It is true that identity fraud has become a buzz word among people and experts, but not in a good way. First coined in 1964, identity theft has since turned prolific, whether it is fraudsters seeking to impersonate others in order to either open a credit card account or gangs laundering money without linking transactions to their real-life identities.

Last year, The National Fraud Intelligence Bureau reported that fraud offences increased by 17% in March 2022, reaching 936,276 compared to the year ending March 2021.

A GlobalData survey conducted in September 2022 stated identity theft came among the top three concerns shared among citizens from the UK, US, France, Germany and Poland. Identity theft, as a practice.

Nir Stern is vice president of product management at AU10TIX, a tech company providing intelligence information and the infrastructure needed to combat fraud.

In an interview with Electronic Payments International, Stern talks about how machine learning, data consortiums as well as external sources can effectively be used to combat identity fraud.

Generally speaking, in the world of financial crime and even beyond, you need to differentiate between account takeover and identity fraud.

In the case of an account takeover, as a consumer, you have an existing relationship with a financial institution, and someone is taking over that account by stealing your credentials (username, password or even one-time password).

This is usually done via social engineering or email scams. In the end, a fraudster tries to make the victim do things for them or provide them with information that they need without the victim knowing about the fraud.

With identity fraud, the purpose usually is to pretend to be another person to start a new interaction with the institution either to commit financial crimes or to accomplish other activities. One example could be money laundering to finance terrorist activity.

Because of regulations, you would need an identity with all relevant information. The methods employed would be different.

First, you need to have access to your victims personal information unless you want to create a synthetic ID.

Secondly, you must create a fake ID that is good quality enough to bypass any security measurements financial institutions have.

There are roughly three types of methods.

There is the very elementary and nave one, like trying to take a photo of the victims ID or download an ID image.

Then there are the more sophisticated fraud techniques. You could go on certain websites, pay a certain amount of money and download any type of digital document. Those should be of sufficient quality so that when you hand them to an expert, they will not notice anything suspicious.

Then you have the top quality fraud, where identity fraudsters steal or buy personal information from databases uploaded on the dark web. Then, they create high-quality, sometimes even physical IDs that are almost impossible to detect with all the information or standard methods of going through fraud. Everything will look perfectly normal because the fraudsters have accessed information through a particular type of data breach.

Those are the trickiest to detect because everything looks completely normal. And when we examine web traffic, we see the most sophisticated fraudsters use this.

So, because fraudsters are using the methods mentioned above, none of the standard ID verification measures employed by banks will be helpful, no matter how sophisticated they are.

The only way to detect it with systems like ours is first to use machine learning that has access to millions of legitimate transactions and can spot different indicators showing what counts as fraudulent activity.

Also, we have a unique system called Instinct, a concerted database where we track transactions worldwide. Through our big customers, we keep an eye on a huge amount of legitimate and fraudulent transactions. We securely store that information, avoiding any storage of sensitive data.

These steps enable us to see repetitions. Because when

these fraudsters invest a lot of time, effort, and money in coming up with a specific tactic, they will not do it just once. They will perform mass attacks.

And then we see cases where everything will look perfectly normal if you look at a single transaction. Still, our systems can detect the same face used on multiple IDs or the same ID with different faces.

The combination of sophisticated machine learning to determine indicators of fraudulent activity and using a data consortium is basically the only way to detect those attacks. Unfortunately, many organisations do not use such capabilities to prevent identity fraud, thereby risking their money.

So, the main differentiator is, first, the fact that many of our competitors and ID-proofing solutions need to provide manual reviews behind the scenes. They will capture the information or images sent over to them. But most of the analysis will be done by human beings. And, as I have mentioned, with all these sophisticated, super complex forgeries, a human being cannot identify ID fraud because the activity looks normal even when you go into detail with your examination.

Thats why first, you need to have a fully automated system based on machine learning/AI to detect those tiny indicators of forgeries that are hard to see otherwise.

Moreover, you also need the capabilities like we do with Instinct to look through a data consortium, detect synthetic fraud or mass attacks, and see whether those attacks are repetitive.

