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 .
Response to Arguments
Applicant's arguments filed 6/22/2026 have been fully considered but they are not persuasive.
Regarding applicant’s argument for 35 USC. 103, applicant argues in page 5-6 “These rejections are respectfully traversed.
As amended, claim 1 requires that the time-varying parameter be applied to the nonlinear modulation function to perform modulation of a state variable. This defines a specific functional relationship in which the time-varying parameter directly affects the nonlinear transformation of the internal state of the reservoir. Neither Okumura nor Yue discloses or suggests such a relationship, as discussed below.
Okumura discloses a reservoir computing structure including a nonlinear element and delay elements, but does not teach applying a time-varying parameter to a nonlinear modulation function to control modulation of a state variable as claimed.
Yue, on the other hand, relates to modulation of an input signal to a physical system. The modulation in Yue is applied to external input or operating conditions, and does not involve applying a time-varying parameter to a nonlinear modulation function to directly control the internal state update of a reservoir element.
Therefore, the combination of Okumura and Yue would not result in the claimed configuration, in which the time-varying parameter is structurally and functionally tied to the nonlinear modulation function for state variable modulation.
Furthermore, there is no teaching, suggestion, or motivation to modify Okumura in view of Yue in such a manner.” Applicant claims that Okumura and Yue do not applying a time-variable parameter to a nonlinear modulation functions. Yet both Okumura and Yue teaches a time-varying parameter (See pervious office action rejection dated 2/19/2026). Further the applicant argues amended limitations and the amend limitations have not been examined. Thus the argument is moot and not convincing.
Claim Rejections - 35 USC § 112 (b)
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.
Claim 1-13 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claim 1, the claim recite “CPU, GPU, DSP, and ASIC” in line 2 of the claim. The term are not defined in the claim making the claim indefinite. The terms should be defined. All dependent claims inherit the issue.
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.
Claim(s) 1, 2, 5, 8 and 10 are rejected under 35 U.S.C. 103 as being unpatentable over Okumura et al. (US20180293495A1) (“Okumura”) in view of Larger, Laurent, et al. "High-speed photonic reservoir computing using a time-delay-based architecture: Million words per second classification." Physical Review X 7.1 (2017): 011015. (“Larger”).
Regarding claim 1, as best understood given the 112(b) issue identified above.
Okumura teaches A reservoir element comprising: a ring-shaped reservoir constituted by a single nonlinear element and a plurality of delay elements (Okumura Fig 2.
PNG
media_image1.png
437
702
media_image1.png
Greyscale
[A reservoir element comprising: a ring-shaped reservoir]
Para 0055, FIG. 2 is a diagram illustrating a concept of the reservoir computing according to Example 1.
Para 0057 line 1-3, The reservoir unit 112 is constituted with one nonlinear node 200 accompanying time delay. The reservoir unit 112 may include two or more nonlinear nodes 200.
Para 0063, The data q(t) is transmitted to a delay network constituted with virtual nodes 201. Specifically, a value of each component of the equation (3) is emulated as a state value of the virtual node 201. [a single nonlinear element and a plurality of delay elements]),
wherein the nonlinear element has a nonlinear modulation function, the nonlinear element being controllable by a time-varying parameter capable of dynamically changing the nonlinear modulation function (Okumura para 0057, The reservoir unit 112 is constituted with one nonlinear node 200 accompanying time delay. The reservoir unit 112 may include two or more nonlinear nodes 200. When the input data x(t) is received from the input unit 111, the nonlinear node 200 divides the input data x(t) into pieces of data each of which consists of a piece of data with a time width T and executes computation processing by using the divided piece of data with the time width T as one processing unit.
Para 0058, Here, T represents a delay time (length of a delay network). The divided input data x(t) is handled as an N-dimensional vector. N represents the number of virtual nodes.
Para 0059, In the computation processing, the reservoir unit 112 executes nonlinear transformation illustrated in the data equation (2) to calculate N-dimensional data q(t). Each component of the data q(t) is expressed by the equation (3) [wherein the nonlinear element has a nonlinear modulation function].
