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 .
Information Disclosure Statement
The IDS(s) has/have been considered and placed of record in file.
CLAIM INTERPRETATION
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) is/are: “a computing device to” in claim(s) 19.
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claim 1-20 rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
STEP 1- STATUTORY CATEGORY
Claims 1 is directed to a process (method) and claims 10 and 19 are directed to systems, which are statutory categories under 35 U.S.C. § 101. However, further analysis is required to
determine whether the claims are directed to a judicial exception.
STEP 2A, PRONG ONE - JUDICIAL EXCEPTION
Claims 1, 10 and 19 recite judicial exceptions including mathematical concepts, mental processes, and abstract ideas for organizing information. Specifically, the claims recites Hebbian and anti-Hebbian learning rules (weight matrix calculations, connection weight strengthening when nodes are co-active, connection weight weakening when representation nodes are co-active), which are mathematical algorithms and relationships. The claims also recite mental processes of classifying object representations and generating classification data, which are concepts that can be performed in the human mind through observation, evaluation, and categorization. The overall process of organizing and classifying visual data using mathematical learning rules constitutes an abstract idea for organizing information.
STEP 2A, PRONG TWO - PRACTICAL APPLICATION
The additional elements in claims 1, 10 and 19 do not integrate the judicial exception into a practical application. The claims recites only generic computer components ("one or more processing devices," "non-transitory computer readable storage devices," "computing device") that perform conventional functions at a high level of generality. The steps of receiving input and generating output amount to insignificant extra-solution activity (data gathering and outputting results). The limitation to image processing is merely a field of use that does not integrate the exception into a practical application.
Most critically, the claim fails to recite an improvement to computer functionality . The alleged improvements described in the specification – computational efficiency, continuous learning capability, and robustness to corrupted inputs-are merely natural byproducts of using particular mathematical learning rules (Hebbian instead of backpropagation). These efficiency gains stem from the mathematical properties themselves, not from any technological innovation in computer systems. This is analogous to claiming an improvement because a faster sorting algorithm is used; the computer processes faster due to different mathematics, not due to any advancement in computer technology. This distinguishes the claims from USPTO Examples 47 (Anomaly Detection) and 48 (Speech Separation), which improve signal processing output quality, whereas these claims merely improve computational efficiency as an intrinsic property of the chosen mathematical algorithm.
STEP 2B - INVENTIVE CONCEPT
Even assuming arguendo, the claims were considered directed to a practical application, Claims 1, 10 and 19 lacks an inventive concept that amounts to significantly more than the judicial exception. Hebbian learning is a well-known mathematical concept dating back to Donald Hebb's 1949 work. Lateral inhibition and multi-layer neural networks are conventional in the art. The claimed tri-layer architecture with specific connectivity patterns merely implements these known mathematical principles using generic computer components in a straightforward manner. The combination of well-understood, routine, and conventional elements does not provide significantly more than the abstract idea itself.
DEPENDENT CLAIMS
Claims 2-9, 11-17, and 20 depend on Claim 1, 10 and 19 and do not add meaningful limitations that integrate the abstract idea into a practical application or provide an inventive concept. Claims 2, 11, and 20 recites additional mathematical concepts ( eigenvectors, variance-covariance matrix). Claims 3 and 12's continuous weight updating is the natural result of the claimed mathematical algorithm. Claims 4 and 13 recites the well-known stochastic gradient descent method, a conventional mathematical optimization technique. Claims 5, 6, 14, and 16 specify mathematical properties and constraints (nonnegative values, symmetrical connection weights, continuous updating). Claims 7 and 16 merely apply the abstract idea to a particular use (identification data). Claims 8, 9, 17 and 18 recite implementation parameters of the mathematical algorithm (global inhibitory input, step size between 0 and 1). None of these dependent claims add elements that amount to significantly more than the judicial
exception. Therefore, claims 1-20 are rejected under 35 U.S.C. § 101.
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.
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.
Claims 1-3, 5-7, 10-12, 14-16, and 19-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Bahroun et al. (Online Representation Learning with Single and Multi-layer Hebbian Networks for Image Classification – hereinafter “Bahroun1”) in view of Bahroun et al. (Building Efficient Deep Hebbian Networks for Image Classification Tasks – hereinafter “Bahroun2”).
Claims 1, 10, and 19.
