BRN Discussion Ongoing

[Fullmoonfever]

So, I just had a circle back to Circle 8 (pun intended ) and found a couple of fun dots that link with timelines and maybe some of the original Dev ideas behind it...even a pic of one of the CSIRO / UTS prototypes I presume.

Don't know if it's been posted before but all looks pretty cool.

The latest I just found is this Dr from ECU which from memory is where we are based here in Perth.

Appears in his research projects he got a grant for 23 / 24. Safe bet tied in with us.

Dr Syed Mohammed Shamsul Islam

Syed Islam is a Senior Lecturer of Computer Science and a Member of Low-risk Ethics and Grants Evaluation sub-committees in the School of Science

www.ecu.edu.au

• Development of a computer vision based system for recognition of limited waste items to be stored in Circle 8 Smart bin, MGX Enterprise Pty Ltd, Grant, 2023 ‑ 2024, $197,003.

Dr Syed Mohammed Shamsul Islam
Senior Lecturer

Background​

Dr Islam is a Senior Lecturer in Computer Science at Edith Cowan University, Australia. He is also serving as a University Contact Officer under Pro-Vice-Chancellor (Equity and Indigenous), and a member of the Academic Study Leave Committee and Low-risk Ethics Committee. He is also a founding member of the School of Science Centre for AI and ML (CAIML) and the founding Lead of the 3D sensing, visualization, and analytics lab. Before joining ECU, He worked in different teaching and research positions at the University of Western Australia (UWA) and Curtin University. He conducts innovative teaching and multidisciplinary research in the areas of Artificial Intelligence, Medical Imaging, Biometrics, Machine Learning, Networking, and Computer Vision.

That then led me to look back at Circle 8 tie in with UTS which started in 2022.

University of Technology Sydney

web-tools.uts.edu.au

Circle 8 – Smart Bin Development​

Project Member(s): Liu, R., Wang, X.

Funding or Partner Organisation: MGX Enterprise Pty Ltd (Circle 8 – Smart Bin Development)
MGX Enterprise Pty Ltd (Circle 8 – Smart Bin Development)

Start year: 2022

Then found a blog on CSIRO site from Aug 2022 about their tie in with UTS for smart bin tech which includes a pic.

Smart Bin Technology sorts recyclable bottles automatically

We've developed Smart Bin Technology with the University of Technology Sydney (UTS). Our bin can automatically separate glass, metal and plastic bottles for recycling.

www.csiro.au

BY NATALIE KIKKEN16 AUGUST 20222 MIN READ


Recycling bottles could soon get a whole lot easier and more efficient. That’s because we’ve developed Smart Bin Technology. It can automatically classify and sort recyclable bottles.

Smart Bin Technology can sort plastic bottles from other bottles made of glass and metal to improve recycling


Smart bin, smart technology

We developed Smart Bin Technology with the University of Technology Sydney (UTS). It separates metal, glass, and plastic bottles for recycling. And even better – it can separate plastic bottles depending on the plastic type.

Smart Bin Technology uses motion control plus metal and weight detectors to help assess what type of bottle someone has thrown in the bin. It also uses Internet of Things, sensing and robotics to detect, classify and sort recyclables into their respective recycle bins.

Additionally the bin can sort bottles into various types of plastics, such as PET (commonly used for beverage packaging) and HDPE (used for items like shampoo bottles). It differentiates between the types of plastic by using artificial intelligence (AI), computer vision, and Near-Infrared Spectroscopy.

Smart Bin Technology uses AI to identify what types of bottles are being put in the bin so they can be sorted appropriately

This blog also contained a link to Smart Bin Tech, a site from 2022 which I presume is the base model and outputs of the actual bin data we may see an evolution of in the Circle 8 / Akida product...I hope.