Para 0063, The data q(t) is transmitted to a delay network constituted with virtual nodes 201. Specifically, a value of each component of the equation (3) is emulated as a state value of the virtual node 201 [the nonlinear element being controllable by a time-varying parameter capable of dynamically changing the nonlinear modulation function].)),
the reservoir element includes a control unit that is implemented by a processor including at least one of a CPU, GPU, DSP, and ASIC and configured to control the time-varying parameter in accordance with a time series having a cycle corresponding to the number of stages of the nonlinear element and the delay elements (Okumura Para 0049 line 1-4, The computation device 101 executes processing according to a program. As the computation device 101, a processor, a field programmable gate array (FPGA), or the like can be considered [is implemented by a processor including at least one of a CPU, GPU, DSP, and ASIC].
Para 0117, The reservoir unit 112 includes a computation unit 721, a laser 722, an MZ optical modulator 723, a photodiode 724, and an amplifier 725. The MZ optical modulator 723 and the photodiode 724 are connected via an optical fiber.
[0118] The computation unit 721 executes computation processing expressed by the equation (2). That is, the computation unit 721 superimposes the input data x(t) input from the input unit 111 and the data q(t) output from the reservoir unit 112. The computation unit 721 outputs the computation result as a signal to the MZ optical modulator 723.
Para 0119, The reservoir unit 112 includes a computation unit 721, a laser 722, an MZ optical modulator 723, a photodiode 724, and an amplifier 725. The MZ optical modulator 723 and the photodiode 724 are connected via an optical fiber.
Para 0120, The MZ optical modulator 723 is hardware for implementing the nonlinear node 200. In Example 2, a fiber coupled LN (LiNbO3)-MZ modulator was used. The MZ optical modulator 723 modulates intensity of laser light input from the laser 722 using the signal input from the computation unit 721. Light transmission characteristic of the MZ optical modulator 723 corresponds to a square of a sine wave with respect to an input electric signal and thus, an amplitude is nonlinearly transformed [and the reservoir element includes a control unit that is configured to control the time-varying parameter].
Para 0122, The length of the optical fiber to be connected between the MZ optical modulator 723 and the photodiode 724 is a length required for a predetermined time to transmit laser light output from the MZ optical modulator 723. The time required for transmission of laser light is a period of the delay network. In Example 2, the MZ optical modulator 723 and the photodiode 724 are connected by using an optical fiber having a length of 20 km. Accordingly, it takes 100 microseconds to transmit the signal.
Para 0126, The read circuit 731 reads the signal output from the reservoir unit 112. The read circuit 731 operates so as to be synchronized with the mask circuit 711. An operation speed of the read circuit 731 varies at an amplification factor of 1 MHz and the read circuit 731 operates at a cycle of 10 kHz. The amplification factor is determined by learning processing. The read circuit 731 outputs the read signal to the integration circuit 731 [in accordance with a time series having a cycle corresponding to the number of stages of the nonlinear element and the delay elements]).
Okumura does not explicitly teach the time-varying parameter is configured to perform a modulation of a state variable of the nonlinear element or a modulation of a state variable of the delay element which is one before the nonlinear element in the ring-shaped reservoir, the modulation being achieved by applying the time-varying parameter to the nonlinear modulation function,
However Larger teaches the time-varying parameter is configured to perform a modulation of a state variable of the nonlinear element or a modulation of a state variable of the delay element which is one before the nonlinear element in the ring-shaped reservoir, the modulation being achieved by applying the time-varying parameter to the nonlinear modulation function, (Larger Page 2 FIG 1,
PNG
media_image2.png
279
631
media_image2.png
Greyscale
page 2, A. Standard RC model, An important computational concept used in RC is the nonlinear dynamical expansion of the information to be processed into a higher-dimensional phase space, such that easy linear read-out of this expansion can be efficiently applied (for a review of the concepts briefly recalled here, see Ref. [4]). Figure 1 shows how the information is encoded and injected (write-in step, also called the input layer) into a nonlinear dynamical system (e.g., a network of firing neurons).
