Search Results

< back AI Case Study Researchers from Warsaw University of Technology identify irises as live or dead with a mean accuracy of 98.94% using convolutional neural networks Researchers from Warsaw University of Technology and other institutions assessed the ability of a deep convolutional neural network to determine whether images of irises belonged to live or deceased subjects. While they achieved a mean classification accuracy of 98.94%, they found that post-mortem irises could be used for a period after death to deceive biometric systems. Industry Public And Social Sector Education And Academia Project Overview Researchers from Warsaw University of Technology, Research and Academic Computer Network, and the Medical University of Warsaw conduct the first research on a "method for iris liveness detection in respect to the post-mortem setting, based on a deep convolutional neural network VGG-16, adapted and fine-tuned to the task of discerning live and dead irises" according to their paper. This involved cropping the iris photos of both live and deceased subjects to avoid bias in the way eyelids are often held open on deceased subjects. Images of deceased eyeballs at various port-mortem timeframes were taken from a publicly available database, and live subjects were photographed using the same equipment. Reported Results Mean accuracy was 98.94%. From the arXiv paper: "We can expect a few false matches (post-mortem samples being classified as live iris samples) with images obtained 5 hours after death, regardless of the chosen threshold. This can be attributed to the fact that these images are very similar to those obtained from live individuals, as post-mortem changes to the eye are still not pronounced enough to allow for a perfect classification accuracy." However, when these recently deceased subjects were excluded, misclassification of post-mortem irises as live dropped to 0, and misclassification of live irises as post-mortem to about 1%. Technol ogy From the arXiv paper: "For our solution, we employed the [deep convolutional neural network] VGG-16 model pretrained on natural images from the ImageNet database, which has been shown to repeatedly achieve excellent results in various classification tasks after minor adaptation and re-training. We thus performed a simple modification to the last three layers of the original graph to reflect the nature of our binary classification into live and post-mortem types of images, and performed transfer learning by fine-tuning the network weights to our dataset of iris images representing both classes. For the network training and testing procedure, 20 subject-disjoint train/test data splits were created by randomly assigning the data from 3 subjects to the test subset, and the data from the remaining subjects to the train subset, both for the live and post-mortem parts of the database. These twenty splits were made with replacement, making them statistically independent. The network was then trained with each train subset independently for each split, and evaluated on the corresponding test subset. This procedure gives 20 statistically independent evaluations and allows to assess the variance of the estimated error rates. The training, encompassing 10 epochs in each of the train/test split run, was performed with stochastic gradient descent as the minimization method with momentum m = 0.9 and learning rate of 0.0001, with the data being passed through the network in mini batches of 16 images. During testing, a prediction of the live or post-mortem class-wise probability was obtained from the Softmax layer, together with a corresponding predicted categorical label." Function Information Technology Security Background From the arXiv research paper: "Law enforcement officers in the U.S. are reportedly already using the fingerprints of the deceased to unlock the suspects’ iPhones, which immediately brings up the topic of whether liveness detection should be one of the components of Presentation Attack Detection implemented in such devices. With a constantly growing market share of iris recognition, and recent research proving that iris biometrics in a post-mortem scenario can be viable, these concerns are also becoming true for iris." IEEE Spectrum reports that "The iris, the colored ring of muscle that controls the contraction and dilation of the pupil, is composed of tiny fibers that form an intricate and unique pattern in each individual’s eye. Iris scanners use both visible and near-infrared light to look at hundreds of points within these patterns, then try to match them with a registered profile." Benefits ​ Data 574 near-infrared iris images from the public Warsaw BioBase PostMortem Iris dataset, taken from 17 subjects at various time points from 5 hours after death till 34 days, as well as 256 iris photos from live subjects, taken using the same iris camera.

