BT/ Apple granted biometrics patents for palm recognition and wearables button

Paradigm

Published in

Paradigm

25 min readDec 23, 2020

Biometrics biweekly vol. 1, 7th December — 23rd December

TL;DR

A pair of researchers from the Idiap Research Institute have developed a metric to use in assessing the fairness of biometric systems, or its lack in providing disparate results in matching people based on gender and skin color.
A recent study from two University of Texas at Dallas researchers and their two colleagues outlined the underlying factors that contribute to race-based deficits in facial recognition accuracy.
A single strand of hair in a crime scene contains many clues that can help identify a perpetrator. In a recent study, scientists have combined two modern techniques, called surface-enhanced Raman spectroscopy and X-ray fluorescence, to distinguish between different colors in individual hair strands. Both these techniques are almost non-destructive and can be conducted with portable devices, making this a promising way to get supportive evidence in forensic investigations.
Apple granted biometrics patents for palm recognition and wearables button: Palm ID could be the next popular biometric authentication method for consumer electronics, as an Apple patent for a contactless palm recognition system, which could measure palm prints or veins, has been granted to the tech giant by the U.S Patent and Trademark Office (USPTO). Apple has also been granted a patent by the USPTO for a fingerprint biometric system for a wearable device, which could be a Touch ID button on an Apple Watch.
IARPA launches tender opportunity for ‘whole-body’ biometric identification from a distance: A new contract opportunity for ‘whole-body’ biometric technology to identify people from long range and elevated platforms like watchtowers or drones has been posted by the U.S. Office of the Director of National Intelligence.
The global market for biometric systems will nearly double to be worth $68.6 billion in 2025, up from an estimated $36.6 billion this year for a 13.4 percent CAGR during the forecast period, according to a report from MarketsandMarkets.
Government biometrics initiatives to boost $18B market growth.
2.5M biometric payment cards in 2021 could be first step for future ‘de facto payment type’
World Health Access (WHA) is looking to provide people with secure immunity credentials in the wake of COVID-19.
Biometric election solution providers pitch Philippines on remote online voting systems.
Passwordless digital identity solution launched to Australia’s myGov platform.
Mexico approves rollout of national biometric digital ID card.
India mulls digital voter ID before assembly elections next year.
Nigeria wants national ID numbers of mobile phone users added to their SIM cards.
Selfd.id launches first decentralized digital identity platform in Romania.
CyberLink’s FaceMe 1:N biometric accuracy among top-ranked by NIST.
Remark Holdings launches edge face biometrics with Nvidia module integration.
Biometrics industry events. And more!

Biometrics market

The Biometric system market size is projected to grow from USD 36.6 billion in 2020 to USD 68.6 billion by 2025; it is estimated to grow at a CAGR of 13.4% during the forecast period. Increasing use of biometrics in consumer electronic devices for authentication and identification purposes, the growing need for surveillance and security with the heightened threat of terrorist attacks, and the surging adoption of biometric technology in automotive applications are the major factor propelling the growth of the biometric system market.

Biometric Research & Development

Latest Researches:

Surface-enhanced Raman Scattering and X-ray Fluorescence Analyses of a Single Hair Colored with a Hair Dye Product

by Momona HORIGUCHI, Shinsuke KUNIMURA in Analytical Sciences

In crime scene investigations, a single strand of hair can make a huge difference in the evolution of a case or trial. In most cases, forensic scientists must look for clues hidden in minuscule amounts of substances or materials found at crime scenes. If a fallen strand of hair with root cells attached is found, a DNA test can reveal the identity of a criminal; unfortunately, this seldom happens. Even though other types of DNA analysis can be conducted using the “mitochondrial DNA” embedded in the hair shaft itself, such tests are not sufficient to reliably identify a person and usually call for additional evidence.

But what if a bit of fashion consciousness could inspire a new forensic technique? In a recent study published in Analytical Sciences, scientists at the Tokyo University of Science, Japan, developed a strategy for identifying criminals from a single strand of hair, leveraging the fact that hair dyes are becoming increasingly common. Their approach involves finding out if two individual strands of hair belong to the same person based on the composition of hair dye products found on them. To do this, they employed two well-known analytical methods: surface-enhanced Raman spectroscopy (SERS) and X-ray fluorescence (XRF) analysis.