Considering a classical network formed by K spatially distributed nodes
PNG
media_image3.png
33
334
media_image3.png
Greyscale
, the discrete time dynamics of the network (as time n is increased) involves the coupling of each node with the other ones according to the network coupling matrix
W
N
. In addition to the internal dynamics, the network also evolves because of the injection of the input information u(n) to be processed. The information injection is ruled by an input layer coupling matrix
W
I
. This results in a network dynamics for each node amplitude xk at time n, which can be expressed as follows:
PNG
media_image4.png
122
606
media_image4.png
Greyscale
where
f
N
L
is the nonlinear node sensitivity on the accumulated stimuli coming from both the network internal connectivity and the input information connectivity. A popular form for
f
N
L
is the tanh function, also known as the sigmoid function [the time-varying parameter is configured to perform a modulation of a state variable of the nonlinear element].
Page 3, B. RC based on delay differential dynamics para 5
PNG
media_image5.png
715
613
media_image5.png
Greyscale
[, the modulation being achieved by applying the time-varying parameter to the nonlinear modulation function]),
Okumura and Larger are considered to be analogous to the claim invention because they are in the same field of reservoir computing. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Okumura to incorporate the teachings of Larger and disclose configuration of time-varying parameter to perform modulation. Doing so to use the photonic delay dynamics and take advantage of the ultrafast and efficient RC processing techniques (Larger V. Conclusion para 1 line 1-17, This work has demonstrated the potential of photonic delay dynamics for ultrafast and efficient RC processing techniques, with a novel electro-optic phase-delay dynamics architecture. A rigorous modeling correspondence of the nonlinear delay dynamics with the original RC concept was proposed, thus theoretically supporting the efficiency of the experimental demonstration. The reported photonic implementation of the RC concept proposes several attractive features: speed related to a structure based on optical telecommunication devices; processing efficiency demonstrated in the particular case of speech recognition; processing universality, according to the universal computing machine concept of RC; design simplification and accurate modeling, through its pure temporal signal processing architecture compared to complex spatiotemporal attempts for photonic RC processors [10,32,33].).
Regarding claim 2, as best understood given the (b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 1.
Okumura further teaches wherein the time- varying parameter is set by the time series which is a value variable per unit time (Okumura para 0071 In the following description, the stream N(t) in one section is described as a stream [AYit). As illustrated in FIG. 4C, the stream [AYit) has a constant value in one section.
Para 0072, Next, the input unit 111 executes mask processing for modulating intensity for each stream [AYit) every time width i:M to calculate an input stream af(t) (step S104). For example, the input stream af(t) as illustrated in FIG. 4D is obtained. In Example 1, intensity modulation is performed in the range from -1 to +1. Here, τM represents a distance between the virtual nodes and satisfies the equation (6).
Para 0076, Next, the input unit 111 executes time shift processing of generating deviation in time based on the counter value m to transform the input stream af(t) into an input stream d(t) (step S105). Thereafter, the input unit 111 proceeds to step S107.
Para 0084, Next, the input unit 111 inputs the input data x(t) to the nonlinear node 200 of the reservoir unit 112 (step S108). Thereafter, the input unit 111 ends processing [by the time series which is a value variable per unit time]).
Regarding claim 5, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 1.
Okumura teaches wherein the nonlinear modulation function is used to perform modulation using a sigmoid as a nonlinear function (Okumura para 0004, As an example of transformation of the hidden layer, there is nonlinear transformation imitating firing phenomenon of a neuron. The firing phenomenon of neuron is known as a nonlinear phenomenon in which a membrane potential rapidly rises and output varies in a case where a potential exceeding a threshold value is input to the neuron. In order to reproduce the phenomenon described above, for example, a sigmoid function expressed by the equation (1) is used.
PNG
media_image6.png
55
123
media_image6.png
Greyscale
Para 0059, In the computation processing, the reservoir unit 112 executes nonlinear transformation illustrated in the data equation (2) to calculate N-dimensional data q(t). Each component of the data q(t) is expressed by the equation (3) [wherein the nonlinear modulation function is used to perform modulation].
PNG
media_image7.png
157
422
media_image7.png
Greyscale
[using a sigmoid as a nonlinear function]).
Regarding claim 8, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 5.
Okumura teaches wherein the nonlinear modulation function is used to perform modulation using an arbitrary function combined with the nonlinear function (Okumura Para 0073, The modulation may be either amplitude modulation or phase modulation. Specific modulation is performed by multiplying the stream N(t) by a random bit sequence.
para 0077, The time shift processing may be processing of delaying the time or processing of advancing the time. For example, time shift processing represented by the equation (8) is performed.