You dont bring a knife to a gunfight. That is probably the key element in fighting identity fraud. If you look at the reports that are always shared, the value of identity theft-related fraud is hundreds of millions. You see specific gangs of identity thieves making millions out of it. As a result, they invest a lot of money into acquiring new technology to commit fraud. So you need to have the equivalent of that technology to fight.

Therefore, you cannot have a solution based on human beings/agents that are supposed to detect identity fraud. It is almost impossible for them.

Next, on top of having a fully automated system, you also need to have a multi-layered approach.

That means not only searching to pinpoint fraudulent activity but also having access to data consortiums, as well as looking into external data sources to determine whether the fraudulent activity is one-off or not.

Thirdly, fraudsters adapt all the time, so you need a fully adaptable system that keeps on learning, a system with access to information not limited to your customers. If you only monitor your customer environment, when the next attack comes, they might not be prepared for it unless they have enough experience.

The other side of this discussion we should have brought up is the legitimate user experience.

If you think about these banks, they want to be secure and protected from fraud. Around 99.9% of their traffic consists of legitimate persons. In that sense, you want to ensure they have the best user experience, especially when you are a digital bank. That is your only channel; you dont have a backup physical branch.

You need to ensure you have a smoother digital user experience and make it secure and comfortable for your customers.

The key for these companies is to have the back-end systems able to detect these forgeries and the front-end solutions that help you, as a legitimate user, take the best profile photos and quickly become a customer.

When using these front-end solutions, make sure that the quality of customer images enables you, as a company, to detect ID theft and prevent fraudsters from using things like external cameras to create deep-fake photos, as well as upload images, or screenshot pictures of people or their IDs.

So all of that, combining a sophisticated front-end solution and a sophisticated back-end solution, is the direction many digital banks and financial institutions take and what traditional banks should do more as well.

When doing ID or digital verification, if you are not a small local bank but a global one willing to extend to other countries, you need companies like AU10TIX to do it for you.

We support over 200 countries and over 4,000 types of IDs. It is virtually impossible for a bank to develop a system that can manage various national ID formats while also successfully detecting forgeries.

Thats not their business. Their business is to give the best user experience, so I believe the best approach to combat fraud is for them to get expert help.

The riskiest thing about blockchain or decentralised financial transactions is that you dont have any indication of where the money goes, no way to turn it back, and no information about who is interacting with the system. So in the case of any account takeovers or social engineering acts that happen, there is nothing you can do.

There are different transaction monitoring techniques for decentralised IDs. A lot of times, these techniques make sure the money doesnt go to unknown wallets. But if its something like a fraudster opening the digital wallet and stealing the money there is very little you can do.

One of the trends we see, and were strategically investing a lot in it, is decentralised ID. It is when you issue a digital ID, which is encrypted and tokenised on your mobile device or digital wallet. That information is not shared with anyone; the signature is kept in the blockchain.

Because it is based on these standards, if you need to use the identification system, you can just share simple information or claims like am I allowed to do it? Am I over 18? Am I a citizen of a particular country?

We see a lot of hype around this technology. Because its all around self-survey identity and significantly related to data privacy, which is key to decentralised finance, the combination of the two may actually be a solution that benefits everyone. You can keep your privacy without the need to be part of the financial institution ecosystem while still making it secure in a way that guarantees only you have access to it.

Continued here:
Machine learning and data consortium can fight identity theft: AU10TIX - Electronic Payments International

Computer vision (CV) has been widely adopted in many Internet of Things (IoT) devices across various use cases, ranging from smart cameras and smart home appliances to smart retail, industrial applications, access control and smart doorbells. As these devices are constrained by size and are often battery powered, they need to wield highly efficient compute platforms.

One such platform is the MCU (microcontroller unit), which has low-power and low-cost characteristics, alongside CV and machine learning (ML) compute capabilities.

However, running CV on the MCU will undoubtedly increase its design complexity due to the hardware and software resource constraints of the platform.

Therefore, IoT developers need to determine how to achieve the required performance, while keeping power consumption low. In addition, they need to integrate the image signal processor (ISP) into the MCU platform, while balancing the ISP configuration and image quality.

