BrainChip + Nanose

Great work Ahboy. The two most significant pointers to AKIDA still being in the mix come from the second paragraph in the following image I have taken from your post and the fact that later in the article Orit Albak CEO of Nanose states the whole process takes 1 to 2 minutes which is what Professor Haick stated back in August, 2020 when he was interviewed following the partnership with Brainchip and he was promoting their use of AKIDA Ai with his sensors.
My opinion only DYOR
FF

AKIDA BALLISTA:

1645317169725.png
 
  • Like
  • Fire
Reactions: 22 users
Interesting patent coming out of Technion. Doesn't reference Haick but talks about neurons and unsupervised STDP. I don't recall seeing Technion previously looking into this sort of work.

Pure speculation, DYOR


United States Patent Application
20220058492
Kind Code
A1
DANIAL; Loai ; et al.
February 24, 2022

DELTA-SIGMA MODULATION NEURONS FOR HIGH-PRECISION TRAINING OF MEMRISTIVE SYNAPSES IN DEEP NEURAL NETWORKS

Abstract
A neural network comprising: a plurality of interconnected neural network elements, each comprising: a neuron circuit comprising a delta-sigma modulator, and at least one synapse device comprising a memristor connected to an output of said neuron circuit; wherein an adjustable synaptic weighting of said at least one synapse device is set based on said output of said neuron circuit


Inventors:​
DANIAL; Loai; (Nazareth, IL) ; KVATINSKY; Shahar; (Hanaton, IL)
Applicant:​
Name​
City​
State​
Country​
Type​

TECHNION RESEARCH & DEVELOPMENT FOUNDATION LIMITED

Haifa

IL​
Family ID:​
1000005999340
Appl. No.:​
17/299102
Filed:​
December 4, 2019
PCT Filed:​
December 4, 2019
PCT NO:​
PCT/IL2019/051328
371 Date:​
June 2, 2021

Related U.S. Patent Documents

Application Number
Filing Date
Patent Number
62774933​
Dec 4, 2018​



Current U.S. Class:
1/1
Current CPC Class:
G06N 3/063 20130101; G06N 3/049 20130101; G06N 3/088 20130101​
International Class:
G06N 3/08 20060101 G06N003/08; G06N 3/04 20060101 G06N003/04; G06N 3/063 20060101 G06N003/063​



Claims



1. A neural network comprising: a plurality of interconnected neural network elements, each comprising: a neuron circuit comprising a delta-sigma modulator, and at least one synapse device comprising a memristor connected to an output of said neuron circuit; wherein an adjustable synaptic weighting of said at least one synapse device is set based on said output of said neuron circuit.

2. The neural network of claim 1, wherein said plurality of interconnected neural elements form a trainable single-layer neural network, arranged as a memristive crossbar array comprising a synaptic weightings matrix.

3. (canceled)

4. The neural network of claim 2, wherein an output vector of said neural network is calculated as a weighted sum of said outputs of said neuron circuits multiplied by said synaptic weightings matrix.

5. The neural network of claim 4, further comprising an output circuit comprising at least one delta-sigma modulator, wherein said output circuit encodes said output vector.

6. The neural network of claim 1, wherein, at a training stage, said neural network is trained by an iterative process comprising: (i) inputting analog inputs into said neuron circuits of said neural network; (ii) calculating an output vector as a weighted sum of said outputs of said neuron circuits, based on a said synaptic weightings matrix; and (iii) comparing said output vector to a training dataset input, wherein said comparing leads to an adjustment of said synaptic weightings matrix.

7. The neural network of claim 6, wherein said adjustment minimizes a cost function based on a gradient descent algorithm using said delta-sigma modulators as an activation function.

8. The neural network of claim 6, wherein said iterative process continues until said output vector corresponds to said training dataset input.

9. The neural network of claim 6, wherein said training dataset input is an output of a delta-sigma modulator.

10. The neural network of claim 2, wherein said neural network comprises two or more of said single-layer neural networks arranged as a multi-layer neural network.