Raman spectroscopy is an analytical technique based on the physical phenomenon of Raman scattering, which models certain energetic interactions that occur when photons collide with matter. SERS is a special type of Raman spectroscopy that provides a “structural fingerprint” of a material even when very few molecules are present in the target sample. On the other hand, XRF analysis involves irradiating a material with X-rays and examining the energies of photons re-emitted when the electrons in the sample leave the excited states. XRF analysis is especially useful to determine which metallic elements are present in a material.

The scientists conducted SERS and XRF analyses using portable devices to see if they could distinguish between single strands of hog hairs dyed with different products. Associate Professor Shinsuke Kunimura, who led the study, explains why both analytical methods had to be used in combination, “SERS can easily detect the overall differences in composition between different types of hair dyes, such as permanent, semi-permanent, or natural dyes. However, it is not enough to distinguish between hair coloring products that contain or produce similar dyes. To do this, we also relied on XRF analysis, which can detect the presence of metallic elements used in the ingredients of hair dye products.” Using both techniques, the scientists were able to easily distinguish between five different dyes applied to individual strands of hog hair.

Because both analytical methods used are almost non-destructive, the strategy proposed in this study could be used to quickly analyze hairs found in crime scenes on-site before they are sent for DNA analysis. “Our approach provides supportive information for more reliably identifying whose hair was found in a crime scene,” remarks first author Momona Horiguchi. “This could help us clarify if someone is a criminal, meaning that our methodology could greatly contribute to forensic investigations.”

Overall, this study showcases how analytical tools normally used in chemistry and materials science can be creatively adapted to vastly different fields, such as forensic investigations. Hopefully, in the future, it will prevent criminals from escaping by a hair’s breadth!

Fairness in Biometrics: a figure of merit to assess biometric verification systems

by Idiap’s Tiago de Freitas Pereira and Sébastien Marcel

A pair of researchers from the Idiap Research Institute have developed a metric to use in assessing the fairness of biometric systems, or its lack in providing disparate results in matching people based on gender and skin color.

The paper describes the use of reference databases which are intended to represent operational conditions in the benchmarking stage of the machine-learning development pipeline.

The accuracy shown in those benchmarking processes, however, often differs between demographic groups, giving rise to fairness concerns.

The researchers discuss the factors which could go into considering a biometric system to be fair, and propose the use of fairness discrepancy rate (FDR) as a measurement of differences in accuracy. The 11-page paper also includes a case study of FDR using facial recognition.

“Most of the works in the biometrics community assess fairness in verification systems by comparing DET curves, and/or ROC curves of different demographic groups separately,” the report authors write. “This type of comparison assumes that decision thresholds are demographic-specific, which is not feasible in operational conditions and doesn’t proxy statistical separation. FDR addresses that by assessing demographic discrepancies assuming single decision thresholds.”

The source code, trained models, and scores are also made publicly available to enable others to reproduce the work.

Much of the paper details the formula for determining FDR, and arrives at a metric which they say indicates fairness the closer the “area under FDR” is to 1. In an example, they call a finding of 0.999 “fair,” and one of 0.777 “unfair.”

Unsurprisingly, in tests of several systems against three public datasets, some significant imbalances are identified.

Researchers describe profound complexity of biometrics ethics puzzle

As facial recognition technology comes into wider use worldwide, more attention has fallen on the imbalance in the technology’s performance across races. In a study researchers from The University of Texas at Dallas School of Behavioral and Brain Sciences (BBS) outlined the underlying factors that contribute to these deficits in facial recognition accuracy and offer a guide to assessing the algorithms as the technology improves.

Dr. Alice O’Toole, the Aage and Margareta Møller Professor in BBS, is the senior author of the study, which she describes as both “profound and unsatisfying” because it clarifies the scale of the challenge.

“Everybody’s looking for a simple solution, but the fact that we outline these different ways that biases can happen — none of them being mutually exclusive — makes this a cautionary paper,” she said. “If you’re trying to fix an algorithm, be aware of how many different things are going on.”

In a study last year conducted by the National Institute for Standards and Technology (NIST), the government agency found that the majority of facial recognition algorithms were far more likely to misidentify racial minorities than whites, with Asians, Blacks and Native Americans particularly at risk.

As a result of their research, the UT Dallas scientists concluded that while there isn’t a one-size-fits-all solution for racial bias in facial recognition algorithms, there are specific approaches that can improve the technology’s performance.

