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 .
This office action is in response to submission of application on 1/16/2024.
Claims 1-20 are presented for examination.
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.
Claims 18 and 19 are rejected under 35 U.S.C. 112(b), second paragraph, as failing to set forth 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.
Claim 19 recites “The method of claim 18” whereas claim 18 recites “The computing device of claim 17”. It is not clear which statutory category the claims belong. Therefore, the claims are indefinite.
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 20 is rejected under 35 USC § 101 because the claim encompasses non-statutory subject matter. Official Gazette Notice 1351 OG 212, dated February 23, 2010, states “the broadest reasonable interpretation of a claim drawn to a computer readable medium…typically covers forms of non-transitory tangible media and transitory propagating signals per se in view of the ordinary and customary meaning of computer readable media.” Given that claim 20 is drawn to a computer readable medium, the claim is construed to cover both transitory and non-transitory media. The Examiner notes that the specification of the instant application recites “The computer readable storage medium may be a tangible or in transitory/non-transitory medium such as optical (e.g., CD, DVD, Blu-Ray, etc.), magnetic, hard disk, volatile or non-volatile, solid state, or any other type of storage medium known in the art.” (Specification, paragraph [00196], line 6.) However, this is open-ended and is not explicitly recited in the claim. Therefore, claim 20 is rejected.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 12, and 20 are rejected under 35 U.S.C. § 103 as being unpatentable over Innatra, (CN 113272828B, (based on translation provided by examiner), Elastic Neural Network, herein Innatra), Iyer, et al (Avoiding Catastrophe: Active Dendrites Enable Multi-Task Learning in Dynamic Environments, herein Iyer), and Zhu, et al (SpikeGPT: Generative Pre-trained Language Model with Spiking Neural Networks, herein Zhu).
Regarding claim 1,
Innatra teaches a method for implementing a Spiking Neural Network (SNN) (Innatra, Abstract, line 1 “The spiking neural network includes a plurality of spiking neurons and a plurality of synaptic elements interconnected to form a network.” In other words, a plurality of spiking neurons and a plurality of synaptic elements interconnected to form a network is a method for implementing a spiking neural network.) comprising:
creating at least one layer of spiking neurons (Innatra, page 4, line 1 “In such event-based impulse neural networks (SNNs), only neurons that change state will generate impulses and may trigger signal processing in subsequent layers, thereby saving computational resources.” In other words, neuron is spiking neuron and subsequent layers is at least one layer of spiking neurons.) , wherein each spiking neuron of the spiking neurons comprises:
[a somatic input of dimensionality d; at least one dendritic input of dimensionality d; and an output of dimensionality d; providing an input [token] through the somatic input; and providing contextual data through the at least one dendritic input]; wherein
[the output of each of the spiking neurons is modulated by the contextual data].
Thus far, Innatra does not explicitly teach a somatic input of dimensionality d; at least one dendritic input of dimensionality d; and an output of dimensionality d; providing an input [token] through the somatic input; and providing contextual data through the at least one dendritic input.
Iyer teaches a somatic input of dimensionality d; at least one dendritic input of dimensionality d; and an output of dimensionality d; providing an input [token] through the somatic input; and providing contextual data through the at least one dendritic input (Iyer, Figure 2, and, page 8, column 2, paragraph 5, line 7 “The dimensionality of the context vector is thus identical to the dimensionality of the input vectors.”
PNG
media_image1.png
582
782
media_image1.png
Greyscale
In other words, from Figure 2, input x is somatic input, context input is dendritic input, output is output, dimensionality of the context vector is identical to the dimensionality of the input vectors is both the somatic input and dendritic input are of dimensionality d, from Figure 2, the output vector has the same dimensionality as the context and input vectors, and context vector is providing contextual data through the at least one dendritic input.)
Iyer teaches the output of each of the spiking neurons is modulated by the contextual data (Iyer, Figure 2, In other words, the input context vector applied to the layers of spiking neural networks (see inset of Figure 2) is the output of each of the spiking neurons is modulated by contextual data.).