< back AI Case Study Obvious' artists create the first GAN-generated artwork to be auctioned and sold for $432,500 An AI artwork sold for $432,500, despite its estimated selling price of $7,000 to $10,000 at an auction at Christie's New York. The piece, titled Portrait of Edmond de Belamy, was created by Paris-based art collective Obvious with the use of Generative Adversarial Network algorithm (GAN). The system was fed information from 15,000 portraits painted between the 14th and 20th centuries. Industry Consumer Goods And Services Entertainment And Sports Project Overview "AI artwork sells for $432,500 — nearly 45 times its high estimate — as Christie’s becomes the first auction house to offer a work of art created by an algorithm This portrait, however, is not the product of a human mind. It was created by an artificial intelligence, an algorithm defined by that algebraic formula with its many parentheses. And when it went under the hammer in the Prints & Multiples sale at Christie’s on 23-25 October, Portrait of Edmond Belamy sold for an incredible $432,500, signalling the arrival of AI art on the world auction stage. The painting, if that is the right term, is one of a group of portraits of the fictional Belamy family created by Obvious, a Paris-based collective consisting of Hugo Caselles-Dupré, Pierre Fautrel and Gauthier Vernier. They are engaged in exploring the interface between art and artificial intelligence, and their method goes by the acronym GAN, which stands for ‘generative adversarial network’. But one of the beguiling things about the depiction of Edmond Belamy is that it departs from a human idea of an 18th-century portrait. There is something weirdly contemporary about him: he looks unnervingly like one of Glenn Brown’s art-historical appropriations. Why might that be? ‘It is an attribute of the model that there is distortion,’ says Caselles-Dupré. ‘The Discriminator is looking for the features of the image — a face, shoulders — and for now it is more easily fooled than a human eye.’ It must also surely be the case that portraiture is an extremely tough genre for AI to take on, since humans are highly attuned to the curves and complexities of a face in a way that a machine cannot be. It turns out that the difficulty was part of the collective’s thinking. ‘We did some work with nudes and landscapes, and we also tried feeding the algorithm sets of works by famous painters. But we found that portraits provided the best way to illustrate our point, which is that algorithms are able to emulate creativity.’" Reported Results The Portrait of Edmond Belamy sold for $432,500. Technol ogy "‘The algorithm is composed of two parts,’ says Caselles-Dupré. ‘On one side is the Generator, on the other the Discriminator. We fed the system with a data set of 15,000 portraits painted between the 14th century to the 20th. The Generator makes a new image based on the set, then the Discriminator tries to spot the difference between a human-made image and one created by the Generator. The aim is to fool the Discriminator into thinking that the new images are real-life portraits. Then we have a result.’" Function R And D Product Development Background "Auctioning the work seemed to be a test by Christie’s of the traditional art market’s interest in AI art." (NYTimes) Benefits ​ Data " the system was fed a data set of portraits painted between the 14th and 20th centuries." (dezeen.com)

< back AI Use Case Compose and conduct market research Automated generation of market research questions and deployment across digital channels, Capture and visualisation of responses. Function Marketing Marketing Research Planning Benefits Data - Data enhancement,Cost - Staff efficiency Case Studies University of Melbourne~University of Melbourne researchers predict consumer beer preference with 82% accuracy using biometric measures ,MARS Marketing~MARS Marketing improved accuracy of purchase intent in marketing research by 8% using facial analysis during ad viewing,Bank of America~Bank of America enhances its published currency research with machine learning ,Liquidnet~Liquidnet assists fund managers in decision making by analysing market and past trading data with machine learning Potential Vendors ​ Industry ​ ​ Data Sets Text,Structured / Semi-structured,Audio,Video AI Technologies Product Type - Vision ,Product - Data Capture - Bio Sensor,Machine Learning (ML),Product - Eye Tracking