Psychological sciences doctoral student Jacqueline Cavazos, the study’s lead author, divided the factors contributing to bias into two categories: data-driven and operationally defined. The former influence the algorithm’s performance itself, while the latter originate with the user.

“Data-driven factors center on the most commonly theorized issues — that the training pool of images is in itself skewed,” Cavazos said. “Are the images being used representative of groups? Are the training images of the same quality across races? Or is there something inherent about the algorithms’ computation of face representations different between race groups?”
O’Toole added, “Our discussion of image difficulty for racial bias is a relatively new topic. We show that as pairs of images become more difficult to distinguish — as quality is reduced — racial bias becomes more pronounced. That hasn’t been shown before.”

Cavazos explained that operational bias can be introduced depending on where the threshold is set between matching and nonmatching decisions, and on what kinds of paired images are chosen.

“Our paper confirms what has been shown previously: Where you set the criterion for what is the same identity versus different identities can influence the error rate, and sometimes the same threshold will give you different error rates for different races,” Cavazos said. “Secondly, you need to be sure that when you test an algorithm, pairs of images that are of different identities should always be matched on demographics; — this assures us that identification accuracy is based only on identity. Human participants are shown two images of different people with matching demographics — same race, same gender and so on. If algorithms aren’t also presented in such similar pairs, algorithm performance can appear better than it really is because the machine’s task is easier.”

While the study outlines how racial bias should be evaluated in the use of facial recognition algorithms, the researchers emphasize that no easy solution to the problem exists.

“One of the novel things about this paper is how it brings all of these factors together,” O’Toole said. “Earlier work has centered on individual issues. But you have to look at them all to know the best way to use these algorithms.”

O’Toole believes their research could help users understand which algorithms should be expected to show bias and how potentially to calibrate for that bias.

“For instance, you can measure the performance of an algorithm in a variety of ways. One measure might indicate that the algorithm is race biased, while another might not. Moreover, the algorithm could be biased in a way that you have not explicitly measured,” O’Toole said. “For example, one measure might be directly indicative of whether the algorithm could falsely accuse an innocent person. It might be aimed at determining how similar the people in two images have to appear for the machine to indicate that they are the same person. Another measure might focus on how many correct identifications the algorithm makes. These measures are made from the same algorithm, but they can easily dissociate, pointing to bias in one case and to equitable performance in the other case.”

Main Development News

Apple granted biometrics patents for palm recognition and wearables button:

Palm ID could be the next popular biometric authentication method for consumer electronics, as an Apple patent for a contactless palm recognition system, which could measure palm prints or veins, has been granted to the tech giant by the U.S Patent and Trademark Office (USPTO). Apple has also been granted a patent by the USPTO for a fingerprint biometric system for a wearable device, which could be a Touch ID button on an Apple Watch.

The patent for an ‘Electronic device including palm biometric sensor layer and related methods’ shows an imaging sensor and potentially Apple’s infrared imaging system being used to perform authentication to the device by sensing “surface distortion.” The image sensing layer could include a substrate, possibly with a photodiode layer, and a layer for narrowing the field of view above that. The system could also use a flood light source operable at between 450 nm and 560 nm, instead of infrared light.

The images depict the biometric scanning process, typically with the IR dot projector system associated with the iPhone’s ‘notch,’ but also on a watch.

Another patent for an ‘Electronic device having sealed button biometric sensing system’ describes the use of fingerprint authentication for user identification, device unlocking, application or transaction authorization.

The button would include a biometric sensor and a flexible conduit which transmits an output signal. It sits in an enclosure in the side of the device, and is connected by a seal with a pressure sensitive adhesive layer.

The patent refers most often to a capaaqcitive fingerprint sensor, but also notes that the biometric button could sense another biometric, like DNA or blood pressure.

Low-power platform for edge biometrics launched by Synaptics and Eta Compute:

Synaptics Inc. announced its newest battery-powered edge computing platform Katana Edge AI in partnership with Eta Compute to enable on-device face biometrics. The new platform combines low power SoC architecture with energy-efficient AI software and is aimed at consumer and IoT applications.