Bahroun1 disclsoes a system for classifying object representations from images derived from inputs, the system comprising:
one or more processing devices (p. 4: “computer”); and
one or more non-transitory computer readable storage devices storing computing instructions configured to be executed on the one or more processing devices and cause the one or more processing devices to execute functions (p. 4: “less computational time and memory”) comprising:
receiving, at a computing device, an input comprising pixelized information (p. 1: “Such an algorithm can take one image”); and
generating, at the computing device, classification data for one or more objects in the input using a tri-layer neural network (p. 3, Fig. 1 discloses a neural network with three layers) comprising: i) an input layer comprising input nodes (Fig. 1, Input layer); ii) a representation layer comprising representation nodes (Fig. 1, Hidden layer, p. 2: “The output matrix Y of encodings is an element of that corresponds to a sparse overcomplete representation of the input”); and iii) a classification layer comprising classification nodes (Fig. 1, Output layer, p. 3: “A multi-class SVM classifies the pictures using output vectors obtained by a simple pooling of the feature vectors, Y*, obtained for the input images from the trained network.”);
wherein: all input nodes are connected to all representation nodes through a first set of weighted connections having differing values and all representation nodes are connected to all other representation nodes through a second set of weighted connections having differing values (p. 3: “WT (green arrows) and MT(blue arrows) can be interpreted respectively as feed
forward synaptic connections between the input and the hidden layer and lateral synaptic
inhibitory connections within the hidden layer.”);
a weight matrix stores connection weights corresponding to the first set of weighted connections between the input nodes of the input layer and the representation nodes of the representation layer (p. 3, Fig. 1, WT), wherein:
a connection weight stored in the weight matrix is strengthened when an input node and a representation node are both active (p. 1: “the Hebbian learning principle which states that connections between two units, e.g., neurons, are strengthened when they are simultaneously activated.”);
in response to detecting that two representation nodes are co-active, the connection weights between input nodes to both representation nodes are weakened (p. 1: “inhibitory connections, which extend the capabilities of such learning rules beyond simple extraction of the principal component of input data”);
the representation nodes of the representation layer are connected to the representation classification nodes of the classification layer in a one-to-one excitatory manner and to the input nodes of the input layer (Fig. 6 of the instant application (left) and Fig. 1 of the prior art (right) show the a similar network topology with one less input node
PNG
media_image1.png
353
875
media_image1.png
Greyscale
);
the classification nodes of (p. 4, 3.1 Multi-layer Hebbian/anti-Hebbian Neural Network: “In the proposed approach, layers of Hebbian/anti-Hebbian network are stacked”; p. 4, 3.2 & p. 8: “deep networks.” Where, deep neural networks are a type of artificial neural network having multiple hidden layers between the input and output layers. A depiction of the proposed multi-layer deep network modification described in § 3.1 is shown here:
PNG
media_image2.png
337
672
media_image2.png
Greyscale
);
the input nodes of the input layer receive a first set of values corresponding to the pixelized information of the input (p. 1: “Such an algorithm can take one image”; p. 3 discloses “p. 1: “Such an algorithm can take one image”);
a second set of values for the representation nodes in the representation layer is calculated based, at least in part, on inputs received via (i) the first set of weighted connections between the input nodes and the representation nodes (Fig. 1, feed-forward connections from input layer) and (ii) the second set of weighted connections among the representation nodes (Fig. 1, lateral synaptic inhibitory connections within the hidden layer);
a third set of values for the classification nodes in the classification layer is calculated based, at least in part, on inputs received by the classification nodes from the input nodes (Fig. 1 shows the direct connections (straight lines) from the input nodes of the input layer to the second hidden layer.), the representation nodes (The modified Fig. 1 showing the disclosed (§ 3.1) deep Hebbian/anti-Hebbian network shows the direct connections (straight lines) from the hidden layer made up of hidden nodes (i.e. representation nodes) to the second hidden layer.) and other classification nodes (The modified Fig. 1 showing the disclosed (§ 3.1) deep Hebbian/anti-Hebbian network with lateral synaptic inhibitory connections within the second hidden layer); and
the classification data for the one or more objects in the input is generated based, at least in part, on the third set of values (p. 5: “The effectiveness of the algorithm is assessed by measuring the performance on an image classification task.”).
Bahroun1 discloses deep networks which inherently have more than one hidden layer; however, Bahroun1 does not specifically teach “the classification layer.” However, Bahroun2 in the same field of endeavor teaches the classification layer (p. 4: Fig. 1, DPL “Depth Pooling Layer (DPL)”).
