN. On the top of Fig 1(b) are independent layers of coarse category component B, which reuses the configuration of rear layers from the building block CNN and produces an intermediate fine prediction {Bfij}Cj=1 for an image xi[*Examiner notes: measured values detected by one or more sensors].”; (page 2745 column 2 section 7.3) “The ILSVRC-2012 ImageNet dataset consists of about 1.2 million training images,”; [*Examiner notes: ImageNet dataset contains images taken by cameras, a type of sensor]
and the second artificial intelligence unit is larger than the first artificial intelligence unit
(page 2741 column 2 last line) “performs end-to-end classification, as illustrated in Fig 1(b). It mainly comprises four parts: (i) shared layers, (ii) a single component B to handle coarse categories[*Examiner notes: mapped to first AI unit], (iii) multiple components {Fk}Kk=1 (one for each group) for fine classification[*Examiner notes: mapped to second AI unit], and (iv) a single probabilistic averaging layer.”; [*Examiner notes: The second AI unit is larger than the first AI unit because it has more components]
obtaining, […], a first output value from the first artificial intelligence unit;
(page 2741 column 1 paragraph 2) “To produce a coarse category prediction {Bik}Kk=1, we append a fine-to-coarse aggregation layer (not shown in Fig 1(b)), which reduces fine predictions into coarse using a mapping[*Examiner notes: mapped to obtaining first output values of the first AI unit] P : [1, C] → [1, K].”
forming a modulation function based on the first output value of the first artificial intelligence unit;
(page 2742 column 1 paragraph 3 line 8) “The coarse category probabilities serve two purposes. First, they are used as weights for combining the predictions[*Examiner notes: modulation function] made by fine category components {Fk}Kk=1.”; (page 2742 column 1 last paragraph) “On the right side of Fig 1 (b) is the probabilistic averaging layer[*Examiner notes: mapped to modulation function], which receives fine as well as coarse category predictions and produces a final prediction based on weighted average [Equation 1] where Bik is the probability of coarse category k for image xi predicted by the coarse category component B[*Examiner notes: output of the first artificial intelligence unit].”
applying the modulation function to one or more parameters of the second artificial intelligence unit, wherein the one or more parameters influence a processing of the input value by the second artificial intelligence unit;
(page 2742 column 1 last paragraph) “On the right side of Fig 1 (b) is the probabilistic averaging layer[*Examiner notes: applying the formed modulation functions to one or more parameters], which receives fine as well as coarse category predictions and produces a final prediction based on weighted average[*Examiner notes: obtaining second output values of the second AI unit]”
obtaining, at a second time subsequent to the first time period, a second output value from of the second artificial intelligence unit; and
(page 2742 column 1 last paragraph) “On the right side of Fig 1 (b) is the probabilistic averaging layer[*Examiner notes: applying the formed modulation functions to one or more parameters], which receives fine as well as coarse category predictions and produces a final prediction based on weighted average[*Examiner notes: obtaining second output values of the second AI unit]”; [*Examiner notes: Obtaining the second output occurs at a subsequent time to the first time because the second output depends on the first output via the weighted average]
Yan does not explicitly teach:
a timer, and an actuator
setting the first artificial intelligence unit is set as a currently dominant artificial intelligence entity;
obtaining, at a first time within a first time period defined by the timer, a first output value
providing a first direct or indirect control signal to the actuator based on the currently dominant artificial intelligence entity and the first output value;
in response to detecting, by the timer, that the first time period has elapsed, setting the second artificial intelligence unit as the currently dominant artificial intelligence entity;
providing a second direct or indirect control signal to the actuator based on the currently dominant artificial intelligence entity and the second output value
However, Hettinger teaches:
a timer
[*Examiner notes: The broadest reasonable interpretation of the term “time period” includes a number of iterations (e.g. a number of epochs in a neural network or the number of times an algorithm is performed) and a and “timer” includes a counter which counts how many iterations/epochs have passed up to a fixed number]; (page 6 last paragraph) “We trained each network (forward thinking and backpropagation) for 100 epochs (complete passes through the data.)”