Katana can make ordinary surveillance systems capable of object and facial recognition and provides various ultra-low power edge devices for use cases in offices, retail stores, factories, farms, and smartphones, the companies say. Other applications include voice biometrics and sound detection, counting and inventory tracking, as well as environmental sensing.

IARPA launches tender opportunity for ‘whole-body’ biometric identification from a distance:

A new contract opportunity for ‘whole-body’ biometric technology to identify people from long range and elevated platforms like watchtowers or drones has been posted by the U.S. Office of the Director of National Intelligence.

The solicitation, which is made under the Intelligence Advanced Research Projects Activity (IARPA), calls the capability the ‘Biometric Recognition and Identification at Altitude and Range (BRIAR) Program.’

BRIAR Program requirements include the capability to identify people with biometrics from 300 meters or further away, “through atmospheric turbulence,” and with sensor platforms observing subjects from angles greater than 20 degrees. Those capabilities are sought by many intelligence and Department of Defense (DoD) agencies, according to the ‘Broad Agency Announcement’ document.

Facial recognition has become the biometric modality of choice for intelligence and DoD missions with unconstrained conditions. Further, the ability of the technology to deal with constraints in the form of pose, illumination and expression characteristics has been developed significantly, IARPA says, referring to its Janus Program.

Itoco and World Health Access launch biometric IDs for vaccination verification:

Itoco Inc., a Bio Tech development, production and distribution company, announced that its patent-pending, biometric Immutable Virus Test Result Verification System is now available as an open-source repository on GitHub. The system was recently deployed via blockchain Smart Contract.

Through the open-source code, users can view the code and verify the exact functionality. Potential partners or customers will have transparency of how their data is being used, for example, what is being retrieved from the blockchain. Mobile application users can verify that the only data used is a hashed public key combination of the blockchain wallet and user biometric, but not the patients’ raw biometric or any personally identifiable information (PII).

The publicly available Smart Contract, written in the Solidity smart contract programming language for Ethereum, allows Itoco’s partners and customers to verify exactly how the blockchain is being leveraged as part of the overall system.

The release of a patented technology service to provide COVID-19 vaccine verification booklets and cards was announced last week by World Health Access, a subsidiary of International Health and Wellness LLC.

EU agency pushing for research into how AI will affect fundamental human rights:

In a world awash in new authoritarian sentiment, it is notable that the European Union has published a lengthy report saying it is key for nascent AI regulation to adequately protect fundamental human rights in the region.

Few nations buying, developing or deploying biometrics and other AI systems are offering more than lip service to the need to shelter people and their personal data from rapaciousness by the private sector or high handedness by government. The new report, by the EU’s Agency for Fundamental Rights, carries 91 interviews with business executives and government officials from several EU member states.

Free mask compliance app powered by biometrics from SAFR by RealNetworks launched:

A partnership has been announced between RealNetworks and the COVID-19 International Research Team to bring the latter’s software product MaskCheck, which is powered by SAFR from RealNetworks, to smartphone users around the world.

MaskCheck runs on Android and iOS devices, and provides a free mask compliance app, along with a service and data platform to help communities and businesses operate more safely. It delivers accurate compliance data in real time, according to the announcement.

Mask detection takes less than one second, according to a video accompanying the news.

Running on a tablet, MaskCheck detects faces as people walk past and determines if they are wearing a mask, and doing so properly. The software provides visual feedback to encourage proper precautions, and aggregates non-personally identifiable information (non-PII) to submit the data to a public health data repository. Data collected includes location and venue type, and non-PII demographic details.

Top five FIDO Korea hackathon projects unveiled as Microsoft reviews strides in passwordless digital ID:

The top five winners of the 2020 FIDO Hackathon — Goodbye Password Challenge have been announced, and organizers have published videos detailing the journeys of the participants throughout the process as well as the winning digital ID projects and how they highlight possibilities for future FIDO authentication.

According to a FIDO blog post, Moses’ Miracle, a team of three university students majoring in computer engineering won the gold award. Their project is a gate access control system based on FIDO authentication, and is a smartphone-based digital identity system that helps people access different security areas much faster and more conveniently without remembering passwords, physical keys, or smart cards, the blog post states. The solution is also said to be less time-consuming and less costly.

Shenzhen BCTC branch laboratory’s biometrics testing gets FIDO certification:

It is now possible for the Shenzhen branch of the Bank Card Test Center (BCTC) to carry out testing services for a number of biometrics products in line with the FIDO Alliance certification program.