< back AI Use Case Identify negative responses to drug trials by for example monitoring social networks for early problem indicators Identifying negative responses (monitor social networks for early problems with drugs) Function R And D Core Research And Development Benefits Operational Support - Situational awareness,Risk reduction - Patient outcomes Case Studies ​ Potential Vendors ​ Industry ​ ​ Data Sets Structured / Semi-structured,Text,Images AI Technologies Product Type - NLP - Text Sentiment Analysis,Product Type - Natural Language Processing (NLP),ML Task - Prediction - Data Translation/Transformation,ML Task - Prediction - Annotation,Machine Learning (ML),ML Task - Prediction - Multi-class Classification,Traditional AI

< back AI Case Study Jerrit Canyon Gold identifies high prospectability areas using Goldspot Discoveries' machine learning geological model Jerritt Canyon enlisted Goldspot Discoveries for a proof of concept study to determine which areas had the highest promise for prospecting at its site, which it did by creating a geological model based on historic and mining data. Industry Basic Materials Mining And Metals Project Overview From GlobeNewsWire: "The Jerritt Canyon project, majority owned by Sprott Mining Inc., through Jerritt Canyon Gold LLC, a private mid-tier gold producer in Northern Nevada has asked Goldspot to assess a significant amount of data in order to assist with continued exploration. Goldspot consolidated over 30 years of historical remote sensing, mining, and exploration data into one comprehensive and functional geological model. Goldspot Artificial Intelligence was then able to use this geological model to identify correlations in the data layers of existing and historically mined deposits." Reported Results The POC is moving onto its final phase where, according to GlobeNewsWire, "Jerritt has agreed to commence the first 1,000-meters of a 5,000-meter drill program as soon as logistically possible", indicating success thus far. Technol ogy "Goldspot has developed a machine‐learning algorithm capable of significantly improving mineral exploration targeting. The Goldspot Algorithm is proven to mitigate investment risk and increase the efficiency and success rate of exploration in data‐rich environments." (GlobeNewsWire) Function Strategy Planning Background According to Goldspot's website, "Goldspot Discoveries is revolutionizing the mineral exploration business by utilizing machine learning to target on a regional and localized scale: Mineral deposits form for a reason. Machine learning links this "reason" to available geoscience data to determine the relationship. With that "relationship" we can predict likelihood of mineralization in new exploration regions". Benefits ​ Data "[H]istorical remote sensing, mining, and exploration data" (GlobeNewsWire).

< back AI Use Case Model and predict customer lifetime value Determine the likely lifetime value of customers to allow better customer management and engagement to increase revenues. It will allow a firm to better target its marketing resources to ensure that customer acquisition costs are managed appropriately by relevant segment. Function Marketing Customer Management Benefits Operational Support - Sales forecasting,Operational Support - Demand forecasting,Operational Support - Production forecasting Case Studies ASOS.com~ASOS.com researchers demonstrate improved customer lifetime value predictions using neural networks and automatic feature selection but do not advise implementation due to increased cost Potential Vendors ​ Industry ​ ​ Data Sets Structured / Semi-structured AI Technologies Machine Learning (ML),Model Architecture - Deep Neural Networks,Algorithm - Embedding,Model Architecture - Recurrent Neural Network (RNN),Model - Unsupervised Learning,Algorithm - Random Forest

< back AI Use Case Generate cloned voices Generate cloned voice - at this stage largely for artistic, trouble-making and media purposes. This has potentially troubling implications for so-called "fake news" applications amongst many potential uses. Function Strategy Data Science Benefits Revenue - New service,Revenue - New product,Cost - Job automation,Data - Data enhancement Case Studies Baidu~Baidu researchers synthesise speech through neural voice cloning with limited data samples,Cereproc~Cereproc generates synthetic speech from text using natural language processing and machine learning Potential Vendors ​ Industry ​ ​ Data Sets Audio AI Technologies Algorithm - Generative Adversarial Network (GAN),Machine Learning (ML),Model Architecture - Artificial Neural Network (ANN),Model Architecture - Deep Neural Networks,ML Task - Prediction - Generation

