Paravision’s Post

Wow! It turns out that most facial recognition actually is biassed. 𝗘𝘅𝗰𝗲𝗽𝘁 𝗳𝗼𝗿 𝗣𝗮𝗿𝗮𝘃𝗶𝘀𝗶𝗼𝗻. It’s often said that, when it comes to #facialrecognition matching, as long as you're using a top 20-ranked NIST-tested vendor, it doesn’t really matter which: they’re all so close. … and at first glance that may appear to be true. But things often are not as they appear at first glance. The National Institute of Standards and Technology (NIST) rankings are based on False Non-Match Rates (FNMR: the error rate at which two photos of the same person do not match) at a pegged False Match Rate (FMR: the error rate at which two photos of two different people do match). (Think of somebody trying to get through an eGate at an airport using their own passport, and the photo doesn’t match and they get referred to secondary.) … and indeed the top 20 vendors do all have close FNMRs. 𝗘𝘅𝗰𝗲𝗽𝘁 𝗵𝗲𝗿𝗲’𝘀 𝘁𝗵𝗲 𝘁𝗵𝗶𝗻𝗴: In order to assess bias, #NIST compares vendors’ FMRs (not FNMRs) across different demographic groups, or the rate at which a photo of somebody within a demographic group falsely matches with a photo of a different person within that same group. (Think of somebody successfully getting through an eGate at an airport using somebody else’s passport, or falsely being identified using facial recognition on a surveillance camera.) NIST doesn't present a topline FMR ranking, and this level of detail is just not apparent when looking at the FNMR leaderboard. When analysing FMR across biographic groups, the differences are startling: 🚀Almost every other vendor except Paravision struggles with all areas of origin outside of Eastern Europe. 🚀Paravision shows exceptional and consistent accuracy across all groups. 🚀Paravision performs better than every other vendor for people of Central American, Western African, Caribbean, Eastern African, South Asian, and East Asian origin, and on par with leading vendors for people of Eastern European origin. NIST only undertakes demographic analysis on their 1:1 tests, but these issues are only compounded on 1:N uses of facial recognition. And if a vendor does not submit to NIST’s 1:1 testing, there is no independent benchmark analysis of #bias and demographic inclusion for that vendor. So, you may think that facial recognition matching is commoditised and there is no real difference between the top vendors. That may be true, unless of course, you are concerned with: ✅Human and Civil rights. ✅Maximising demographic inclusion. ✅Minimising legal and reputational risk. ✅Remaining compliant with emerging regulations. ✅Increasing operational efficiency across all demographic groups. ✅Minimising security risks. In which case, the choice is clear. 𝗗𝗼𝘄𝗻𝗹𝗼𝗮𝗱 𝘁𝗵𝗲 𝘄𝗵𝗶𝘁𝗲𝗽𝗮𝗽𝗲𝗿 𝗶𝗻 𝘁𝗵𝗲 𝗹𝗶𝗻𝗸 𝗯𝗲𝗹𝗼𝘄 𝗳𝗼𝗿 𝗳𝘂𝗿𝘁𝗵𝗲𝗿 𝗱𝗲𝘁𝗮𝗶𝗹. Get in touch if you’d like to discuss running Paravision in Challenger Mode with your existing technology to compare the results.

This chart speaks volumes about our team's commitment to inclusion in face recognition. If it doesn't work for everyone, it doesn't work for anyone. This work is never done, and we take the same approach with our Deepfake Detection, Liveness, and Age Estimation products (the latter two benchmarking this fall (unfortunately no objective public benchmark for DF yet - we'll be there when there is)). If you are undertaking a project in face recognition today, you need to be thinking about excellent performance across demographics. It's not only the right thing to do from an ethical perspective, it's the right thing to do from a business and legal perspective. If you notice that there are some familiar names missing on this list, it's either because they weren't in the top 25, or they don't subject themselves to this kind of public benchmarking (and you should ask yourself, "why not?" - we've got some thoughts). #NIST #facerecognition #facialrecognition #bias #IDV #fintech #banking #payments #SSO #cybersecurity #healthID #smartborders #fanexperience #automotive #liveness #antispoofing #deepfake #ageestimation

If you care about #demographic performance (a.k.a. #bias) in #ai and face recognition, please consider reading this report. The analysis is a breakthrough in understanding demographic performance across face recognition false match rates (i.e. the likelihood of a false positive, which has security, customer experience, civil rights, and human rights implications). National Institute of Standards and Technology (NIST) FRTE has created a treasure trove of data in the form of vendor "Report cards" that we were able to extract and analyze. Bottom line: Paravision's false match rate performance across demographics (same age, gender, and region of origin) is singular, and it's not really even close. If you're interested, we'd be happy to walk through this data, how it was extracted from NIST, and / or why it matters. If you don't see a certain vendor in the graph below, they are not top-25 on NIST FRTE 1:1 or they didn't submit to NIST FRTE 1:1, which is where the demographic analysis is performed. Feel free to DM me here or email me at my first name @paravision.ai to discuss more!