Both Innatra and Iyer are directed to spiking neural networks, among other things. Innatra teaches a method for implementing a Spiking Neural Network (SNN) comprising:
creating at least one layer of spiking neurons, wherein each spiking neuron of the spiking neurons; but does not explicitly teach a somatic input of dimensionality d; at least one dendritic input of dimensionality d; and an output of dimensionality d; providing an input through the somatic input; and providing contextual data through the at least one dendritic input wherein the output of each of the spiking neurons is modulated by the contextual data. Iyer teaches a somatic input of dimensionality d; at least one dendritic input of dimensionality d; and an output of dimensionality d; providing an input through the somatic input; and providing contextual data through the at least one dendritic input wherein the output of each of the spiking neurons is modulated by the contextual data.
In view of the teaching of Innatra, it would be obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teaching of Iyer into Innatra. This would result in a method for implementing a Spiking Neural Network (SNN) comprising: creating at least one layer of spiking neurons, wherein each spiking neuron of the spiking neurons; and a somatic input of dimensionality d; at least one dendritic input of dimensionality d; and an output of dimensionality d; providing an input through the somatic input; and providing contextual data through the at least one dendritic input wherein the output of each of the spiking neurons is modulated by the contextual data.
One of ordinary skill in the art would be motivated to do this to build systems that can adapt to changing environments by using context through dendritic input. (Iyer, page 1, paragraph 1, line 1 “A key challenge for AI is to build embodied systems that operate in dynamically changing environments. Such systems must adapt to changing task contexts and learn continuously. Although standard deep learning systems achieve state of the art results on static benchmarks, they often struggle in dynamic scenarios. In these settings, error signals from multiple contexts can interfere with one another, ultimately leading to a phenomenon known as catastrophic forgetting. In this article we investigate biologically inspired architectures as solutions to these problems. Specifically, we show that the biophysical properties of dendrites and local inhibitory systems enable networks to dynamically restrict and route information in a context-specific manner.”)
Thus far, the combination of Innatra and Iyer does not explicitly teach the input is token input. Zhu teaches input token (Zhu, Figure 1, and, abstract, line 13 “To the best of our knowledge, SpikeGPT is the largest backpropagation-trained SNN model to date, rendering it suitable for both the generation and comprehension of natural language. We achieve this by modifying the transformer block to replace multi-head self-attention to reduce quadratic computational complexity O(N2) to linear complexity O(N) with increasing sequence length. Input tokens are instead streamed in sequentially to our attention mechanism (as with typical SNNs).”).
PNG
media_image2.png
528
762
media_image2.png
Greyscale
In other words, input tokens is input tokens.)
Both Zhu and the combination of Innatra and Iyer are directed to spiking neural networks, among other things. The combination of Innatra and Iyer teaches a method for implementing a Spiking Neural Network (SNN) comprising: creating at least one layer of spiking neurons, wherein each spiking neuron of the spiking neurons comprises: a somatic input of dimensionality d; at least one dendritic input of dimensionality d; and an output of dimensionality d; providing an input
In view of the teaching of the combination of Innatra and Iyer, it would be obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teaching of Zhu into the combination of Innatra and Iyer. This would result in a method for implementing a Spiking Neural Network (SNN) comprising: creating at least one layer of spiking neurons, wherein each spiking neuron of the spiking neurons comprises: a somatic input of dimensionality d; at least one dendritic input of dimensionality d; and an output of dimensionality d; providing an input token through the somatic input; and providing contextual data through the at least one dendritic input; wherein the output of each of the spiking neurons is modulated by the contextual data.
One of ordinary skill in the art would be motivated to do this to use spiking neural networks to more efficiently implement language models that take token input by reducing energy requirements and computational overhead (Zhu, abstract, line 1 “As the size of large language models continue to scale, so does the computational resources required to run it. Spiking Neural Networks (SNNs) have emerged as an energy-efficient approach to deep learning that leverage sparse and event-driven activations to reduce the computational overhead associated with model inference.” )
Claim 12 is a computing device claim corresponding to method claim 1. Otherwise, they are not patentably distinct. The combination of Innatra, Iyer, and Zhu teaches a computing device (Innatra, page 7, paragraph 11, line 1 “The microcontroller integrated circuit 100 further includes a microprocessor core 101 to perform computations and control of the integrated circuit 100. For example, microprocessor core 101 may oversee communication between memory mapped control and configuration interface 113 and memory device 102.” In other words, integrated circuit is computing device.). Therefore, claim 12 is rejected for the same reasons as claim 1.