One processor that fulfills these requirements is Arms Cortex-M85, which is Arms most powerful Cortex-M CPU to date. With vector extension of SIMD (single instruction, multiple data) 128-bit vector processing, Cortex-M85 accelerates CV alongside the overall MCU performance. For IoT developers, they can leverage Arms software ecosystem, ML embedded evaluation kit and guidance on how to integrate the ISP with the MCU in order to unlock CV and ML easily and quickly on the highly-efficient MCU platform.

As a first step, being able to run CV compute workloads requires improved performance on the MCU. Focusing on the CPU architecture, there are several ways to enhance the MCUs performance, including superscalar, VLIW (very long instruction word), and SIMD. For the Cortex-M85, Arm chose to adopt SIMD which is a single instruction set that can operate multiple data as its the best option for balancing performance and power consumption.

Figure 1: The comparison between VLIW and SIMD

Arms Helium technology, which is the M-Profile Vector Extension (MVE) for the Cortex-M processor series, brings vector processing to the MCU. Helium is an extension in the Armv8.1-M architecture to significantly enhance performance for CV and ML applications on small, low-power IoT devices. It also utilises the largest software ecosystem available to IoT developers, including optimised sample code and neural networks.

Supporting the Cortex-M CPUs, Arm has published various materials to make it easier to start running CV and ML. This includes the Arm ML embedded evaluation kit.

The evaluation kit provides ready-to-use ML applications for the embedded stack. As a result, IoT developers can experiment with the already-developed software use cases and then create their own applications. The example applications with ML networks are listed in the table below.

The Arm ML embedded evaluation kit

The ISP is an essential technology to unlock CV, as the image stream is the input source. However, there are certain points that we must consider when integrating ISP on the MCU platform.

For IoT edge devices, there will be a smaller image sensor resolution (<1-2MP; 15-30fps) and even lower frame rate. Also the image signal processing is not always active. Therefore, using a higher quality scaler within the ISP will drop the resolution to sub-VGA, which is 640 x 480, to, for example, minimise the data ingress to the NPU. This means that the ISP only uses the full resolution when needed.

ISP configurations can also affect power, area, and efficiency. Therefore, it is worth asking the following questions to save power and area.

Whether its for human vision, computer vision, or both?

What is the required memory bandwidth?

How many ISP output channels will be needed?

An MCU platform is usually resource-constraint with limited memory size. Integrating with an ISP requires the MCU to run the ISP driver, including the ISPs code, data, and control LUT (loop up table). Therefore, once the ISP configuration has been decided, developers need to tailor the driver firmware accordingly, removing unused code and data to accommodate the memory limitation on the MCU platform.

Figure 2: An example of concise ISP configuration

Another consideration when integrating the ISP with the MCU is lowering the frame rate and resolution In many cases, it would be best to consider the convergence speed of the 3As auto-exposure, auto-white balance and auto-focus. This will likely require a minimum of five to ten frames before settling. If the frame rate is too slow, it might be problematic for your use case. For example, this could mean a two to five second delay before a meaningful output can be captured and, given the short power-on window, there is a risk of missing critical events. Moreover, if the clock frequency of the image sensor is dropped too low, it is likely to introduce nasty rolling shutter artifacts.

Enabling CV and ML on MCU platforms is part of the next wave of the IoT evolution. However, the constraints of the MCU platform can increase the design complexity and difficulty. Enabling vector processing on the MCU through the Cortex-M85 and leveraging Arms software ecosystem can provide enough computing and reduce this design complexity. In addition, integrating a concise ISP is a sensible solution for IoT devices to speed up and unlock CV and ML tasks on low-power, highly efficient MCU platforms.

Here is the original post:
Unlocking computer vision and machine learning on highly efficient ... - Imaging and Machine Vision Europe

When Canadian artificial intelligence company Darwin AI was founded in 2017, machine learning and deep learning were still relatively new terms. In the past five years, CEO Sheldon Fernandez and his team have been working with this technology to develop some foundational IP to simplify implementation. About a year ago, Sheldon took a part happenstance, part deliberate opportunity to develop a vertical offering for EMS manufacturing. Heres what happened.

Sheldon, its nice to meet you. Would you briefly introduce your company?