PNG
media_image8.png
30
383
media_image8.png
Greyscale
Para 0078 line 1-3, The equation (8) is time shift processing that gives a delay to another input stream af(t) by using an arbitrary input stream af(t) as a reference.
Para 0015, The shift register 712 executes computation processing corresponding to processing of step S105 for the input stream af(t). The shift register 712 outputs the calculated input stream d(t) to the computation unit 713. In Example 2, a delay circuit for generating delay in the input stream af(t) using the shift register 712 is implemented. However, the delay circuit may be a delay circuit constituted with a ladder type transmission circuit network constituted with a capacitor and an inductor
Para 0116 The computation unit 713 executes computation processing corresponding to processing of step S107 using the input stream a/t) input from each shift register 712. The computation unit 713 outputs a computation result to the reservoir unit 112 [using an arbitrary function].
Para 0118 The computation unit 721 executes computation processing expressed by the equation (2). That is, the computation unit 721 superimposes the input data x(t) input from the input unit 111 and the data q(t) output from the reservoir unit 112. The computation unit 721 outputs the computation result as a signal to the MZ optical modulator 723. [wherein the nonlinear modulation function is used to perform modulation] (i.e. in combination with the nonlinear function)).
Regarding claim 10, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 1.
Okumura teaches wherein the nonlinear element is constituted by an optical modulation element (Okumura Para 0117, The reservoir unit 112 includes a computation unit 721, a laser 722, an MZ optical modulator 723 [an optical modulation element], a photodiode 724, and an amplifier 725. The MZ optical modulator 723 and the photodiode 724 are connected via an optical fiber.)).
Claim(s) 3 and 4 are rejected under 35 U.S.C. 103 as being unpatentable over Okumura in view of Larger and further in view of Benjamin Schrauwen, Marion Wardermann, David Verstraeten, Jochen J. Steil Dirk Stroobandt, Improving reservoirs using intrinsic plasticity, Neurocomputing, Volume 71, Issues 7–9, 2008, Pages 1159-1171, ISSN 0925-2312 (“Schrauwen”).
Regarding claim 3, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 1.
Okumura and Larger are combine in the same rational as set forth above with respect to claim 1.
Okumura does not explicitly teach wherein the time-varying parameter is set by the time series determined by IP learning.
However Schrauwen teaches wherein the time-varying parameter is set by the time series determined by IP learning (Schrauwen Page 1160 1. Introduction para 6, The ESN consists of a randomly connected recurrent network of analog neurons—the reservoir—that is driven by a (one- or multi-dimensional) temporal input signal. On the output level, the activations of the entire network are treated as high-dimensional spatio-temporal input features for a linear classification/regression algorithm—the linear readout. The ESN was introduced as an improved way to use the computational power of RNNs without training of the internal weights. From a formal viewpoint, the reservoir acts as a complex non-linear dynamic filter that transforms the input signals using a high-dimensional temporal mapping, not unlike the operation of an explicit, temporal kernel function. It is even possible to solve several classification tasks on a single input signal simultaneously by adding multiple readouts to one reservoir.
page 1161, The original work of Triesch assumes a constraint on the mean of the output distribution and derives a gradient descent learning rule from these principles for fermi nonlinearities. In this work, we extend the formalism and study the effects on two types of transfer functions:
Page 162 2.1 Effect of bounded activation values,
We now want to evaluate whether the IP learning rules are able to converge to the desired output distributions. There is however a problem due to the bounded nature of the output of the neurons we use: they are unable to output arbitrary values. The non-linearities we consider in this contribution are bounded to [0,1] for fermi and [-1,1] for tanh. Due to these bounds, when we impose a certain constraint on the moments of the infinite distribution, these moments will not be accurately approximated by a learning rule controlling a neuron with finite output range. The effects of these bounds on the moments of the actual output distribution can be computed as follows
Page 1165 3.1 Experimental Setting para 5, For pre-training a reservoir, the IP rule is applied with a learning rate of 0.0005 for 100 000 time steps (equal to 10 epochs when we use 10 time series, used for the cross validation, each consisting of 1000 time steps). To check whether IP has had sufficient time to adapt after this time, we verified that a and b had converged to small regions and compared the expected probability density with the one estimated from the reservoir’s output [wherein the time-varying parameter is set by the time series determined by IP learning]).