< back AI Case Study Kik, a popular messaging platform, uses machine learning to integrate and scale different systems Kik, a popular messaging platform, uses Databricks' AI platform to integrate and scale different systems as it was having difficulty keeping pace with the exponential growth in user base and contents. Industry Technology Internet Services Consumer Project Overview "Databricks has allowed Kik to improve team productivity while significantly reducing data engineering overhead at scale: Support for Multiple Languages: Familiarity with SQL – finding expertise in SQL is much more cost effective than other languages Performant and Reliable Data Pipelines: Ability to handle 5TB new data per day streaming into the data lake. Fully Managed Platform: Managed Apache Spark™ takes the burden off of needing Spark expertise and having to understand the internals of Spark. Collaborative Workspace: Streamlined data analysis and fostered collaboration through support for multiple languages, easy search, and real-time commenting." Reported Results According to Databricks: * Data engineering efforts reduced by 70% * Big jobs run 2x faster * Improved Collaboration - Sharing data is instant via links to notebooks * Ability to combine and ensure clean data to trust for analysis Technol ogy ​ Function Digital Data Digital Data Management Background "Kik is a messaging platform, already at 300 million users and is incredibly popular in the youth market, where it is growing fastest among teenagers. But it has to compete with giants like Google and Facebook with small team and developing fast came at the expense of scalable engineering resulting in disjointed systems and tools for data analysis and lack of standard. Amazon Redshift was starting to break at the seams with 300 data pipelines with 5TB of new data per day." Benefits ​ Data ​

< back AI Case Study International Centre for Missing and Exploited Children The International Centre for Missing and Exploited Children has launched a pilot project using Amazon's facial recognition system to identify missing children from images posted online. Industry Public And Social Sector Ngo Project Overview "Called the GMCNgine, the system leverages machine learning and artificial intelligence technology to automatically scan the web for images of missing children... The system also features a dynamic advertizing capability that uses geo-targeting to deliver child endangerment alerts to specific communities as necessary. A key component of the GMCNgine is Rekognition, Amazon’s controversial computer vision platform. Amazon has faced intense criticism this year from privacy and civil rights groups, shareholders, and even Amazon employees over its sale of Rekognition to law enforcement agencies seeking to use it in public surveillance; the company has defended itself in part by arguing that Rekognition offers a valuable tool for organizations seeking to track the victims of human trafficking and sexual exploitation. The GMCNgine was launched this week in Cordoba, Spain, at the 10th annual Global Missing Children’s Network conference". Reported Results Project just launched at time of reporting Technol ogy Amazon's Rekognition facial recognition machine learning platform. Function Operations General Operations Background "Facial recognition technology’s growing role in the fight against child exploitation could see a substantial expansion after the International Centre for Missing and Exploited Children’s launch of a new AI-driven system this week." Benefits ​ Data The data comes from "the 33 members of the ICMEC’s Global Missing Children’s Network, which includes law enforcement agencies and NGOs spanning 29 countries."