11. The neural network of claim 1, further comprising a plurality of input neuron circuits, a plurality of synapse devices, and at least one output neuron circuit, wherein, at a training stage, said neural network is trained by an unsupervised spike-time-dependent plasticity (STDP) process, wherein outputs of said neuron circuits reflect spikes encoded in time.

12. The neural network of claim 11, wherein said STDP process comprises comparing pre-synaptic and post-synaptic outputs of said neuron circuits, wherein a difference detected in said comparison leads to long-term potentiation or long-term depression.

13. A method comprising: providing a neural network comprising a plurality of interconnected neural network elements, each of said neural network elements comprising: a neuron circuit comprising a delta-sigma modulator, and at least one synapse device comprising a memristor connected to an output of said neuron circuit, wherein an adjustable synaptic weighting of said at least one synapse device is set based on said output of said neuron circuit; and at a training stage, training said neural network by an iterative process comprising: (i) inputting analog inputs into said neuron circuits, (ii) calculating an output vector of said neural network as a weighted sum of said outputs of said neuron circuits, based on a said synaptic weightings, and (iii) comparing said output vector to a training dataset input, wherein said comparing leads to an adjustment of said synaptic weightings.

14. The method of claim 13, wherein said plurality of interconnected neural elements form a trainable single-layer neural network, arranged as a memristive crossbar array comprising a synaptic weightings matrix.

15. (canceled)

16. The method of claim 14, wherein said output vector is calculated as a weighted sum of said outputs of said neuron circuits multiplied by said synaptic weightings matrix.

17. The method of claim 16, wherein said neural network further comprises an output circuit comprising at least one delta-sigma modulator, wherein said output circuit encodes said output vector.

18-30. (canceled)

31. A method comprising: providing a memristor driver circuit representing a trainable neural network circuit, wherein said memristor driver circuit comprises: a delta-sigma modulator configured to receive an input voltage and output a binary sequence representing an amplitude of said input signal; a memristive device; and at least one subtractor; training said memristor driver circuit by an iterative process comprising: (i) a read stage wherein said input voltage is a read voltage selected to produce a desired duty cycle of said delta-sigma modulator, and (ii) an update stage wherein said input voltage is an updating voltage reflecting a subtraction operation between a reference voltage and an output signal of said memristive device.

32. The method of claim 31, wherein said memristor driver circuit comprises a plurality of interconnected said memristor driver circuits arranged as a trainable single-layer neural network, arranged as a memristive crossbar array comprising a synaptic weightings matrix.

33. (canceled)

34. The method of claim 31, wherein said read stage reflects a feedforward operation of the neural network, and said update stage reflects an error backpropagation operation of the neural network.

35. The method of claim 31, wherein said iterative process minimizes a cost function based on a gradient descent algorithm using said delta-sigma modulators as an activation function.

36. The method of claim 31, wherein said memristor driver circuit further comprises at least one operational amplifier configured to amplify said output signal of said memristive device.
 
  • Like
  • Fire
Reactions: 12 users

Diogenese

Top 20
Just on the accuracy thing for AKIDA it has been published by Brainchip that working with Nanose initially it was 94% having increased to 95% to 98%. I am not sure where the 90% accuracy is coming from.

My opinion only DYOR
FF

AKIDA BALLISTA
The priority date of the quoted patent is 21 April 2020.

The data was analyzed by Brainchip with a Spiking Neural Network, the adjacent confusion matrix shows the results on the test set. The test set included 31 samples- 21 positives and 10 negatives from 21 tested subjects. Zero out of 21 positive samples were identified correctly which represents 100% sensitivity and 4 out of 10 negative samples were identified correctly which represents 40% specificity. The overall accuracy was 80.65% The second study was performed with the multiuse NaNose sensors installed in Sniffphone device. The dataset included 165 samples taken from 141 subjects tested with Sniffphone device at Zayed Military Hospital - 65 samples from 65 COVID-19 positive subjects and 100 samples from 76 COVID-19 negative subjects (Several negative subjects were sampled two or three times). A Linear discriminative analysis was performed. The adjacent confusion matrix shows the results on the test set that that was completely blind to the training and validation of the model. The test set included 37 samples - 8 positive and 29 negative samples from 27 tested subjects. Seven out of eight positive samples were identified correctly which represents 87.5% sensitivity, and 25 out of the 29 negative samples were identified correctly which represents 86.2% specificity. The overall accuracy was therefore 86.5%.