This is according to information posted on the LinkedIn account of the financial technology and security services company.

With the certification, the Shenzhen BCTC office can conduct tests on aspects including registration, verification accuracy (FRR/FAR), presentation attack detection, log preservation, security information and storage to be able to figure out whether the biological component product meets the FIDO Biometrics Requirements V1.1 or V2.0, the post notes. The lab can also issue test reports with the approval of FIDO, it adds.

Automotive biometrics solutions with Cipia, ams and Cerence technology launched:

Three new solutions are joining the growing automotive biometrics market, as Cipia and ams AG have announced driver monitoring capabilities powered by biometrics, while Cerence’ voice biometric authentication has been integrated by a major auto manufacturer.

Eyesight Technologies has been renamed Cipia and announced that its Fleet Sense biometrics-based driver monitoring service has reached general availability.

Fleet Sense is meant to help fleets and telematics service providers (TSPs) reduce the accidents and related costs associated with fatigue and distraction, according to the announcement.

Remark Holdings launches edge face biometrics with Nvidia module integration:

Remark Holdings, Inc. has announced that its KanKan AI subsidiary has released two new edge computing systems to perform facial recognition and other applications locally. The S and T Series Smart Boxes are AI-powered edge computers equipped with NVIDIA‘s newest group of high-performance edge computing modules.

KanKan AI’s software will integrate with NVIDIA’s CUDA architecture to power edge computing. The NVIDIA modules promise improved size and energy efficiency.

Ariadnext adds video ID document capture to digital identity solution:

Digital identity solutions firm Ariadnext has announced dynamic document capturing among upgrades to its remote ID verification solution. The new functionalities are compliant with the new ANSSI criteria published in the December 1st reference framework draft that is projected to be enacted in March 2021.

Ariadnext’s IDCheck.io and MobileID.io offer an array of biometric and ID document checks via automated services. The company further promises fraud detection and streamlining for client onboarding services to be fully KYC and AML compliant.

The newly added features include dynamic document capture with Officially Valid Document (OVD) verification and a hybrid verification using a specially trained fraud-detection operator. Another feature, the dynamic capture capability for ID documents augments fraud detection capabilities by making use of security features previously undetectable in static mode.

TypingDNA announces progress on typing biometrics-based app to analyze productivity:

TypingDNA says it is in the research phase of developing an application to analyze people’s behavioral patterns and attitudes while they are typing to help them improve focus and productivity.

The company said in a blog post that the application, TypingDNA Focus, is unique as it is designed to employ a new biometrics technology to enable users have a better understanding of their state of mind to make use of their typing data.

TypingDNA Focus is a continuation of a project that was started in 2016 when the company sought to develop its core proprietary artificial intelligence algorithms which are trained to analyze human typing behavior with the view to better understanding users’ device interaction and transforming typing patterns into compelling data.

Infinity Optics to use Rank One facial recognition algorithms in True Biometric Hash technology:

Rank One Computing and Infinity Optics have formed a partnership under which Rank One’s facial recognition algorithms will be used in the development of Infinity’s QuantumCrypt biometric face verification solution.

The companies note the increase in enterprise phishing attacks during the pandemic in the announcement.

The company’s biometric cryptographic technology gathers many stable and repeatable feature vectors directly from the biometric device to generate a True Biometric Hash code, which is used as input for cryptographic authentication keys that can be used for digital signatures or secure access to private data. Verification with the QuantumCrypt platform is not based on stored images or encrypted templates, and feature extraction from the biometric data does not involve distance calculation.

QuantumCrypt produces a stable binary code, registration is revocable, and the platform is device-agnostic.

CyberLink’s FaceMe 1:N biometric accuracy among top-ranked by NIST:

CyberLink Corp. announced its ranking as a global top ten vendor after its FaceMe AI facial recognition engine earned high marks during the latest Facial Recognition Vendor Test (FRVT). The National Institute of Standards and Technology (NIST)-administered FRVT is the most rigorous assessment for facial recognition precision within the industry.

CyberLink’s automated face recognition algorithm FaceMe promises flexible edge computing architecture optimized across devices and platforms, enabling a variety of use cases for smart banking, security, retail, and smart cities, the company says. Driven by deep learning algorithms, FaceMe brings reliable, high-precision, and real-time facial recognition crucial to AIoT applications.