Risk AI Case Studies RyanAir Audit Transportation Ryanair monitors the performance of its online service in real time and identifies root cause of issues in minutes compared to hours previously using machine learning Read More Facebook Audit Technology Facebook and CrowdAI researchers develop a machine learning method to identify geographic areas heavily affected by natural disasters Read More eBay Audit Technology eBay research identifies 40% of credit card fraud with high precision automatically using machine learning Read More The Zoological Society of London Audit Public And Social Sector The Zoological Society of London is preventing poaching with machine vision and machine learning Read More Choozle Audit Professional Services Choozle announces partnership with Oracle's Grapeshot to determine optimal online ad placement and ensure brand safety using AI Read More Infinity Property & Casualty Audit Professional Services Infinity Property & Casualty processes claims 35% faster by developing a solution to flag potentially fraudulent claims using machine learning Read More Influential Audit Professional Services Influential assesses suitability of social media influencers for use in marketing campaigns by analysing social media content using natural language processing Read More Man Group Plc Audit Financial Services Man Group extends AI models to four large funds for predicting financial market pricing and risk Read More Moody's Analytics Audit Financial Services Moody's Analytics find loan risk prediction is improved using AI models over its proprietary RiskCalc model Read More Nasdaq Audit Financial Services Nasdaq purchases AI software company Sybenetix to monitor and flag unusual trader activity using AI Read More Insurance Company Anonymous Audit Financial Services An Anonymous UK Insurance firm demonstrates that it can better assess the risk of insuring companies by automatically determining relevant UK court cases with deep neural networks Read More Direct Line Insurance Audit Financial Services Direct Line Insurance uses machine learning to automatically flag fraudulent claims in automotive section with 75% hit rate Read More MS & AD Insurance Group Audit Financial Services MS & AD Insurance Group detects 2.5 times more fraudulent claims than industry average using machine learning Read More Thrivent Financial Audit Financial Services Thrivent Financial attracts 250% more visitors to their website and acquires 22% more customers using Hearsay Read More American Express Audit Financial Services American Express identifies $2 billion in potential annual incremental fraud incidents with machine learning Read More Chime Audit Financial Services Chime decreases basis point loss by 40% using a machine learning fraud detection platform Read More Danske Bank Audit Financial Services Danish Danske Bank increases payment fraud detection by 60% and reduces false positives by 50% with machine learning Read More Danske Bank Audit Financial Services Dankse Bank identifies fraudulent online banking customers with 99.7% accuracy in a pilot with the BehavioSec behavioural biometrics system Read More Earthport Audit Financial Services Earthport Payment Network reduces false positives of automated suspicious transaction detection using AML risk data in real-time Read More Holvi Payment Services Audit Financial Services Holvi reduces time spent investigating false positives for customer risk using AI platform ComplyAdvantage Read More Monzo Audit Financial Services Monzo decreased pre-paid card fraud to 0.1% and false positive rate to 25% using machine learning Read More NatWest Audit Financial Services NatWest Bank prevents over £7m worth of corporate fraud by using machine learning to detect suspicious invoice payment activity Read More OCBC Bank Audit Financial Services OCBC bank reduces number of false positive financial transaction alerts by 35% with machine learning Read More Royal Bank of Scotland Audit Financial Services RBS identifies invoice fraud saving £7 million of losses to customers with artificial intelligence Read More SMFG Audit Financial Services SMFG, the Japanese financial services company, uses deep neural network to analyse credit card transactions and predict fraud with an 80-90% accuracy Read More Predictim Audit Consumer Goods And Services Predictim offers trustworthy background checks for babysitters using natural language processing and computer vision Read More New York Times Audit Consumer Goods And Services New York Times increases article commenting capacity by 3x using machine learning to automatically flag inappropriate content for moderator review Read More Tesla Inc Audit Consumer Goods And Services Drivers demonstrate Tesla's Autopilot limitations in recreation of fatal Model X crash at exact same location Read More Uber Audit Consumer Goods And Services Uber's self driving car failed to visualise and recognise a pedestrian and the potential risk in real-time resulting in a fatal accident Read More Lyft Security Transportation Lyft delivered a 40% increase in potentially fraudulent users detected without increasing false positives by using neural networks Read More 1 2 3 1 ... 1 2 3 ... 3

< back AI Use Case Leverage a deep learning library-SDK-API to support developers Leverage deep learning library-SDK-APIs to quickly and cost-effectively build capacity and support new system deployments. Function Information Technology Development Benefits Operational - Faster system build / test / deploy Case Studies Lobe~Lobe is a visual tool that can be used to build custom deep learning models without writing any code Potential Vendors Lobe Industry Technology Software And It Services Data Sets Structured / Semi-structured AI Technologies Machine Learning (ML)