I have a vague recollection of Akida being subsequently fine tuned and achieving better results, but that may just be in my imagination.

In any event, the 4-bit version of Akida would be more accurate than the 2019 version, and the 4-bit version has ben reported as performing better than expected.

In
 
  • Like
Reactions: 5 users

Diogenese

Top 20
Interesting patent coming out of Technion. Doesn't reference Haick but talks about neurons and unsupervised STDP. I don't recall seeing Technion previously looking into this sort of work.

Pure speculation, DYOR


United States Patent Application
20220058492
Kind Code
A1
DANIAL; Loai ; et al.
February 24, 2022

DELTA-SIGMA MODULATION NEURONS FOR HIGH-PRECISION TRAINING OF MEMRISTIVE SYNAPSES IN DEEP NEURAL NETWORKS

Abstract
A neural network comprising: a plurality of interconnected neural network elements, each comprising: a neuron circuit comprising a delta-sigma modulator, and at least one synapse device comprising a memristor connected to an output of said neuron circuit; wherein an adjustable synaptic weighting of said at least one synapse device is set based on said output of said neuron circuit


Inventors:​
DANIAL; Loai; (Nazareth, IL) ; KVATINSKY; Shahar; (Hanaton, IL)
Applicant:​
Name​
City​
State​
Country​
Type​

TECHNION RESEARCH & DEVELOPMENT FOUNDATION LIMITED

Haifa

IL​
Family ID:​
1000005999340
Appl. No.:​
17/299102
Filed:​
December 4, 2019
PCT Filed:​
December 4, 2019
PCT NO:​
PCT/IL2019/051328
371 Date:​
June 2, 2021

Related U.S. Patent Documents

Application Number
Filing Date
Patent Number
62774933​
Dec 4, 2018​



Current U.S. Class:
1/1
Current CPC Class:
G06N 3/063 20130101; G06N 3/049 20130101; G06N 3/088 20130101​
International Class:
G06N 3/08 20060101 G06N003/08; G06N 3/04 20060101 G06N003/04; G06N 3/063 20060101 G06N003/063​



Claims



1. A neural network comprising: a plurality of interconnected neural network elements, each comprising: a neuron circuit comprising a delta-sigma modulator, and at least one synapse device comprising a memristor connected to an output of said neuron circuit; wherein an adjustable synaptic weighting of said at least one synapse device is set based on said output of said neuron circuit.

2. The neural network of claim 1, wherein said plurality of interconnected neural elements form a trainable single-layer neural network, arranged as a memristive crossbar array comprising a synaptic weightings matrix.

3. (canceled)

4. The neural network of claim 2, wherein an output vector of said neural network is calculated as a weighted sum of said outputs of said neuron circuits multiplied by said synaptic weightings matrix.

5. The neural network of claim 4, further comprising an output circuit comprising at least one delta-sigma modulator, wherein said output circuit encodes said output vector.

6. The neural network of claim 1, wherein, at a training stage, said neural network is trained by an iterative process comprising: (i) inputting analog inputs into said neuron circuits of said neural network; (ii) calculating an output vector as a weighted sum of said outputs of said neuron circuits, based on a said synaptic weightings matrix; and (iii) comparing said output vector to a training dataset input, wherein said comparing leads to an adjustment of said synaptic weightings matrix.

7. The neural network of claim 6, wherein said adjustment minimizes a cost function based on a gradient descent algorithm using said delta-sigma modulators as an activation function.

8. The neural network of claim 6, wherein said iterative process continues until said output vector corresponds to said training dataset input.

9. The neural network of claim 6, wherein said training dataset input is an output of a delta-sigma modulator.

10. The neural network of claim 2, wherein said neural network comprises two or more of said single-layer neural networks arranged as a multi-layer neural network.

11. The neural network of claim 1, further comprising a plurality of input neuron circuits, a plurality of synapse devices, and at least one output neuron circuit, wherein, at a training stage, said neural network is trained by an unsupervised spike-time-dependent plasticity (STDP) process, wherein outputs of said neuron circuits reflect spikes encoded in time.