setting the first artificial intelligence unit is set as a currently dominant artificial intelligence entity;
(page 7 first paragraph) “To train this using forward thinking we first train 256 3x3 convolutions along with a fully-connected layer[*Examiner notes: setting the first artificial intelligence unit as currently dominant] of 150 ReLU neurons (FC 150) and a final 10-class softmax layer (Softmax 10).”
obtaining, at a first time within a first time period defined by the timer, a first output value
(page 7 paragraph 1) “To train this using forward thinking we first train 256 3x3 convolutions along with a fully-connected layer of 150 ReLU neurons (FC 150) and a final 10-class softmax layer (Softmax 10). For the second iteration, we begin by pushing the data through the 256 convolutions to create a new synthetic dataset[*Examiner notes: output at first time within a first time period defined by the timer]”.; [*Examiner notes: During machine learning training, an output is generated every time an input data is passed through the network, and particularly at the end of training. Examiner interprets “within” as inclusive of the last moment (iteration) of the timer]
in response to detecting, by the timer, that the first time period has elapsed, setting the second artificial intelligence unit as the currently dominant artificial intelligence entity;
(page 7 paragraph 1) “For the second iteration, we begin by pushing the data through the 256 convolutions to create a new synthetic dataset. Using this transformed data, we train an identical network[*Examiner notes: setting the second artificial intelligence unit as the currently dominant artificial intelligence entity]: 256 3x3 convolutions followed by FC 150 and Softmax 10. As before, we push the data through the newly learned filters.”
PNG
media_image2.png
337
533
media_image2.png
Greyscale
Yan Hettinger, and the instant application are analogous because they are all directed to artificial intelligence and neural networks.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the present invention to modify the artificial intelligence units of Yan in with the timer measuring the elapse of a predetermined time period taught by Hettinger because (Hettinger page 7 below figure 4) “Our forward thinking neural network trained with a rate of 24 sec per epoch. Traditional backprop took 53 sec per epoch” and (Hettinger page 7 Figure 4 caption) “A comparison of the training and test accuracy per epoch of a convolutional neural network trained using forward thinking (thicker, green) and traditional backpropagation (thinner, blue).” That is, the method proposed by Hettinger has better performance than traditional neural networks, and using the timer allows for comparison between different methods.
And Perumallapalli teaches:
and an actuator
(page 64 section 3.5) “The control system uses control theory to convert the planned path into commands that can be used by the vehicle's actuators (braking, steering, acceleration).”
providing a first direct or indirect control signal to the actuator based on the currently dominant artificial intelligence entity and the first output value;
(page 64 section 3.5) “The control system uses control theory to convert the planned path[*Examiner notes: based on the dominant artificial intelligence entity and the first output value] into commands that can be used by the vehicle's actuators (braking, steering, acceleration).”
providing a second direct or indirect control signal to the actuator based on the currently dominant artificial intelligence entity and the second output value.
(page 64 section 3.5) “The control system uses control theory to convert the planned path[*Examiner notes: based on the dominant artificial intelligence entity and the second output value] into commands that can be used by the vehicle's actuators (braking, steering, acceleration).”
Yan, Hettinger, Perumallapalli, and the instant application are analogous because they are all directed to artificial intelligence and neural networks.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the present invention to modify the artificial intelligence units of Yan in view of Hettinger with the actuators of Perumallapalli because (Perumallapalli page 1 abstract) “Using real-time AI algorithms, AVs can predict and respond to unpredictable traffic events and road conditions, resulting in safer transportation networks. The results of the study demonstrate the potential of AI-driven solutions to raise vehicle safety and provide the framework for fully autonomous, green transportation systems.”
Regarding Claim 2
Yan in view of Hettinger and Perumallapalli teaches:
The method of claim 1
(see rejection of claim 1)
Yan further teaches:
wherein the first artificial intelligence unit comprises a neural network having a plurality of nodes
(page 2740 abstract line 8) “However, existing deep convolutional neural networks (CNN) are trained as flat N-way classifiers […] In this paper, we introduce hierarchical deep CNNs (HD-CNNs)[*Examiner notes: mapped to neural network] by embedding deep CNNs into a two-level category hierarchy”; (page 2741 column 2 last line) “It mainly comprises four parts: (i) shared layers, (ii) a single component B to handle coarse categories, (iii) multiple components {Fk}K k=1 (one for each group) for fine classification[*Examiner notes: mapped to plurality of nodes], and (iv) a single probabilistic averaging layer.”