< back AI Case Study The LawGeex AI algorithm outperforms human lawyers by 9% in identifying issues with historic non-disclosure agreements The LawGeex AI algorithm was tested against a group of human lawyers in finding issues commonly identified in non-disclosure agreements for a set of corporate NDAs in a given amount of time Industry Professional Services Legal Project Overview "In a landmark study, US lawyers with decades of experience in corporate law and contract review were pitted against the LawGeex AI algorithm to spot issues in five Non-Disclosure Agreements (NDAs), which are a contractual basis for most business deals. Twenty US-trained lawyers, with decades of legal experience ranging from law firms to corporations, were asked to issue-spot legal issues in five standard NDAs. They competed against a LawGeex AI system that has been developed for three years and trained on tens of thousands of contracts. The research was conducted with input from academics, data scientists, and legal and machine-learning experts, and was overseen by an independent consultant and lawyer." Reported Results "LawGeex Artificial Intelligence achieved an average 94% accuracy rate, ahead of the lawyers who achieved an average rate of 85%." Technol ogy "The LawGeex AI has been trained to detect issues on more than a dozen different legal contracts, ranging from software agreements to services agreements to purchase orders. This specific research focused solely on NDAs – the most common form of business contract. NDAs are typically used to create a legal obligation to secrecy, and compel those who agree to them to keep information confidential and secure. The LawGeex AI was trained on tens of thousands of NDAs, using custom-built machine learning and deep learning technology. The machine was trained based on an exclusive corpus of documents that presented the LawGeex algorithm with a variety of examples, which allowed it to distinguish between different legal concepts. This level of technology for analyzing legal documents has only been possible with advances in computing over the last five years. Computers convert the text into a numeric representation. The image below is a visualization of how computers read text. Each dot represents one paragraph in the semantic space. The different colors shown represent different legal issues. Pink dots, for example, represent samples of non-compete issues, and purple ones represent governing law sections. LawGeex created proprietary Legal Language Processing (LLP) and Legal Language Understanding (LLU) models for the task. Teams of lawyers and engineers taught LawGeex AI legalese by exposing the AI to a wide range of legal documents. Once the AI learned legalese, legal trainers pointed out the concepts it is required to recognize. The LLP technology allows the algorithm to identify these concepts even if they were worded in ways never seen before. Monitoring concepts, not keywords — LawGeex AI operates in a far more sophisticated manner than a blunt “keyword search.” Keyword searches can be over- and under-inclusive, as words may be absent from relevant documents, or present in irrelevant documents. True AI recognizes a concept however it is phrased or wherever it appears in a document. Unsupervised learning was used for teaching the AI engine the core legalese language. Thereafter, supervised learning, using deep learning multi-layer LSTM and convolution technology, was used to train the system for the fine-tuned issue-spotting. Supervision was performed based on human-annotated documents, using legal experts. A unique augmentation algorithm was applied to boost learning from these examples." Function R And D Core Research And Development Background "The study is a response to a major business problem experienced by every company of any size that requires contracts to engage with partners, suppliers, or vendors. The typical Fortune 1000 company maintains 20,000 to 40,000 active contracts at any given time, while The International Association for Contract & Commercial Management (IACCM) has found that 83% of businesses are dissatisfied with their organization’s contracting process. In addition, NDAs take companies a week or longer to approve – a process that frustrates other departments and slows down deals. Businesses have reduced their reliance on outside law firms, as they want to pay less for legal services, but they are seeing no reduction in legal work." Benefits ​ Data "Five publicly available NDA agreements from the Enron Data Set, which has become the industry standard corpus for common documents for technology providers, scientists, and researchers, were selected by consultant and referee, Christopher Ray. The NDAs were real, everyday agreements used by companies in the US, including Enron, InterGen, Pacific Gas and Electric Company, and Cargill. The five contracts were various forms of commercial NDAs – one 2-page NDA, one 3-page NDA, two 4-page NDAs, and one 5-page NDA."