12. The neural network of claim 11, wherein said STDP process comprises comparing pre-synaptic and post-synaptic outputs of said neuron circuits, wherein a difference detected in said comparison leads to long-term potentiation or long-term depression.

13. A method comprising: providing a neural network comprising a plurality of interconnected neural network elements, each of said neural network elements comprising: a neuron circuit comprising a delta-sigma modulator, and at least one synapse device comprising a memristor connected to an output of said neuron circuit, wherein an adjustable synaptic weighting of said at least one synapse device is set based on said output of said neuron circuit; and at a training stage, training said neural network by an iterative process comprising: (i) inputting analog inputs into said neuron circuits, (ii) calculating an output vector of said neural network as a weighted sum of said outputs of said neuron circuits, based on a said synaptic weightings, and (iii) comparing said output vector to a training dataset input, wherein said comparing leads to an adjustment of said synaptic weightings.

14. The method of claim 13, wherein said plurality of interconnected neural elements form a trainable single-layer neural network, arranged as a memristive crossbar array comprising a synaptic weightings matrix.

15. (canceled)

16. The method of claim 14, wherein said output vector is calculated as a weighted sum of said outputs of said neuron circuits multiplied by said synaptic weightings matrix.

17. The method of claim 16, wherein said neural network further comprises an output circuit comprising at least one delta-sigma modulator, wherein said output circuit encodes said output vector.

18-30. (canceled)

31. A method comprising: providing a memristor driver circuit representing a trainable neural network circuit, wherein said memristor driver circuit comprises: a delta-sigma modulator configured to receive an input voltage and output a binary sequence representing an amplitude of said input signal; a memristive device; and at least one subtractor; training said memristor driver circuit by an iterative process comprising: (i) a read stage wherein said input voltage is a read voltage selected to produce a desired duty cycle of said delta-sigma modulator, and (ii) an update stage wherein said input voltage is an updating voltage reflecting a subtraction operation between a reference voltage and an output signal of said memristive device.

32. The method of claim 31, wherein said memristor driver circuit comprises a plurality of interconnected said memristor driver circuits arranged as a trainable single-layer neural network, arranged as a memristive crossbar array comprising a synaptic weightings matrix.

33. (canceled)

34. The method of claim 31, wherein said read stage reflects a feedforward operation of the neural network, and said update stage reflects an error backpropagation operation of the neural network.

35. The method of claim 31, wherein said iterative process minimizes a cost function based on a gradient descent algorithm using said delta-sigma modulators as an activation function.

36. The method of claim 31, wherein said memristor driver circuit further comprises at least one operational amplifier configured to amplify said output signal of said memristive device.
Hi IDD,

The word DELTA is used to identify the difference between two values, and the word SIGMA is used to refer to the sum of two values.

The delta-sigma modulator is used in an attempt to overcome problems in the variability of manufacturing memristors which affects the accuracy of these analog devices.

It is also said to provide sparsity to reduce power consumption and improve speed.

[0061] A potential advantage of the present invention is, therefore, in that it utilizes the resemblance between a biological neuron and a .DELTA..SIGMA. modulator to improve training and inference accuracy, by precise programming control of memristive synapses. In some embodiments, this approach may overcome the intrinsic variability of the memristors, in a solution which encodes information using firing rates and timing spikes and consumes less power budget than PWM. In some embodiments, the present approach eliminates data movements out of the memory structure, and therefore reduces energy consumption, improves execution time, and is thus suitable for low power applications.


[0082] A .DELTA..SIGMA. modulator may be used to modulate the amplitude of input signals to a binary sequence, with the percentage of is proportional to the amplitude. The .DELTA..SIGMA. modulator encodes the amplitude of the input signal into frequency [sequence] of 1s. If the amplitude of the pulse is positive, the output will have a higher number of 1s and vice versa. At 0 input, we get 50% duty cycle output. The difference between the input and the previous output passes through an integrator. A comparator samples the integrated voltages and produces the 1s and 0s. A digital-to-analog converter (DAC) is the used to convert these 1s and 0s back to V.sub.plus and V.sub.minus. The input to the modulator can be a continuous voltage signal, varying between the V.sub.plus and V.sub.minus values of the DAC. This is so that the integrator does not become unstable.