Claim 20 is a computer readable medium claim corresponding to method claim 1. Otherwise, they are not patentably distinct. The combination of Innatra, Iyer, and Zhu teaches a computer readable medium (Innatra, see above mapping. In other words, memory device is computer readable medium.). Therefore, claim 20 is rejected for the same reasons as claim 1.
Claims 2-6, and 13-17 are rejected under 35 U.S.C. § 103 as being unpatentable over Innatra, Iyer, Zhu, and Ferrand, et al (Context-Dependent Computations in Spiking Neural Networks with Apical Modulation, herein Ferrand).
Regarding claim 2,
The combination of Innatra, Iyer, and Zhu teaches the method of claim 1, wherein
[the at least one dendritic input comprises N basal dendritic inputs and N apical dendritic inputs], and wherein
a maximum input sequence length is N (Innatra, FIGURE 2. And, page 4, column 2, paragraph 7, line 1“ Building on the original HTM neuron model (Hawkins and Ahmad, 2016), our Active Dendrites Neuron [Figure 2 (right inset)] receives two sources of input, analogous to the proximal and distal inputs in pyramidal neurons. Feedforward activation is computed by a linear weighted sum of the feedforward input vector, identical to the mechanism in a point neuron.” Examiner notes that the specification of the instant application recites “In at least some implementations of the first aspect, the second aspect, or the third aspect, the at least one dendritic input comprises N basal dendritic inputs and N apical dendritic inputs, and the maximum input sequence length is N.” (Specification, paragraph [0009], line 1.) Based on this, examiner is interpreting that the maximum input sequence refers to the length of the input vector since a large language model can have continuous input. In other words, input vector which is the size of the context vector is the maximum input sequence of a size (vector size) N.).
Thus far, the combination of Innatra, Iyer, and Zhu does not explicitly teach the at least one dendritic input comprises N basal dendritic inputs and N apical dendritic inputs.
Ferrand teaches the at least one dendritic input comprises N basal dendritic inputs and N apical dendritic inputs (Ferrand, Fig. 1, and, page 384, paragraph 1, line 2 “In the second variant, recurrent connections to both the basal and the apical compartment exist.”
PNG
media_image3.png
542
802
media_image3.png
Greyscale
In other words, the vector size in Fig. 1 B that remains the same for all inputs is N inputs, and connections to both the basal and the apical compartment exist is the at least one dendritic input comprises N basal dendritic inputs and N apical dendritic inputs.).
Both Ferrand and the combination of Innatra, Iyer, and Zhu are directed to spiking neural networks (SNN), among other things. The combination of Innatra, Iyer, and Zhu teaches the method of claim 1, but does not explicitly teach the at least one dendritic input comprises N basal dendritic inputs and N apical dendritic inputs. Ferrand teaches the at least one dendritic input comprises N basal dendritic inputs and N apical dendritic inputs.
In view of the teaching of the combination of Innatra, Iyer, and Zhu, it would be obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teaching of Ferrand into the combination of Innatra, Iyer, and Zhu. This would result in the method of claim 1, where the at least one dendritic input comprises N basal dendritic inputs and N apical dendritic inputs.
One of ordinary skill in the art would be motivated to do this to enhance context dependent computations to improve accuracy of spiking neural networks (Ferrand, abstract, line 8 “The model consists of a basal and an apical compartment, where the latter modulates activity of the former in a multiplicative manner. We show that this model captures the experimentally observed properties of top-down modulated activity of cortical pyramidal neurons. We evaluated recurrently connected networks of such neurons in a series of context-dependent computation tasks. Our results show that the resulting novel spiking neural network model can significantly enhance spike-based context-dependent computations.”)
Regarding claim 3,
The combination of Innatra, Iyer, Zhu, and Ferrand teaches the method of claim 2, wherein
the contextual data comprises an input sequence (Iyer, Table 2
PNG
media_image4.png
516
786
media_image4.png
Greyscale
In other words, feedforward input size for active dendrites network is the contextual data comprises an input sequence. Examiner notes that the mapping of claim 1 also teaches an input sequence. See Zhu, abstract, line 13, “sequence length” is input sequence.) .