PNG
media_image3.png
242
825
media_image3.png
Greyscale
and wherein the one or more parameters is at least one of: a weighting (wi) for a first node of the neural network, an activation function (fakt) of a second node of the neural network, an output function (fout) of a third node of the neural network, or a propagation function of a fourth node of the neural network.
(page 2742 column 1 last paragraph) “On the right side of Fig 1 (b) is the probabilistic averaging layer, which receives fine as well as coarse category predictions and produces a final prediction[*Examiner notes: output function] based on weighted average[*Examiner notes: weighting] [Equation 1] where Bik is the probability of coarse category k for image xi predicted by the coarse category component B.”; Equation 1
Regarding Claim 3
Yan in view of Hettinger and Perumallapalli teaches:
The method according to claim 1
(see rejection of claim 1)
Yan further teaches:
wherein a classification memory is associated with the first artificial intelligence unit
(page 2745 column 1 “Parameter compression”) “As fine category CNNs have independent layers from conv2 to cccp6, we compress them and reduce the memory footprint from 447MB to 286MB[*Examiner notes: classification memory associated with each of the artificial intelligence units]”
wherein the first artificial intelligence unit performs a classification of the input value into one or more classes (Ki,K2, ..., K., Km) which are stored in the classification memory, the one or more classes each being structured in one or more dependent levels
(page 2741 column 2 section 3.1) “The dataset consists of a set of pairs {xi, yi}, where xi is an image and yi its category label. C denotes the number of fine categories, which will be automatically grouped into K coarse categories[*Examiner notes: one or more classes being structured in one or more dependent levels].”
and a number of first classes (n) in the classification memory is less than a number of second classes (m) in a second classification memory of the second artificial intelligence unit.
(page 2742 column 1) “(ii) a single component B[*Examiner notes: first AI unit] to handle coarse categories, (iii) multiple components {Fk}Kk=1 (one for each group)[*Examiner notes: second AI unit] for fine classification”; (page 2741) “The dataset consists of a set of pairs {xi, yi}, where xi is an image and yi its category label. C denotes the number of fine categories, which will be automatically grouped into K coarse categories.”; [*Examiner notes: The coarse component classifies input data into K coarse categories (1 level) and the fine classifier classifies input data into C fine categories based on the K coarse categories (2 categories, see figure 1(a) below). C is divided into K categories (i.e. C is a multiple of K) so C is larger than K.]
PNG
media_image4.png
169
192
media_image4.png
Greyscale
Regarding Claim 4
Yan in view of Hettinger and Perumallapalli teaches:
The method according to claim 1,
(see rejection of claim 1)
And Yan further teaches:
wherein applying the modulation function causes a time-dependent superposition of parameters of the second artificial intelligence unit
[*Examiner notes: The superposition principle is a mathematical and scientific principle stating that the cumulative effect is equal to a summation of the effects. Therefore, a weighted summation of the fine classifiers can be interpreted as a superposition.]; (page 2743 column 2 last paragraph) “At test time[*Examiner notes: mapped to time-dependent], for a given image, it is not necessary to evaluate all fine category classifiers, as most of them have insignificant weights Bik, […] Those fine category classifiers with Bik = 0 are not evaluated.”; (page 2742 column 1 last paragraph) “On the right side of Fig 1 (b) is the probabilistic averaging layer, which receives fine as well as coarse category predictions and produces a final prediction based on weighted average[*Examiner notes: mapped to superposition] [equation 1] where Bik is the probability of coarse category k”
PNG
media_image5.png
59
344
media_image5.png
Greyscale
and wherein the modulation function (fmoa f, fmoaw) comprises one of: a periodic function, a step function, a function with briefly increased amplitudes, a damped oscillation function, a beat function as a superposition of several periodic functions, a continuously increasing function, or a continuously decreasing function.