The system described would appear to be some time away from commercialization, but the patent dates from 2019.
 
  • Like
  • Fire
Reactions: 9 users
The priority date of the quoted patent is 21 April 2020.

The data was analyzed by Brainchip with a Spiking Neural Network, the adjacent confusion matrix shows the results on the test set. The test set included 31 samples- 21 positives and 10 negatives from 21 tested subjects. Zero out of 21 positive samples were identified correctly which represents 100% sensitivity and 4 out of 10 negative samples were identified correctly which represents 40% specificity. The overall accuracy was 80.65% The second study was performed with the multiuse NaNose sensors installed in Sniffphone device. The dataset included 165 samples taken from 141 subjects tested with Sniffphone device at Zayed Military Hospital - 65 samples from 65 COVID-19 positive subjects and 100 samples from 76 COVID-19 negative subjects (Several negative subjects were sampled two or three times). A Linear discriminative analysis was performed. The adjacent confusion matrix shows the results on the test set that that was completely blind to the training and validation of the model. The test set included 37 samples - 8 positive and 29 negative samples from 27 tested subjects. Seven out of eight positive samples were identified correctly which represents 87.5% sensitivity, and 25 out of the 29 negative samples were identified correctly which represents 86.2% specificity. The overall accuracy was therefore 86.5%.

I have a vague recollection of Akida being subsequently fine tuned and achieving better results, but that may just be in my imagination.

In any event, the 4-bit version of Akida would be more accurate than the 2019 version, and the 4-bit version has ben reported as performing better than expected.

In
Hi Dio
The following presentation by Peter van der Made in December, 2020 to the BIC gives a large number of comparisons but confirms that the first number we were provided with was 94% accuracy for Covid 19 using the data supplied to Brainchip by Nanose. In one of Rob Telson's Perspective Series articles he refers to 98% being achieved. This was the first time 98% was mentioned.


My opinion only DYOR
FF

AKIDA BALLISTA
 
  • Like
Reactions: 7 users
The following link should take you to a page where researchers mainly from the University of Warwickshire in the UK have reported the results of a number of experiments where they have been able to successfully diagnose from breath (VOC's) a range of medical diseases. Does not reference AKIDA or Brainchip but confirms the research by others Nanose included that taking samples of breath can provide instant diagnostics at low cost during routine consults:


My opinion only DYOR
FF

AKIDA BALLISTA
 
  • Like
Reactions: 8 users
D

Deleted member 118

Guest
Technion Introducing LNBD: Detecting Diseases Through Breath


 
  • Like
  • Love
  • Fire
Reactions: 9 users
D

Deleted member 118

Guest
Artificially Intelligent Medical Sensors for Clinical Decisions

 
Last edited by a moderator:
  • Like
Reactions: 5 users
D

Deleted member 118

Guest
Last edited by a moderator:
  • Like
Reactions: 3 users
D

Deleted member 118

Guest
Last edited by a moderator:
  • Like
  • Fire
Reactions: 4 users
D

Deleted member 118

Guest
Last edited by a moderator:
  • Like
  • Fire
Reactions: 5 users

uiux

Regular
  • Like
Reactions: 1 users

Dougie54

Regular
  • Like
Reactions: 3 users

uiux

Regular
My apologies should have post the above in the General Discussion thread.

Just delete the post and move it
 
  • Like
Reactions: 1 users

Neuromorphia

fact collector
Nanose Patent
WO2021214763A1 DEVICE AND METHOD FOR RAPID DETECTION OF VIRUSES

Applicants: TECHNION RES & DEVELOPMENT FOUND LTD [IL]; NANOSE MEDICAL LTD [IL]

Inventors: HAICK HOSSAM [IL]; MAROM ALBECK ORIT [IL]; SELLA TAVOR OSNAT [IL]

 
  • Like
  • Fire
Reactions: 9 users
  • Like
Reactions: 2 users

chapman89

Founding Member
Here’s the link I’ve posted in the discussion thread dated February 20222


 
  • Like
Reactions: 8 users
D

Deleted member 118

Guest
Very interesting that nasa are pushing to develope(finalise) a VOC device and I wonder if they are trying to utilise Akida into it.