The newest FRVT 1:N Identification report ranked the FaceMe algorithm behind those of nine other global vendors overall, after it scored a true acceptance rate (TAR) of 98.11 percent on its identity recognition against a collection of 1.6 million webcam and Visa photos. The company also notes that only three vendors from outside of China and Russia are ranked higher on the overall leaderboard.

CyberLink states FaceMe delivers efficient processing and higher levels of security than cloud-based solutions, and can run on low-powered CPUs for cost-efficient facial recognition on IoT and AIoT devices. The solution can also be integrated with high-end servers, workstations, and GPU-equipped personal computers.

Biometric shoe insole provides new way to measure health-insights:

A new biometric insole will make monitoring and tracking health insights simple and efficient. Analytic consulting service Pinnacle Solutions has announced its partnership with California-based wearable biometric technology company Autonomous_ID to create the BIO_SOLE.

BIO_SOLE is a shoe insert, designed to record an individual’s unique way of standing, walking, running, or gait; which creates a unique Pedo Biometric signature linked to their identity — much like a fingerprint. The insole uses pressure sensors, an accelerometer, a gyroscope and circuitry to track every part of the movement, which can be influenced by body position, weight, injury, or disease.

The insole technology also promises data security and privacy, and eliminates the security risks associated with lost swipe cards, according to Autonomous_ID’s website. Pinnacle’s analytic platform then gains insights from the insole, with the application of machine learning and artificial intelligence tools to rapidly build thousands of advanced data models for multiple use cases.

Aculab broadens implementation options for voice biometrics compliance:

Aculab has announced a new update to its Communications Platform as a Service (CPaaS) solution Aculab Cloud.

Aculab Cloud will now feature a range of flexible implementation services, including technical review of application code, writing and deploying voice capabilities, and SMS or fax applications as a fully managed service.

According to the company, the new update is designed to help companies build up their offering more efficiently while also reducing time-to-market and development costs. To support the lowest possible call latencies, Aculab Cloud has also been designed with regional, local cloud architecture.

By operating in a fully-regionalized way, Aculab Cloud can keep users’ voice recognition and other data within the region of their choice to meet data sovereignty and regulatory compliance requirements. The company runs multiple clouds to provide companies with a cloud service based in their country of operation.

Report: Huawei and Megvii Designed Facial Recognition System to Target Uighur Minorities:

The Washington Post has obtained a report that indicates that Huawei and Megvii worked together to build a surveillance system that would identify members of China’s Uighur Muslim minority group.

These Weeks’ News by Categories

Access Control:

Consumer Electronics:

Financial Services:

Civil / National ID:

Government Services & Elections:

Facial Recognition:

Fingerprint Recognition:

Iris / Eye Recognition

Voice Biometrics

Behavioral Biometrics

Wearables

Liveness Detection

Mobile Biometrics

Biometrics Industry Events

Deep Learning 2.0 Virtual Summit: Jan 28, 2021 — Jan 29, 2021

Border Management & Technologies Summit Asia: Feb 23, 2021 — Feb 25, 2021

IFINTEC Finance Technologies Conference and Exhibition: Mar 9, 2021 — Mar 10, 2021

SECON 2021: Mar 10, 2021 — Mar 12, 2021

2nd Annual Facial Recognition Summit: Apr 7, 2021 — Apr 8, 2021

Secure Identification 2021: Apr 14, 2021 — Apr 16, 2021

Identity Management Symposium: Apr 21, 2021 — Apr 22, 2021

Critical Infrastructure Protection & Resilience Europe: May 11, 2021 — May 13, 2021

5th India Homeland Security: May 13, 2021 — May 14, 2021

MISC

Are You Comfortable with Your Government’s Use of Biometrics? The annual FindBiometrics Year in Review survey is now open, offering industry insiders the opportunity to share their views on some of the most important issues facing the industry today. And among the most fraught of those topics is probably the question of whether industry professionals agree with the statement, “I am comfortable with the government’s use of biometrics in the country where I live.”
ID Talk Podcast: FaceTec CEO Kevin Alan Tussy on the State of Face Authentication and the $100K Spoof Bounty

Subscribe to Paradigm!

Medium. Twitter. Telegram. Telegram Chat. Reddit. LinkedIn.

Main sources

Research articles