(page 2744 column 1 paragraph 1) “Conditional execution of top relevant fine components can accelerate the HD-CNN classification. Therefore, we threshold Bik using a parametric variable Bt = (βK)−1 and reset Bik to zero when Bik < Bt[*Examiner notes: mapped to step function]. Those fine category classifiers with Bik = 0 are not evaluated.”
Regarding Claim 5
Yan in view of Hettinger and Perumallapalli teaches:
The method of claim 1
(see rejection of claim 1)
Yan further teaches:
wherein the second artificial intelligence unit comprises a second neural network having a plurality of nodes
(page 2741 column 1 paragraph 2) “Third, we make the HD-CNN scalable by compressing the layer parameters and conditionally executing the fine category classifiers.”; (2743 column 1 last paragraph) “Fine category components {Fk}Kk=1[*Examiner notes: second neural network] can be independently pretrained in parallel […] . For each Fk, we initialize all the rear layers[*Examiner notes: mapped to plurality of nodes] except the last convolutional layer by copying the learned parameters from the pretrained model Fp.”; [*Examiner notes: Yan teaches on a second artificial intelligence unit having a plurality of fine category components which are convolutional neural networks. Each of these neural networks have a plurality of layers, which are nodes in the neural network.]
and wherein applying the one or more modulation function causes deactivation of at least a portion of the plurality of nodes.
(page 2742 column 1 paragraph 3 line 11) “The coarse category probabilities[*Examiner notes: applying modulation function] serve two purposes. First, they are used as weights for combining the predictions made by fine category components {Fk}Kk=1. Second, when thresholded, they enable conditional execution of fine category components whose corresponding coarse probabilities are sufficiently large.”; (page 2743 column 2 last paragraph) “At test time, for a given image, it is not necessary to evaluate all fine category classifiers, as most of them have insignificant weights Bik, as in Eqn 1. Their contributions to the final prediction are negligible. Conditional execution of top relevant fine components can accelerate the HD-CNN classification. Therefore, we threshold Bik using a parametric variable Bt = (βK)−1 and reset Bik to zero when Bik < Bt. Those fine category classifiers with Bik = 0 are not evaluated[*Examiner notes: applying modulation function causes deactivation of at least a portion of the nodes]”
Claims 8-9 are rejected under 35 U.S.C. 103 as being unpatentable over Yan in view of Hettinger and Perumallapalli, and further in view of Pfeil et al. (PGPUB no. US20220019874A1) herein referred to as Pfeil.
Regarding Claim 8
Yan in view of Hettinger and Perumallapalli teaches:
The method according to claim 1
(see rejection of claim 1)
Yan in view of Hettinger and Perumallapalli does not explicitly teach:
further comprising determining a comparison between the input value and a previous input value and, if the comparison results in a deviation that is above a predetermined input threshold, setting the first artificial intelligence unit to the currently dominant artificial intelligence entity
However, Pfeil teaches:
further comprising determining a comparison between the input value and a previous input value and, if the comparison results in a deviation that is above a predetermined input threshold, setting the first artificial intelligence unit to the currently dominant artificial intelligence entity
[*Examiner notes: Specification page 2 recites “The input values or features can
also be considered as layers.” Specification page 4 further recites “An important application of neural networks is the classification of input data or inputs into certain categories or classes, i.e. the recognition of correlations and assignments.”; Therefore, the broadest reasonable interpretation of “input value” includes representations of inputs processed by neural networks. The output variables of Pfeil are representations of input variables and thus may reasonably interpreted as input variables in view of Applicant’s specification.]; (paragraph [0075]) “Preferably, the criterion therefore characterizes a threshold value, for example a confidence or variance of the output variable (y) and/or the resource contingent. In addition or alternatively, the criterion can characterize a change in the respectively outputted output variables of the different paths[*Examiner notes: different paths mapped to previous and current]. If, for example, the output variable[*Examiner notes: mapped to input] differs by less than 5%,[*Examiner notes: mapped to threshold] the criterion is met[*Examiner notes: first AI unit is the dominant unit].”; (paragraph [0082]) “If the criterion is met in step S35, there follows step S36. In step S36, for example, an at least partly autonomous robot can be controlled as a function of the output variable of the deep neural network.”