Therefore, it would have been obvious to one of ordinary skill in the art to combine Bahroun1 and Bahroun2 before the effective filing date of the claimed invention. The motivation for this combination of references would have been to:
Enhanced Feature Learning and Accuracy: Bahroun1 demonstrates that increasing the number of output neurons in a single-layer network improves classification performance. Bahroun2 further confirms this, showing that classification accuracy increases with the number of layers and when features from multiple layers are combined in a Deep Hebbian Network (DHN). “The different layers can be stacked in various manners to construct a variety of DHNs.” (Bahroun2, p. 4). This provides a clear motivation to build a multi-layer system to achieve higher classification accuracy and learn more discriminative representations of image data.
Overcoming Limitations of Simpler Models: While single-layer Hebbian networks are effective, multi-layer architectures offer the potential for learning more complex, hierarchical features. The goal of building an efficient deep Hebbian network that combines overcomplete sparse coding and dimensionality reduction within one architecture, as presented in Bahroun2, would motivate a POSITA to stack layers and adapt learning mechanisms to effectively process image inputs for classification.
Therefore, the motivation for combining these references is to create a robust, high-performing, and biologically inspired multi-layer system that overcomes the limitations of single-layer models while retaining their efficiencies for online, unsupervised image classification.
Claims 2, 11, and 20.
The combination of Bahroun1 and Bahroun2 discloses the system of claim 1, wherein the first set of connection weights associated with the weighted connections are initially calculated using estimates of the eigenvectors of the variance-covariance matrix based on an input matrix created from vector representations of a selected set of inputs (Bahroun2 p. 6: “Prior to feeding the DHN, basic preprocessing is performed on the inputs, namely brightness and contrast normalization, and whitening”; Bahroun1 p. 5: “Although there exist Hebbian networks that can perform online whitening [10], an offline technique based on singular value decomposition [2] is applied in these experiments.” It is well-known in the art that whitening and Principal Component Analysis (PCA) methods used to decorrelate data and are fundamentally based on computing eigenvectors of covariance matrices. See also, at least p. 1251 of reference [9] of Bahroun2 attached to this office action.).
Claims 3 and 12.
The combination of Bahroun1 and Bahroun2 discloses the system of claim 1, wherein a learning mechanism continuously updates the first set of connection weights as additional inputs are processed by the tri-layer neural network (Both Bahroun1 and Bahroun2 focus on "online unsupervised learning" and describe the "continuous and local update dynamics of Hebbian learning". See at least Bahroun1, pp. 1-2).
Claims 5 and 14.
The combination of Bahroun1 and Bahroun2 discloses the system of claim 1, wherein the third set of values for the classification nodes in the classification layer, the second set of values for the representation nodes in the representation layer, and the first set of values for the input nodes in the input layer are all non-negative values (Bahroun1 p. 1: “The rule implemented is derived from a nonnegative classical multi-dimensional scaling cost-function, and is applied to both single and multi-layer architectures.”).
Claims 6 and 15.
The combination of Bahroun1 and Bahroun2 discloses the system of The system of claim 1, wherein: a second set of connection weights for the second set of weighted connections is determined such that the connection weights between any two representation nodes in the representation layer are the same in both directions; and
the second set of connection weights for the second set of weighted connections is continuously updated based, at least in part, on changes in the first set of connection weights (Bahroun 1 and Bahroun2 teach the system of claim 1, wherein the second set of connection
weights for the second set of weighted connections is determined such that the connection
weights between any two representation nodes in the representation layer are the same in both directions, and are continuously updated based, at least in part, on changes in the first set of connection weights. Both references describe "competitive learning" and "inhibitory
connections" that extend the capabilities of Hebbian/anti-Hebbian rules for sparse coding. Furthermore, both Bahroun1 and Bahroun2 describe "continuous and local update dynamics of Hebbian learning" and "online unsupervised learning" where the learning mechanism continuously updates weights as new inputs are processed.).
Claims 7 and 16.
The combination of Bahroun1 and Bahroun2 discloses the system of claim 1, wherein the classification data comprises identification data related to at least one object in the inputs (Both Bahroun1 and Bahroun2 explicitly focus on "image classification" using datasets like CIFAR-10, which consists of images of various objects.).
Claims 4, 9, 13, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Bahroun1 in view of Bahroun2 as applied to claim 1, 10, and 19 above, and further in view of Tavanaei et al. (Representation Learning using Event-based STDP – hereinafter “Tavanaei”).