Sheldon Fernandez:Were based out of Waterloo, Ontario, Canada. We're organically connected to the University of Waterloo, which is kind of like Canada's MIT. Two of our co-founders are professors at the institution, including Professor Alexander Wong, Canadas Research Chair in AI and Medical Imaging.

Weve been working on foundational machine learning and deep learning technology for the past five years. A couple of years ago, our large industrial and aerospace clients were telling us about their supply chain challenges during the pandemic and reshoring sensitive electronics manufacturing work back to North America, specifically printed circuit boards (PCB). We thought that created an opportunity for us.

When we looked at PCB manufacturing, it became apparent that while the SMT placement workflow was highly automated, there was a need in automating back-end production and final assembly. This laborious part of the process was where EMS companies and OEMs were still employing manual inspection. These inspection tasks are tough to crack from a traditional machine vision perspective, and we wondered, Can AI bring anything to bear on this problem? We spent about a year developing a hardware and software solution which fits into the typical assembly line for PCB manufacturing. It also does post-assembly analysis, and what's really fascinating is how quickly an operator can program our product.

We often hear that AOIs are good at what they do but are laborious to program and maintain. With our system, theres not a lot of manual work. You give the system a good (i.e., golden) boardor a couple of good boards if there's a union of different componentsand our AI system creates a map of where components should be in less than a minute; away you go. You can tweak it after that, and its striking how quickly you can configure the product.

We brought our mini system to IPC APEX EXPO, and the response was fascinating. So many companies were intrigued by finally automating back-end production, and we're really excited about becoming a part of the community.

To read this entire conversation, which appeared in the March 2023 issue of SMT007 Magazine,click here.

More here:
Darwin AI Evolving the Islands of Automation - SMT 007

Recent achievements in supervised tasks of deep learning can be attributed to the availability of large amounts of labeled training data. Yet it takes a lot of effort and money to collect accurate labels. In many practical contexts, only a small fraction of the training data have labels attached. Semi-supervised learning (SSL) aims to boost model performance using labeled and unlabeled input. Many effective SSL approaches, when applied to deep learning, undertake unsupervised consistency regularisation to use unlabeled data.

State-of-the-art consistency-based algorithms typically introduce several configurable hyper-parameters, even though they attain excellent performance. For optimal algorithm performance, it is common practice to tune these hyper-parameters to optimal values. Unfortunately, hyper-parameter searching is often unreliable in many real-world SSL scenarios, such as medical image processing, hyper-spectral image classification, network traffic recognition, and document recognition. This is because the annotated data are scarce, leading to high variance when cross-validation is adopted. Having algorithm performance sensitive to hyper-parameter values makes this issue even more pressing. Moreover, the computational cost may become unmanageable for cutting-edge deep learning algorithms as the search space grows exponentially concerning the number of hyper-parameters.

Researchers from Tsinghua University introduced a meta-learning-based SSL algorithm called Meta-Semi to leverage the labeled data more. Meta-Semi achieves outstanding performance in many scenarios by adjusting just one more hyper-parameter.

The team was inspired by the realization that the network may be trained successfully using the appropriately pseudo-labeled unannotated examples. Specifically, during the online training phase, they produce pseudo-soft labels for the unlabeled data based on the network predictions. Next, they remove the samples with unreliable or incorrect pseudo labels and use the remaining data to train the model. This work shows that the distribution of correctly pseudo-labeled data should be comparable to that of the labeled data. If the network is trained with the former, the final loss on the latter should also be minimized.

They defined the meta-reweighting objective to minimize the final loss on the labeled data by selecting the most appropriate weights (weights throughout the paper always refer to the coefficients used to reweight each unlabeled sample rather than referring to the parameters of neural networks). The researchers encountered computing difficulties when tackling this problem using optimization algorithms.

For this reason, they suggest an approximation formulation from which a closed-form solution can be derived. Theoretically, they demonstrate that each training iteration only needs a single meta gradient step to achieve the approximate solutions.