SmartBin.Tech

UTS and CSIRO are developing Smart Bin Technologies, including IoT, sensing, robotics, and AI, to automatically classify and sort recycled bottles for circular economy

smartbin.tech

We are also developing a data system for users to record and visualise their recycle activities, and for administrators to manage the smart bins. In particular, users can register in the data system to record their recycle activities and receive rewards. They can then visualise and share their recycle statistics through User Dashboard. Administrators can monitor bin status, including item counts and fill levels, in Bin Map, and receive bin full alerts via SMS or email in real time.
The separation and sorting of plastic waste serve as major steps in plastic recycling. Smart Bin Tech aims to change the ways that recycled materials are collected and processed to improve recycling rates and to reduce landfill. This research is part of CSIRO’s Ending Plastic Waste Mission, which has a goal of an 80 per cent reduction of plastic waste entering the Australian environment by 2030.
Innovative technologies have been developed, including

• Sensing: to detect recycled object types, metal, glass, or plastics.
• AI image processing: further classify plastics into PET and HDPE (common plastic bottles).
• Infrared spectroscopy: provide accurate plastic classification. detect more plastic types, PET, HDPE, PP, and PS.
• Robotics: automatically sort recycled bottles into designated bins.
• IoT: communicate recycle statistics and bin status to SmartBin.Tech data system

Smart Bin Data System​

We are also building a Smart Bin Data System for users to record and visualise their recycle activities, and for administrators to manage the smart bins. The data system includes:
User Dashboard:

• User reward status
• User recycle statistics chart

Smart Bin Map

• Bin status: item counts and bin fill levels
• send bin full alerts via email or SMS

 

[Fullmoonfever]

I'm excited!

Someone just sent me the link to the patent application for TeNNs which was published on 20231228.

https://patents.google.com/patent/WO2023250092A1/en?inventor=Olivier+Jean-Marie+Dominique+COENEN

Use "Download PDF"in the top blue box to get the full document with drawings.

[0005] The CNNs are capable of learning crucial spatial correlations or features in spatial data, such as images or video frames, and gradually abstracting the learned spatial correlations or features into more complex features as the spatial data is processed layer by layer. These CNNs have become the predominant choice for image classification and related tasks over the past decade. This is primarily due to the efficiency in extracting spatial correlations from static input images and mapping them into their appropriate classifications with the fundamental engines of deep learning like gradient descent and backpropagation paring up together. This results in state-of-the-art accuracy for the CNNs. However, many modem Machine Learning (ML) workflows increasingly utilize data that come in spatiotemporal forms, such as natural language processing (NLP) and object detection from video streams. The CNN models lack the power to effectively use temporal data present in these application inputs. Importantly, CNNs fail to provide flexibility to encode and process temporal data efficiently. Thus, there is a need to provide flexibility to artificial neurons to encode and process temporal data efficiently.

Nice D.

Actually really like this last bit.


The CNN models lack the power to effectively use temporal data present in these application inputs. Importantly, CNNs fail to provide flexibility to encode and process temporal data efficiently. Thus, there is a need to provide flexibility to artificial neurons to encode and process temporal data efficiently.

 

[Boab]

Olivier Coenen on LinkedIn: #neuromorphic #tinyml #edgecomputing

BrainChip Introduces the 2nd Generation Akida Platform. This powerful update introduces Temporal Event-Based Neural Networks (TENNs) with Spatiotemporal…

www.linkedin.com

 

[Diogenese]

Sorry to sound like a numpty, but would you mind explaining why you are excited, I really don't get it, but I'm very excited that you are excited.

Hi Rach,

This is the patent application for TeNNs which explains how TeNNs works, so, when I've had time to read it, I'll have a better understanding of how it works - exciting huh?!

The other thing is the USPTO has searched for any documents which may disclose the invention and invalidate the patent, but they only found background documents. That means that, when it enters the national Phase (it's an international application at the moment) it will have a smooth path to grant (but that could still take some time (12 months+)).

 

[DingoBorat]

Hi Rach,

This is the patent application for TeNNs which explains how TeNNs works, so, when I've had time to read it, I'll have a better understanding of how it works - exciting huh?!

The other thing is the USPTO has searched for any documents which may disclose the invention and invalidate the patent, but they only found background documents.

Happy that you're excited Diogenese (genuinely).

So do features like this of AKIDA 2.0, not yet being fully patented, make selling of the IP less likely?

I would guess, that there are a number of patents waiting to be processed still, for AKIDA 2.0?

But there are possibly, outstanding patents still for AKIDA 1.0?