Okumura and Schrauwen are considered to be analogous to the claim invention because they are in the same field of reservoir computing. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Okumura to incorporate the teachings of Schrauwen use IP learning. Doing to make reservoir computing more robust in respect to weights or input scaling (Schrauwen Abstract line 6-12, The IP rule is evaluated in a reservoir computing setting, which is a temporal processing technique which uses random, untrained recurrent networks as excitable media, where the network’s state is fed to a linear regressor used to calculate the desired output. We present an experimental comparison of the different IP rules on three benchmark tasks with different characteristics. Furthermore, we show that this unsupervised reservoir adaptation is able to adapt networks with very constrained topologies, such as a 1D lattice which generally shows quite unsuitable dynamic behavior, to a reservoir that can be used to solve complex tasks. We clearly demonstrate that IP is able to make reservoir computing more robust: the internal dynamics can autonomously tune themselves— irrespective of initial weights or input scaling—to the dynamic regime which is optimal for a given task.).
Regarding claim 4, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 1.
Okumura and Larger are combine in the same rational as set forth above with respect to claim 1.
Okumura and Schrauwen are combine in the same rational as set forth above with respect to claim 3.
Schrauwen further teaches wherein the time-varying parameter is set by the time series determined by a logarithmic normal distribution (Page 1160 1. Introduction para 6, The ESN consists of a randomly connected recurrent network of analog neurons—the reservoir—that is driven by a (one- or multi-dimensional) temporal input signal. On the output level, the activations of the entire network are treated as high-dimensional spatio-temporal input features for a linear classification/regression algorithm—the linear readout. The ESN was introduced as an improved way to use the computational power of RNNs without training of the internal weights. From a formal viewpoint, the reservoir acts as a complex non-linear dynamic filter that transforms the input signals using a high-dimensional temporal mapping, not unlike the operation of an explicit, temporal kernel function. It is even possible to solve several classification tasks on a single input signal simultaneously by adding multiple readouts to one reservoir.
Page 1161, 2. Derivation of generalize IP Para 3 line 6-16 The ME distribution for a given mean (first moment) and support in the interval [0,∞] is the exponential distribution. Likewise, the ME distribution for a given mean and standard deviation with support in [-∞,∞] is the Gaussian. Therefore, to target ME output distribution in formal neurons, we use for the first moment case fermi neurons having a positive output range [0,1] and for the second order case tanh neurons with output range [-1,1]. We express the amount the empirical output distribution differs from the desired ME distribution using the Kullback–Leibler divergence:
PNG
media_image9.png
75
302
media_image9.png
Greyscale
[ by a logarithmic normal distribution]).
Claim(s) 6 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Okumura in view of Larger and further in view of Kudithipudi D, Saleh Q, Merkel C, Thesing J, Wysocki B. Design and Analysis of a Neuromemristive Reservoir Computing Architecture for Biosignal Processing. Front Neurosci. 2016 Feb 1;9:502. doi: 10.3389/fnins.2015.00502 (“Kudithipudi”).
Regarding claim 6, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 1.
Okumura and Larger are combine in the same rational as set forth above with respect to claim 1.
Okumura does not explicitly teach wherein the nonlinear modulation function is used to perform modulation using tanh as a nonlinear function.
However Kudithipudi teaches wherein the nonlinear modulation function is used to perform modulation using tanh as a nonlinear function (Kudithipudi page 7-8,
PNG
media_image10.png
539
534
media_image10.png
Greyscale
[using tanh as a nonlinear function]
PNG
media_image11.png
151
528
media_image11.png
Greyscale
[wherein the nonlinear modulation function is used to perform modulation]).