Regarding claim 4,
The combination of Innatra, Iyer, Zhu, and Ferrand teaches the method of claim 3, wherein
the input token comprises a Query for a selected token of the input sequence (Zhu, abstract, line 17 “Input tokens are instead streamed in sequentially to our attention mechanism
(as with typical SNNs). Our preliminary experiments show that SpikeGPT remains competitive with non-spiking models on tested benchmarks, while maintaining 20× fewer operations when processed on neuromorphic hardware that can leverage sparse, event-driven activations.” And, page 3, paragraph 2, line 3 “The first Spiking Transformer model was proposed in [22], which proposes spiking self-attention to model visual features using sparse Query, Key and Value
matrices.” Examiner notes that one of ordinary skill in the art would know that the input for a language model during inference is typically called a query. In other words, input token is input token, input tokens are streamed is input sequence, and query is query.)
Regarding claim 5,
The combination of Innatra, Iyer, Zhu, and Ferrand teaches the method of claim 4, wherein
the N basal dendritic inputs are configured to receive N Keys, wherein the N apical dendritic inputs are configured to receive N Values (Zhu, Figure 1, and, page 3, paragraph 2, line 3 “The first Spiking Transformer model was proposed in [22], which proposes spiking self-attention to model visual features using sparse Query, Key and Value matrices.” In other words, from prior mapping, query is input tokens, key and value matrices is N keys and N values. Examiner notes that apical dendritic inputs is previously mapped to Ferrand.) , wherein
each of the N Keys and each of the N Values are derived from N tokens in the input sequence (Zhu, See above mapping, In other words, key and value matrices is N keys and N values, and query matrix is N tokens in the input sequence.).
Regarding claim 6,
The combination of Innatra, Iyer, Zhu, and Ferrand teaches the method of claim 5, wherein
the N basal dendritic inputs and the N apical dendritic inputs are shared dendrites whose output is received by each of the spiking neurons (Ferrand, Fig. 1, and, page 384, paragraph 1, line 2 “In the second variant, recurrent connections to both the basal and the apical compartment exist. Finally, all network neurons project to an output layer that consists of (non-spiking) leaky integrators. The activation of these neurons determines the output of the network.” In other words, from Fig. 1B, the basal and apical inputs are from a shared dendrite, and, from mapping of claim 1, layers of spiking neurons send output to successive layers of spiking neurons is the output is received from each of the spiking neurons.).
Claims 13-17 are computing device claims that correspond to method claims 2-6, respectively. Otherwise, they are not patentably distinct. Therefore, claims 13-17 are rejected for the same reasons as claims 2-6, respectively.
Allowable Subject Matter
Claim 7 is objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims. Claim 8 is objected to for depending from a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Claim 9 is objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims. Claims 10 and 11 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims
Claim 18 recites the same patentably distinct limitations as claim 9 and would be objected to for depending from a rejected base claim if not rejected under 35 U.S.C. § 112(b) for being indefinite. Claim 19 depends from claim 18 and would be objected to for depending from a rejected base claim if not rejected under 35 U.S.C. § 112(b) for being indefinite.
The prior art made of record and not used is considered pertinent to applicant’s disclosure:
Cleland, et al (WO 2020/210673 A1) “Neuromorphic Algorithm for Rapid Online Learning and Signal Restoration” discloses a computer-implemented method of training a neural network to recognize sensory patterns includes obtaining input data, preprocessing the input data in one or more preprocessors of the neural network, and applying the preprocessed input data to a core portion of the neural network.
Bicknell, et al “A synaptic learning rule for exploiting nonlinear dendritic computation” discloses a plasticity rule that optimally adjusts the strengths of interacting synapses to control somatic spiking, that shows that neurons can learn to harness the biophysical properties of their dendrites to perform nonlinear computations.
Georgescu, et al “Nonlinear neurons with human-like apical dendrite activations” discloses a new model of artificial neuron along with a novel activation function enabling the learning of nonlinear decision boundaries using a single neuron.”
Capone, et al “Beyond spiking networks: The computational advantages of dendritic amplification and input segregation” discloses an architecture that supports a burst-dependent learning rule, based on the comparison between the target bursting activity triggered by the teaching signal and the one caused by the recurrent connections, providing support for target-based learning.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BART RYLANDER whose telephone number is (571)272-8359. The examiner can normally be reached Monday - Thursday 8:00 to 5:30.
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, Miranda Huang can be reached at 571-270-7092. 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.
/Bart I Rylander/Examiner, Art Unit 2124
Prosecution Insights