Time is running short. Therefore, financial entities and their ICT partners subject to #DORA requirements must make up their minds and define priorities. The first step is to have a clear picture of the gaps that need to be filled to achieve #compliance by the deadline. In parallel, it's time to identify and plan for improvements to increase efficiency in mitigating systemic risks and improve third-party risk management. Several solutions can ease Digital Operation Resilience, so having a plan is of paramount importance to leverage DORA as an inspiration to future-proof #governance processes #thirdpartyriskmanagement #digitaloperationalresilience

🚨 The final countdown to DORA is on! 🚨 With the EU’s Digital Operational Resilience Act (DORA) going live in January 2025, financial institutions must act now. Ensure your organization is prepared to comply by prioritizing gaps, planning tools and processes, and strengthening digital resilience. Don't let time run out—discover the steps needed to meet these new regulatory requirements. Learn more here: https://lnkd.in/dF_Brhgd #DORA #DigitalResilience #FinancialRegulation #Compliance Maria Adele Di Comite

Meet Tom-Martijn Roelofs. He’s ING’s global head of Security Strategy and Data and last week, he was a speaker at the Gartner conference in London, covering DORA! Missed it? Here’s a snapshot to get you up to speed: ❓       What exactly is #DORA? Digital Operational Resilience Act (DORA), is European Union’s new regulation that comes into force in January 2025. 💡      What problem is it trying to solve? With ever-increasing digitalisation in the financial services industry, it becomes apparent to design and operate our services so that we are digitally resilient to the various root causes of outages, including cyber-attacks. 🚀     What has #ING been doing to prepare for DORA? ING is implementing the requirements from DORA via a COO-lead program. One of the first steps was to assess the impact of DORA and distinguish between new requirements and those that we were already undertaking. Then, we took the quite complex step to define our critical business services and register the relevant applications and infrastructure connected to these. The last part will be to ensure compliance to the regulatory technical standards of all critical business services, and this is still ongoing.   😎      Will DORA have a direct impact on customers and/or clients? The Act will not directly impact customers and clients. It is all about financial institutions making sure their critical business services are sufficiently resilient. 📅      When will DORA be introduced? The law passed the European Parliament in November 2022, and becomes effective on 17 January 2025. 🌎      What impact will there be outside of the EU? As an EU law, DORA will have a big impact across the EU, the effects of which will be noticed by non-EU based providers. EU-based financial institutions will implement requirements uniformly, e.g. via global services, policies, and internal standards, which will have a big impact on service providers, , such as telecommunications companies, and this will be noticed outside of the EU. For the full picture, head here: https://lnkd.in/eAWuxenC

𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱𝗶𝘇𝗮𝘁𝗶𝗼𝗻 – 𝘁𝗵𝗲 𝗸𝗲𝘆 𝘁𝗼 𝗶𝗻𝘁𝗲𝗿𝗼𝗽𝗲𝗿𝗮𝗯𝗶𝗹𝗶𝘁𝘆 ➡️ https://lnkd.in/e2T5FFhm Our vision for shaping the future data economy requires international standards for #DataSpaces. Such standards are crucial for enabling sovereign and trustworthy data sharing, as demanded by European stakeholders, and are essential for ensuring interoperability across the global landscape. Swipe and figure out why IDS(A) plays a pivotal role in the standardization process, enabling organizations to unlock the full potential of data spaces. Silvia Castellvi

ICYMI... our panel on Digital Public Infrastructure yesterday generated some new ideas to ensure customer-centricity. A few questions from my own notebook that I'm keeping in my mind as I move forward with conversations: 1. Who does DPI leave out, and how might we think about engaging them? 2. What are the consumer research budgets for ongoing DPI design, to avoid unintended consequences and surface risks? 3. Is DPI designed with women's digital financial capability in mind? Huge thanks to Jayshree Venkatesan, David Porteous, Deepti George, and Nanjira Sambuli for sharing your big questions and early thoughts on this topic as we move forward with more inclusive infrastructure for financial services. And if you want to hear more about Women's World Banking's perspective on this topic, see a great resource in the comments aimed at policymakers.

Consolidated tape regulations in Europe and the U.S. are on parallel tracks with similar objectives. But there is no clear path to consistency and convergence. https://lnkd.in/epXnEU52 #riskmanagement #financialrisk #financialmarkets #data

Responsible data use is a business imperative, but do you really know what it means? Data&Storage Asia consulted the experts, and we not only got some great insights to define this seemingly abstract concept, but also some practical recommendations on how to actually operationalise responsible data use. And in an era when data is said to be "the new oil," we believe organisations need to be guided on how to use this most valuable asset. PS. A big thank you to the experts at Sophos, Wasabi Technologies, Hewlett Packard Enterprise, Affinidi, GreaterHeat, and Ping Identity for the wonderful insights! Click the link below to read more about responsible data use: https://lnkd.in/gUf7qBax.