Yan, Hettinger, Perumallapalli, Pfeil, and the instant application are analogous because they are all directed to artificial intelligence and neural networks.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the present invention to modify the artificial intelligence units of Yan with the determining of a currently dominating unit taught by Pfeil because (Pfeil paragraph [0004]) “Because mobile terminal devices usually have a limited energy budget, it is desirable for as little energy as possible to be consumed during operation of deep neural networks on these terminal devices. The advantage of the method according to the present invention is that the deep neural networks can be carried out at least partly successively in order to save energy while nonetheless providing accurate results. A further advantage of the method according to the present invention is that a deep neural network that is executed successively requires less memory than a plurality of deep neural networks that are optimized for respective (energy or time) budgets. An advantage of the at least partly successive execution of the deep neural network is a low latency of input signals to output signals. The first “coarse” result is then refined through the addition of further paths.”
Regarding Claim 9
Yan in view of Hettinger and Perumallapalli teaches:
The method of claim 1
(see rejection of claim 1)
Yan in view of Hettinger and Perumallapalli does not explicitly teach:
further comprising determining a comparison between the first output value of the first artificial intelligence unit and a previous first output value of the first artificial intelligence unit, and, if the comparison results in a deviation that is above a predetermined output threshold, setting the first artificial intelligence unit to the currently dominant artificial intelligence entity.
However, Pfeil teaches:
further comprising determining a comparison between the first output value of the first artificial intelligence unit and a previous first output value of the first artificial intelligence unit, and, if the comparison results in a deviation that is above a predetermined output threshold, setting the first artificial intelligence unit to the currently dominant artificial intelligence entity.
(paragraph [0075]) “Preferably, the criterion therefore characterizes a threshold value, for example a confidence or variance of the output variable (y) and/or the resource contingent. In addition or alternatively, the criterion can characterize a change in the respectively outputted output variables of the different paths[*Examiner notes: different paths mapped to previous and current]. If, for example, the output variable[*Examiner notes: mapped to output values] differs by less than 5%,[*Examiner notes: mapped to threshold] the criterion is met[*Examiner notes: first AI unit is the dominant unit].”; (paragraph [0082]) “If the criterion is met in step S35, there follows step S36. In step S36, for example, an at least partly autonomous robot can be controlled as a function of the output variable of the deep neural network.”
Claim 13 is rejected under 35 U.S.C. 103 as being unpatentable over Yan in view of Hettinger and Perumallapalli, and further in view of Han et al. (US 20190050736 A1) herein referred to as Han.
Regarding Claim 13
Yan in view of Hettinger and Perumallapalli teaches:
The method according to claim 1
(see rejection of claim 1)
Yan in view of Hettinger and Perumallapalli does not explicitly teach:
wherein the input values (Xi) further comprise at least one of data detected by a user interface, data retrieved from a memory, data received via a communication interface, and data output by a computing unit.
However, Han teaches:
wherein the input value further comprises at least one of data detected by a user interface, data retrieved from a memory, data received via a communication interface, or data output by a computing unit.
(paragraph [0007]) “Another example aspect of the present disclosure provides an example method for data pruning at a maxout layer of a neural network. The example method may include retrieving, by a load/store unit, input data from a storage module, wherein the input data is formatted as a three-dimensional vector that includes one or more feature values stored in a feature dimension of the three-dimensional vector”
Yan, Hettinger, Perumallapalli, Han, and the instant application are analogous because they are all directed to machine learning.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the present invention to modify the artificial intelligence units of Yan in view of Hettinger and Perumallapalli with the storage taught by Han because (Han paragraph [0023]) “For example, the data conversion unit 108 may be configured to prioritize the reading/writing of the feature values by adjusting the read/write sequence. That is, feature values in the feature dimension of the input data may be read from or written into the register unit 104 or other storage components prior to the reading or writing of data in other dimensions of the input data.” That is, a storage can be used for reading and writing of data.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Ezra J Baker whose telephone number is (703)756-1087. The examiner can normally be reached Monday - Friday 10:00 am - 8:00 pm ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, David Yi can be reached at (571) 270-7519. 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.
/E.J.B./Examiner, Art Unit 2126
/VAN C MANG/Primary Examiner, Art Unit 2126