In conclusion, they suggest a dynamic weighting approach to reweight previously pseudo-labeled samples with 0-1 weights. The results show that this approach eventually reaches the stationary point of the supervised loss function. In popular image classification benchmarks (CIFAR-10, CIFAR-100, SVHN, and STL-10), the proposed technique has been shown to perform better than state-of-the-art deep networks. For the difficult CIFAR-100 and STL-10 SSL tasks, Meta-Semi gets much higher performance than state-of-the-art SSL algorithms like ICT and MixMatch and obtains somewhat better performance than them on CIFAR-10. Moreover, Meta-Semi is a useful addition to consistency-based approaches; incorporating consistency regularisation into the algorithm further boosts performance.

According to the researchers, Meta-Semi requires a little more time to train is a drawback. They plan to look into this issue in the future.

Check out thePaperandReference Article.All Credit For This Research Goes To the Researchers on This Project. Also,dont forget to joinour 15k+ ML SubReddit,Discord Channel,andEmail Newsletter, where we share the latest AI research news, cool AI projects, and more.

Tanushree Shenwai is a consulting intern at MarktechPost. She is currently pursuing her B.Tech from the Indian Institute of Technology(IIT), Bhubaneswar. She is a Data Science enthusiast and has a keen interest in the scope of application of artificial intelligence in various fields. She is passionate about exploring the new advancements in technologies and their real-life application.

Read the original post:
Researchers From Tsinghua University Introduce A Novel Machine Learning Algorithm Under The Meta-Learning Paradigm - MarkTechPost

Machine learning development startup Weights & Biases Inc., whose software is used by the likes of OpenAI LLC and Nvidia Corp. to develop new artificial intelligence models, announced two major enhancements to its platform today.

Weights and Biases has created a platform for teams to build and collaborate on machine learning models and operations, or MLOps. The platform enables teams to keep track of their machine learning experiments. It also provides tools for evaluating the performance of different machine learning models, dataset versioning and pipeline management.

The W&B platform is designed to improve the efficiency of the trial-and-error process through which AI software is developed. According to the startup, it helps by increasing developer productivity, as it solves one of the main challenges of AI initiatives: organizing and processing project data.

The new additions to its platform were announced at Fully Connected, W&Bs inaugural user conference, and include W&B Launch and W&B Models.

Available in public preview starting today, W&B Launch provides an easier way for users to package code automatically and launch a new machine learning job on any target environment. In this way, machine learning practitioners gain easier access to the compute resources they need, while simplifying infrastructure complexity. In addition, W&B Launch makes it easier for teams to reproduce runs and scale up and scale out those activities.

Orlando Avila-Garcia, AI principal researcher at ARQUIMEA Research Center, said W&B Launch will make it easier for his organization to accelerate research on a variety of deep learning techniques, including neural radiance fields and graph neural networks. Abstracting away the complexity of using the infrastructure for our researchers is very beneficial to our overall team, he said. Launch greatly simplifies our work optimizing, experimenting with and benchmarking ML methods, letting us focus on reliability and reproducibility of results.

As for W&B Models, this is generally available now and provides a more scalable way for teams to govern machine learning model lifecycles in a centralized repository, while allowing for cross-functional discovery and collaboration. With its reproducibility and lineage tracking functionality that makes it possible for users to track exactly when a model was moved from staging to production, W&B Models can help teams to maintain higher-quality models over time.

Andy Thurai, vice president and principal analyst of Constellation Research Inc., told SiliconANGLE that Weights & Biases is one of the older players in the MLOps segment, and its especially strong inmodel creation, data set versioning and experiment tracking. He added that W&B Models looks like it will be a good addition to the companys solution set. It offers ML model registry and model governance, allowing users to collaborate more easily by making models discoverable across large enterprises, he said. This, combined with model lineage tracking, enables users to track models in the ML pipeline from inception to production.

Were confident that W&B users will quickly see the benefits of Launch and Models in accelerating model training, utilizing compute resources efficiently, managing models with more confidence, and having a more cohesive end-to-end ML workflow, said W&B Vice President of Product Phil Gurbacki.

The platform enhancements follow what W&B says has been a very promising 12 months, during which it has enjoyed significant traction and momentum. The company, which last raised $135 million at a $1 billion valuation in October 2021, claims to have doubled its employee base in the last year, opening new offices in Berlin, London and Tokyo to help boost its international presence.

See more here:
Weights & Biases simplifies machine learning production and ... - SiliconANGLE News