Okumura and Kudithipudi are considered to be analogous to the claim invention because they are in the same field of reservoir computing. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Okumura to incorporate the teachings of Kudithipudi to use a nonlinear modulation using tanh function. Doing to have richer reservoir dynamics leading to better classification accuracies in the ESNs (Kudithpudi page 7 5.2 Reservoir Neuron Circuits para 1 line 8-10, In this work, we’ve found that tanh activation functions also result in richer reservoir dynamics leading to better classification accuracies in ESNs.).
Regarding claim 12, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 1.
Okumura and Larger are combine in the same rational as set forth above with respect to claim 1.
Okumura and Kudithipudi are combine in the same rational as set forth above with respect to claim 6.
Okumura further teaches wherein the nonlinear modulation function is used to perform modulation using an arbitrary function combined with the nonlinear function (Okumura Para 0073, The modulation may be either amplitude modulation or phase modulation. Specific modulation is performed by multiplying the stream N(t) by a random bit sequence.
para 0077, The time shift processing may be processing of delaying the time or processing of advancing the time. For example, time shift processing represented by the equation (8) is performed.
PNG
media_image8.png
30
383
media_image8.png
Greyscale
Para 0078 line 1-3, The equation (8) is time shift processing that gives a delay to another input stream af(t) by using an arbitrary input stream af(t) as a reference.
Para 0015, The shift register 712 executes computation processing corresponding to processing of step S105 for the input stream af(t). The shift register 712 outputs the calculated input stream d(t) to the computation unit 713. In Example 2, a delay circuit for generating delay in the input stream af(t) using the shift register 712 is implemented. However, the delay circuit may be a delay circuit constituted with a ladder type transmission circuit network constituted with a capacitor and an inductor
Para 0116 The computation unit 713 executes computation processing corresponding to processing of step S107 using the input stream a/t) input from each shift register 712. The computation unit 713 outputs a computation result to the reservoir unit 112 [using an arbitrary function].
Para 0118 The computation unit 721 executes computation processing expressed by the equation (2). That is, the computation unit 721 superimposes the input data x(t) input from the input unit 111 and the data q(t) output from the reservoir unit 112. The computation unit 721 outputs the computation result as a signal to the MZ optical modulator 723. [wherein the nonlinear modulation function is used to perform modulation] (i.e. in combination with the nonlinear function)).
Claim(s) 7 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Okumura in view of Larger and further in view of Manneschi, L., A. C. Lin, and E. Vasilaki. "SpaRCe: sparse reservoir computing." arXiv:1912.08124v1 [cs.NE] 4 Dec 2019 (2019) (“Maneschi”).
Regarding claim 7, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 1.
Okumura and Larger are combine in the same rational as set forth above with respect to claim 1.
Okumura does not teach wherein the nonlinear modulation function is used to perform modulation using ReLu as a nonlinear function.
Manneschi teaches wherein the nonlinear modulation function is used to perform modulation using ReLu as a nonlinear function (Manneschi Page 2 1 Introduction para 3 line 16-29, Analogously to the concept of firing thresholds, SpaRCe exploits learnable thresholds to optimize the level of sparsity inside the network. Both the learnable thresholds and the read-out weights (but not the recurrent connections within the reservoir) are optimised by minimising an error function without exploiting any normalization term. We analysed the learning rule derived from this error minimization and found that learning occurs by two antagonist factors: the first raises the thresholds proportionally to the correlated activity of the nodes (thus silencing nodes that are correlated and therefore redundant), while the second lowers the thresholds of nodes that contribute to the correct classification (Fig. 3).
Page 3 2.1 SpaRCe, Let us consider the mean square cost function,
given by
PNG
media_image12.png
264
488
media_image12.png
Greyscale
[ReLu as a nonlinear function.]
Page 7 3.1 Odor Sequence Learning para 4 line 16-26, It is clear that there is an optimal sparsity level of about 50% where the error is minimized for all the training instance. Furthermore, the change in the threshold values obtained through the learning rule is highlighted by the black dashed lines that connect dots of training instances from the top to the bottom of the graph. All these lines tend approximately toward the optimal representation, showing how the learning rule appropriately modulates the percentage of active nodes [wherein the nonlinear modulation function is used to perform modulation]).