Claims 4 and 13.
The combination of Bahroun1 and Bahroun2 discloses the system of claim 3.
Bahroun1 and Bahroun2 discloses all of the subject matter as described above except for specifically teaching “wherein the learning mechanism includes a stochastic gradient descent method.” However, Tavanaei in the same field of endeavor teaches wherein the learning mechanism includes a stochastic (Tavanaei p. 2: “stochastic neurons”) gradient descent method (Tavanaei p. 6: “gradient descent”).
Therefore, it would have been obvious to one of ordinary skill in the art to combine Bahroun1, Bahroun2, and Tavanaei before the effective filing date of the claimed invention. Bahroun2 discloses that the "solutions of Eq. 3 [ for online learning] … found in using coordinate descent." Coordinate descent is an iterative optimization algorithm used to find the minimum of a function, which adjusts one parameter at a time. It would have been obvious to a person of ordinary skill in the art to implement a stochastic gradient descent method as a learning mechanism for such a system. Stochastic gradient descent is a well-known and widely applied optimization technique in neural networks for minimizing cost functions, especially in online learning settings where data is processed sequentially. Given the goal of minimizing a cost function and the online nature of the learning, a POSITA would have been motivated to use stochastic gradient descent as an alternative or complementary optimization method to coordinate descent for updating network weights, thereby providing a robust and efficient way to continuously adapt the system as additional inputs are processed
Claims 9 and 18.
The combination of Bahroun1, Bahroun2, and Tavanaei teaches the system of claim 4, wherein the stochastic gradient descent method uses a step with a step size between 0 and 1 (Tavanaei p. 6: “The learning rates for STDP learning, a, (Eq. 18) and for threshold adjustment, b, (Eq. 25) were set to 0.0005 and 0.0001, respectively”; It would be obvious to use a step size (also known as a learning rate) between 0 and 1. Learning rates for gradient-based optimization algorithms are conventionally positive values, typically set between 0 and 1 to control the magnitude of weight updates and ensure stable convergence. For example, other biologically plausible learning rules, such as STDP, use learning rates like 0.0005 and 0.0001.).
Claims 8 and 17 are rejected under 35 U.S.C. 103 as being unpatentable Bahroun1 in view of Bahroun2 as applied to claim 1, 10, and 19 above, and further in view of de Silva et al. (A survey of adaptive resonance theory neural network models for engineering applications – hereinafter “Silva”).
Claims 8 and 17.
The combination of Bahroun1 and Bahroun2 teaches system of claim 1, wherein the classification nodes of the classification layer (p. 5: “The effectiveness of the algorithm is assessed by measuring the performance on an image classification task.”).
Bahroun1 and Bahroun2 discloses all of the subject matter as described above except for specifically teaching “receive a global inhibitory input that is utilized to limit spurious activities in the classification layer.” However, Silva in the same field of endeavor teaches receive a global inhibitory input that is utilized to limit spurious activities in the classification layer (Adaptive Resonance Theory (ART) models, recognized in the prior art for bioinspired
Networks. ART models employ lateral inhibition (only one category active at a time) – p. 16: “the current resonant category J is inhibited (lateral reset).” This lateral inhibition is within the classification layer (F2 p. 2) and utilize a "vigilance parameter" (p. 2, Fig. 1
ρ
) that provides a global inhibitory signal to recruit a new category if an active category does not sufficiently match the input.).
Therefore, it would have been obvious to one of ordinary skill in the art to combine Bahroun1, Bahroun2, and Silva before the effective filing date of the claimed invention. Fig. 1 shows this "global reset (inhibitory) signal" serves to stabilize learning and enforce competition within the classification layer. A person of ordinary skill in the art would have found it obvious to apply such a global inhibitory input to the classification layer in a Hebbian-based competitive learning system, to enhance classification by limiting unwanted or redundant activations and ensuring a clear winner, thereby limiting spurious activities
Conclusion
The prior art made of record but not relied, yet considered pertinent to the applicant’s disclosure, is listed on the PTO-892 form.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Ross Varndell whose telephone number is (571)270-1922. The examiner can normally be reached M-F, 9-5 EST.
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, O’Neal Mistry can be reached at (313)446-4912. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see https://ppair-my.uspto.gov/pair/PrivatePair. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Ross Varndell/Primary Examiner, Art Unit 2674