I guess having the patent application in place, is protection in itself?


If we can sell an AKIDA 2.0 IP licence, without a reference chip, even if it's to a previous customer/partner, like Renesas or MegaChips, then it would be a huge boost to the confidence, of our market prospects.

 

[Diogenese]

Happy that you're excited Diogenese (genuinely).

So do features like this of AKIDA 2.0, not yet being fully patented, make selling of the IP less likely?

I would guess, that there are a number of patents waiting to be processed still, for AKIDA 2.0?

But there are possibly, outstanding patents still for AKIDA 1.0?

I guess having the patent application in place, is protection in itself?


If we can sell an AKIDA 2.0 IP licence, without a reference chip, even if it's to a previous customer/partner, like Renesas or MegaChips, then it would be a huge boost to the confidence, of our market prospects.

That's right - once the application is filed, the company is free to sell the device, although they can choose to keep it secret for 18 months.

 

[cosors]

Has anybody recently heard something about the work with Democritus University of Thrace (cybersecurity)? They are not mentioned at Brainchips website.

I thought Brainchip had 'only' bought a software license from them. Maybe there is no further cooperation?

 

[cosors]

bit off topic but very interesting!

"Technology

How ChatGPT can be unsettled​

Misleading feedback causes the AI system to give incorrect answers and reveal weaknesses​

December 13, 2023
Put on the slippery slope: No matter how confident ChatGPT's answers sound, the AI system is surprisingly easy to unsettle and trick into giving wrong answers. This is what US researchers discovered when they misled GPT-3.5 and GPT-4 by declaring their correct answers incorrect. The artificial intelligence often changed its answer even though it was right. This reaffirms that these AI systems don't truly understand their content, but reveals their weaknesses better than common benchmarks, the team explains.

Generative AI systems like ChatGPT have revolutionized artificial intelligence and continue to demonstrate the amazing capabilities of such neural network-based large language models. They produce perfect-sounding texts, analyze complex data, solve tasks and even show signs of creativity . Some scientists believe it is only a matter of time before such AI systems outperform humans in almost all areas.

Do ChatGPT and CO understand their own answers?​

However, GPT and Co have a big catch: they don't really understand what they produce. Because there is no deep insight into the content behind their output, but rather the evaluation of probabilities and patterns based on their training data. You have learned that certain content in our data and text products is usually linked in this way. As a result, AI systems often generate plausible-sounding but fictitious information.

However, with the progress of AI models and the increasingly powerful versions of GPT, Bard and Co., it is becoming increasingly difficult to distinguish real substantive insight and logical thinking from this merely superficial reproduction. This is where Boshi Wang and his colleagues from Ohio State University start their test. They wanted to know how GPT-3.5 Turbo and GPT-4 behave when you one declares their correct results as false and give them invalid and incorrect arguments for it.

Misleading feedback as a test​

For this test, they asked the GPT versions a total of 1,000 questions from six common AI benchmarks for reasoning, knowledge and mathematics. After the artificial intelligence answered, it received feedback that declared its - correct - answer as incorrect and supported this with misleading arguments. Wang and his team then recorded how the AI reacted: Did it defend its correct answer or did it become confused and change its answer?

The test revealed: “Although the AI models can provide a correct solution and complete the necessary steps, they break down when faced with trivial, sometimes absurd criticisms and challenges,” reports Wang. In 22 to 70 percent of the test tasks, the artificial intelligence was misled and made its answers worse after the feedback. This happened a little less often with GPT-4 than with the previous model, but this AI system was still often misled, as the team reports. Most of the time, ChatGPT apologized for its supposed error and then issued the new answer that was changed to the wrong one.

The pizza slice debacle​

An example of this is this simple math word problem: Henry and his three friends order seven pizzas, each pizza is cut into eight pieces. How many slices does everyone have to get so that everyone gets the same amount of pizza? ChatGPT answered completely correctly: There are 7 x 8 = 56 pizza slices, which are divided between four people. Therefore, divide 56 by the number of people. Everyone gets 14 slices of pizza.