Okumura and Manneschi are considered to be analogous to the claim invention because they are in the same field of reservoir computing. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Okumura to incorporate the teachings of Manneschi to use a nonlinear modulation using relu function. Doing to use thresholds learned by error functions on the outputs of the network (Maneschi Abstract line 14-28, This approach, which we term SpaRCe, optimizes the sparseness level of the reservoir and applies the threshold mechanism to the information received by the read-out weights. Both the read-out weights and the thresholds are learned by a standard online gradient rule that minimises an error function on the outputs of the network. Threshold learning occurs by the balance of two opposing forces: reducing inter-neuronal correlations in the reservoir by deactivating redundant neurons, while increasing the activity of neurons participating in correct decisions. We test SpaRCe in a set of classification problems and find that introducing threshold learning improves performance compared to standard reservoir computing networks.).
Regarding claim 13, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 7.
Okumura and Larger are combine in the same rational as set forth above with respect to claim 1.
Okumura and Manneschi are combine in the same rational as set forth above with respect to claim 7.
Okumura further teaches wherein the nonlinear modulation function is used to perform modulation using an arbitrary function combined with the nonlinear function (Okumura Para 0073, The modulation may be either amplitude modulation or phase modulation. Specific modulation is performed by multiplying the stream N(t) by a random bit sequence.
para 0077, The time shift processing may be processing of delaying the time or processing of advancing the time. For example, time shift processing represented by the equation (8) is performed.
PNG
media_image8.png
30
383
media_image8.png
Greyscale
Para 0078 line 1-3, The equation (8) is time shift processing that gives a delay to another input stream af(t) by using an arbitrary input stream af(t) as a reference [using an arbitrary function].
Para 0015, The shift register 712 executes computation processing corresponding to processing of step S105 for the input stream af(t). The shift register 712 outputs the calculated input stream d(t) to the computation unit 713. In Example 2, a delay circuit for generating delay in the input stream af(t) using the shift register 712 is implemented. However, the delay circuit may be a delay circuit constituted with a ladder type transmission circuit network constituted with a capacitor and an inductor
Para 0116 The computation unit 713 executes computation processing corresponding to processing of step S107 using the input stream a/t) input from each shift register 712. The computation unit 713 outputs a computation result to the reservoir unit 112.
Para 0118 The computation unit 721 executes computation processing expressed by the equation (2). That is, the computation unit 721 superimposes the input data x(t) input from the input unit 111 and the data q(t) output from the reservoir unit 112. The computation unit 721 outputs the computation result as a signal to the MZ optical modulator 723. [wherein the nonlinear modulation function is used to perform modulation] (i.e. in combination with the nonlinear function)).
Claim(s) 9 are rejected under 35 U.S.C. 103 as being unpatentable over Okumura in view of Larger and further in view of B. Barazani, G. Dion, J. -F. Morissette, L. Beaudoin and J. Sylvestre, "Microfabricated Neuroaccelerometer: Integrating Sensing and Reservoir Computing in MEMS," in Journal of Microelectromechanical Systems, vol. 29, no. 3, pp. 338-347, June 2020, doi: 10.1109/JMEMS.2020.297846 (“Barazani”).
Regarding claim 9, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 1.
Okumura and Larger are combine in the same rational as set forth above with respect to claim 1.
Okumura does not explicitly teach wherein the nonlinear element is constituted by a MEMS element.
However Barazani teaches wherein the nonlinear element is constituted by a MEMS element (Barazani page 339, II. Desing A necessary property of physical RC is the ability to map their input signals into a high-dimensional state, via non-linear dynamics [10]. This mapping allows signals that are originally not linearly separable to be represented in a space where they can be processed by linear models. In this study, the non-linear expansion of the input results from the dynamical response of a clamped-clamped beam oscillating at large amplitudes [11], [12]. We have shown previously [8] that this dynamical response could be exploited to achieve significant neuromorphic computational capabilities, in a very small and energy efficient device. In this work, we leverage the mechanical nature of the clamped-clamped beam computing system by coupling it to a suspended proof mass that implements the sensing functions of the neuromorphic MEMS.