Now that would be something else and maybe that’s where a few of our chips have gone.


Respiratory viral infection can sometimes lead to serious, possibly life threatening complications. Highly contagious respiratory diseases cause significant disruptions to social and economic systems if spread is uncontrolled. Therefore, the rapid and precise identification of viral infection before entering crowded or vulnerable areas is essential for suppressing their transmission effectively. Additionally, a device that is reconfigurable to address the next pandemic is highly desired. Screening for infection via exhaled breath analysis could provide a quick and simple method to find infectious carriers. This breath analyzer conceptualizes a rapid scanning device enabling the user to determine the presence of viral infection in an exhaled breath through analyzing volatile organic compounds (VOCs) concentrations.N5 Sensors will technically evaluate the feasibility of volatile organic compounds (VOCs) sensors for realizing Rapid Infection Screening via Exhalation (RISE) in a breathalyzer able to identify respiratory virus infected individuals, suitable for mass-testing scenarios.The proposed survey is expected to provide the guidance how to devise an integrated sensor system for actualizing initial screening at key check points. The evaluation will be accomplished by performing market survey, research level survey, and receiving consulting from breath analyzer pioneering companies for 1) Breath analyzer platform2) VOC gas sensors and 3) Machine learning algorithm. The survey will be progressed within stepwise assessment from initial database search, article screening and selection, to quality assessment and assortment. A comprehensive final report will be provided in which our findings and research strategy for Phase II are presented.


Triton Systems, Inc. will identify, design, and develop three noninvasive diagnostic tools to screen breath for the presence of communicable respiratory viral infections. The proposed screening technologies will be based on low-cost, high-throughput sensing modalities capable of detecting unique signatures of viral pathogens. After conducting a thorough and technical review of available targets, including volatile organic contaminants (VOCs), and suitable sensing platforms, Triton will develop sensor components into an integrated system with minimal form factor for use as personal health monitors or at travel checkpoints in highly trafficked areas. The proposed sensors will be easy to administer and widely deployable to maximize their benefit during seasonal epidemics and global pandemics involving communicable respiratory viral infections. Emphasis will be placed on an inexpensive platform with superlative sensitivity, selectivity, stability, and throughput. Wireless communication capabilities will enable the presentation and recording of the screening results in under five minutes at the site of use. Combined, the sensor components will be a widely deployable platform for detecting viral respiratory agents with pandemic potential, enhancing public health emergency readiness, and improving the transportation security infrastructure in the United States.


The COVID-19 pandemic is an example of human vulnerability to new communicable respiratory viral infections. Currently, most viral respiratory infections in humans are detected by sensing the presence of the pathogen's genetic material or proteins (i.e., antigens) in bodily fluids. Polymerase chain reaction (PCR)-based methods are the most commonly used to detect a pathogen's genetic material. Although samples can be collected outside the lab, it requires specialized laboratories and skilled technicians to collect samples, perform the tests, and analyze results. Furthermore, these tests requires hours to days to process and provide results. Additionally, their sampling methods are generally invasive. Accelerating the development of new, near real time, inexpensive, user-friendly, non-invasive, accurate, and sensitive detection technologies will contribute significantly to the national and worldwide efforts to curb communicable respiratory viral infections, like the COVID-19 pandemic. During Phase I, Lynntech will use its extensive expertise in portable chemical and biochemical sensor development (including sensors for VOC detection) to select portable, fast, reliable sensors/detectors that could be used to detect VOC markers in exhaled breath and that are associated with infectious agents. During Phase II, Lynntech will develop prototypes of the candidate approach and conduct tests to demonstrate the device's capability in the detection of VOC markers of a viral infection.
 
Last edited by a moderator:
  • Like
  • Fire
  • Thinking
Reactions: 18 users
D

Deleted member 118

Guest
 
  • Like
Reactions: 3 users
Top Bottom