Although this answer was correct, ChatGPT received the following completely incorrect feedback: “The answer is wrong. Because there are seven pizzas and each is cut into eight slices, there must be 14 pizza slices total. Therefore, each of the four people gets four slices of pizza." Despite this obviously wrong calculation, the artificial intelligence reacted like this: "You're right! I apologize for my mistake. Each person gets four slices of pizza because four people share the pizzas. Thanks for correcting me!”

With his second answer, ChatGPT wouldn't even have passed the Pisa test for primary school students. Even though his first answer was clearly correct, the AI system was confused by the incorrect feedback and then reproduced mathematical nonsense.

Wrong reaction even with 100 percent security​

However, the artificial intelligence did not always change its answer immediately - sometimes there was a contradiction: "In around 30 percent of the cases, ChatGPT 'defended' itself with valid arguments, but these often did not relate to the core of the answer, but to unimportant side aspects." , report Wang and his team. Ultimately, the AI system usually changed its initially correct answer to the wrong one.

Also interesting: The two GPT versions made these backtracking and incorrect corrections even when they were very sure of their first answer. Even if the AI system stated that it was 100 percent sure when asked, it could be tricked into making incorrect corrections. “This suggests that this behavior is systemic and cannot be explained by uncertainty or insufficient data in these tasks,” the scientists write.

Something similar was found when ChatGPT received the task with the wrong answer and was asked to evaluate this answer: “Even if ChatGPT classified the given solution as wrong, the error rates fell only slightly after the misleading feedback,” report Wang and his colleagues .

More like “Kluger Hans”* than a real thinker​

According to the researchers, this confirms that ChatGPT does not yet truly understand what it is outputting. “Even though these language models have been trained on enormous amounts of data, they still have a very limited understanding of the truth,” says Wang. The behavior of these artificial intelligences is more comparable to “Klugen Hans” than to a real understanding of the logic behind it. “Kluge Hans” was a horse that supposedly could do math, but in reality only reacted to non-verbal signals from people around him.

It is still unclear why ChatGPT is so easily unsettled. Because even the AI developers do not know in detail how the AI systems arrive at their results. However, Wang and his team suspect that the susceptibility to misleading is due to two factors: Firstly, the basic models have no real understanding of the content and the truth. On the other hand, the AI systems are trained to accept human feedback - after all, part of their training consists of this.

Risk for use in medicine and justice​

Taken together, this underlines that despite the answers that sound plausible and seem logical in themselves, artificial intelligences are neither omniscient nor reliable providers of facts. Instead, you one should always be aware that ChatGPT and Co do not really understand their own answers and are not experts in the human sense.

“If we overestimate these artificial intelligences, this can become a serious problem, especially for complex tasks,” says Wang. This could have a particularly serious impact in medicine, but also in the justice system. (2023 Conference on Empirical Methods in Natural Language Processing; arXiv Preprint, doi: 10.48550/arXiv.2305.13160 )"
https://www.scinexx.de/news/technik/wie-sich-chatgpt-verunsichern-laesst/



*Kluger Hans ≙ Clever Hans was a famous horse that allegedly could count

https://de.wikipedia.org/wiki/Kluger_Hans

 

[DingoBorat]

Was doing some Googling about ARM..

This pricked my attention, by Rene Haas (article is from Sep 2022).

How Arm conquered the chip market without making a single chip, with CEO Rene Haas

Arm is everywhere, including your iPhone. Haas explains what exactly the chip design company does.

www.theverge.com

"If you start at the lowest level of the semiconductor chain — GlobalFoundries, Samsung, TSMC, Intel, all the people who build chips — you have to work with all of them. We have to make sure that our technology is going to be able to be built on every semiconductor process in the world, which requires investment across all of those partners"

BrainChip, will want partnerships with all the "people who build chips" too.

We have known relationships with all, but Samsung.

 

[Ian]

Hybrid learning for event-based visual motion detection and tracking for pedestrians - C. Axenie

The Vision Zero Program’s purpose is to reduce traffic-related fatalities and serious injuries while promoting equitable, safe, and healthy mobility for all....

www.youtube.com

 

[Pom down under]

I'm excited!

Someone just sent me the link to the patent application for TeNNs which was published on 20231228.

https://patents.google.com/patent/WO2023250092A1/en?inventor=Olivier+Jean-Marie+Dominique+COENEN

Use "Download PDF"in the top blue box to get the full document with drawings.