The neuroaccelerometer thus comprises two principal mechanical elements: the non-linear oscillating beam, which has a high natural frequency (section II-B); and a larger suspended inertial mass with a much lower natural frequency, designed to be sensitive to external accelerations (section II-A). When in operation, a pump voltage applied to the inertial mass induces an electrostatic force over the beam, driving it near resonance with large displacements, in its non-linear regime. External accelerations displace the inertial mass, thus modulating the amplitude of the driving force over the beam and consequently the beam oscillation amplitude. The displacement of the beam is measured with piezoresistive strain gauges. The signal from the gauges is digitized, delayed and fed back to the pump voltage, in a scheme described in section V-A that is useful to increase the computational power of simple dynamical systems, at the cost of reduced processing speed (wherein the nonlinear element is constituted by a MEMS element)).
Okumura and Barazani are considered to be analogous to the claim invention because they are in the same field of reservoir computing. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Okumura to incorporate the teachings of Barazani to use MEMS neuroacceleromter. Doing to accurately emulate non-linear autoregressive moving models (Barazani page 338 Abstract line 18-20, The neuromorphic MEMS accelerometer was able to accurately emulate non-linear autoregressive moving average models and compute the parity of random bit streams).
Claim(s) 11 are rejected under 35 U.S.C. 103 as being unpatentable over Okumura in view of Larger and further in view of D. Marković, N. Leroux, M. Riou, F. Abreu Araujo, J. Torrejon, D. Querlioz, A. Fukushima, S. Yuasa, J. Trastoy, P. Bortolotti, J. Grollier; Reservoir computing with the frequency, phase, and amplitude of spin-torque nano-oscillators. Appl. Phys. Lett. 7 January 2019; 114 (1): 012409 (“Markovic”).
Regarding claim 11, as best understood given the 112(b) issue identified above, Okumura in view of Larger teach the reservoir element according to claim 1.
Okumura and Larger are combine in the same rational as set forth above with respect to claim 1.
Okumura does not explicitly teach wherein the nonlinear element is constituted by a spin element.
However Markovic teaches wherein the nonlinear element is constituted by a spin element (Markovic Page 1 para 1 line 20-31 Spin-torque induced magnetization dynamics indeed takes place in nanoscale magnetic volumes, which makes them sensitive to thermal fluctuations. In addition, phase noise is enhanced by amplitude noise due to the inherent coupling between the phase and the amplitude of magnetization oscillations.13 In this work, we show that these issues can be circumvented by working in a regime where the oscillator is synchronized to the input waveform that it has to process which considerably reduces magnetization fluctuations.14 For this purpose, we use a sinusoidal input waveform that carries information encoded in its modulated frequency, chosen close to the spin-torque oscillator frequency.
Page 2 FIG. 1
PNG
media_image13.png
513
590
media_image13.png
Greyscale
[a spin element]
Page 3, The fact that the frequency, amplitude, and phase are all non-linear functions of input frequency enables us to use them for neuromorphic computing. We now demonstrate this capability on a task that consists in classification of sine and square waves of equal periods but different amplitudes. For this, we use a method called single node reservoir computing.6,7,24,25 This method uses time multiplexing in order to emulate a reservoir with a single nano-oscillator that plays a role of a different effective virtual neuron at each time step (i.e. the spin element is the nonlinear element)).
Okumura and Markovic are considered to be analogous to the claim invention because they are in the same field of reservoir computing. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Okumura to incorporate the teachings of Markovic and use a Spin element. Doing to classify sine and square waveforms with high accuracy (Markovic page 1 Abstract line 5-8, We show that this method allows classifying sine and square wave forms with an accuracy above 99% when decoding the output from the oscillator amplitude, phase, or frequency. We find that recognition rates are directly related to the noise and non-linearity of each variable. These results prove that spin-torque nano-oscillators offer an interesting platform to implement different computing schemes leveraging their rich dynamical features).
Pertinent Prior Art
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Petre et al. (US10404299B1) – teaches a delay embedding for gradient learning used in reservoir computing.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALFREDO CAMPOS whose telephone number is (571)272-4504. The examiner can normally be reached 7:00 - 4:00 pm M - F.
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, Michael J. Huntley can be reached at (303) 297-4307. 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.
/ALFREDO CAMPOS/Examiner, Art Unit 2129
/MICHAEL J HUNTLEY/Supervisory Patent Examiner, Art Unit 2129
Prosecution Insights