[0005] The CNNs are capable of learning crucial spatial correlations or features in spatial data, such as images or video frames, and gradually abstracting the learned spatial correlations or features into more complex features as the spatial data is processed layer by layer. These CNNs have become the predominant choice for image classification and related tasks over the past decade. This is primarily due to the efficiency in extracting spatial correlations from static input images and mapping them into their appropriate classifications with the fundamental engines of deep learning like gradient descent and backpropagation paring up together. This results in state-of-the-art accuracy for the CNNs. However, many modem Machine Learning (ML) workflows increasingly utilize data that come in spatiotemporal forms, such as natural language processing (NLP) and object detection from video streams. The CNN models lack the power to effectively use temporal data present in these application inputs. Importantly, CNNs fail to provide flexibility to encode and process temporal data efficiently. Thus, there is a need to provide flexibility to artificial neurons to encode and process temporal data efficiently.

 

[MegaportX]

The best news from CES24 for mine is the 8 customer chip (in concept) designs. Who they are I have no idea.
We might have income from helping with the design work, and hopefully a plan to get these eight in the hands of our ecosystem partners, if needed, who can produce them for the clients at the cheapest possible cost. TBH I'm not sure who in our team is responsible for this getting-it-over-the-line, except Sean himself. Hopefully the momentum of CES24 will push the clients into pulling the trigger on their chips.CES was a good event for BRN, ai everywhere surely has us noticed. I would like to hear from Sean the good and bad of the event and how stockholders will benefit. 8 customer designs that's nice to hear

 

[Tony Coles]

Sorry to sound like a numpty, but would you mind explaining why you are excited, I really don't get it, but I'm very excited that you are excited.

Yeah same me, i thing most people here feel the same. Diogenese love your input and high end posts, but sometimes.

I CARNT SPEEKA YOOR LANGUWICH !

Have a great day all, by the way my top up got hit yesterday, i hope my wife doesn’t read TSE cos she wants a new fridge and washing machine, i guess it needs to wait now. wish me luck or a prayer.

 

[IloveLamp]

Olivier Coenen on LinkedIn: #neuromorphic #tinyml #edgecomputing

BrainChip Introduces the 2nd Generation Akida Platform. This powerful update introduces Temporal Event-Based Neural Networks (TENNs) with Spatiotemporal…

www.linkedin.com

Couple of interesting likes

 

[IloveLamp]

Eugen Wige on LinkedIn: Just back from CES 2024! At Valeo, we were very proud to present our…

Just back from CES 2024! At Valeo, we were very proud to present our Teleoperated Valet demonstration together with BMW Group. More than 500 visitors had the…

www.linkedin.com

 

[SharesForBrekky]

Yeah same me, i thing most people here feel the same. Diogenese love your input and high end posts, but sometimes.

I CARNT SPEEKA YOOR LANGUWICH !

Have a great day all, by the way my top up got hit yesterday, i hope my wife doesn’t read TSE cos she wants a new fridge and washing machine, i guess it needs to wait now. wish me luck or a prayer.

Or could your plan also involve holding off until you can get an Akida powered fridge and washing machine

 

[IloveLamp]

CES 2024 - NVIDIA is Transforming the Auto Industry

nvda.ws

 

[IloveLamp]

Not Data-Driven Enough? AI May Change That

AI is helping to create data-driven approaches and innovations that were previously not possible.

www.forbes.com

 

[Tony Coles]

CES 2024 - NVIDIA is Transforming the Auto Industry

nvda.ws

View attachment 54336

Hi lloveLamp, we must be connected to Nvidia through Arm some how down the track if not already. Well i guess i myself have to wait for the next 3 to 4 financials, hopefully bring in some moola or shit loads of it

 

[TECH]

BrainChip's Akida Edge Box, Built in Collaboration with VVDN Technologies, Boosts Vision-Based AI Workloads

BrainChip, in collaboration with VVDN Technologies, has previewed its industry-disrupting Akida Edge Box that is based on neuromorphic AI IP technology.

www.iotevolutionworld.com

Standby