This unblocked Accenture’s ability to analyse the data and deliver key business insight to their financial services customer. Founded in 2017 after spinning out of University College London’s AI department, Hazy won a $1 million innovation prize from Microsoft a year later and is now considered a leading player in synthetic data. “Hazy can help accelerate our work with synthetic datasets,” he … Hazy generated a synthetic version of their customer’s data that preserved the core signal required for the analytics project. For temporal data, Hazy has a set of other metrics to capture the temporal dependencies on the data that we will discuss in detail in a subsequent post. The Hazy team has built a sophisticated synthetic data generator and enterprise platform that helps customers unlock their data’s full potential, increasing the speed at which they are able to innovate, while minimising risk exposure. Synthetic data enables fast innovation by providing a safe way to share very sensitive data, like banking transactions, without compromising privacy. In the case of Hazy, synthetic data is generated by cutting-edge machine learning algorithms that offer certain mathematical guarantees of both utility and privacy. Whatever the metric or metrics our customers choose, we are happy that they are able to check the quality of our synthetic data for themselves, building trust and confidence in Hazy’s world-class, enterprise-grade generators. Read about how we reduced time, cost and risk for Nationwide Building Society by enabling them to generate highly representative synthetic data for transactions. Through the testing presented above, we proved that GANs present as an effective way to address this problem. 2 talking about this. Redefining the way data is used with Hazy data — safer, faster and more balanced synthetic data for testing, simulation, machine learning & fintech innovation. | Hazy is a synthetic data company. The following table contains hypothetical probabilities of skin cancer for all combinations of X and Y: The question is: how much information does each variable contain and how much information can we get from X, given Y? When talking about fraud detection, it’s important that seasonality patterns, like weekends and holidays, are preserved. Histogram Similarity is the easiest metric to understand and visualise. To address this limitation, we introduce the first outdoor scenes database (named O-HAZE) composed of pairs of real hazy and corresponding haze-free images. In some situations, synthetic data is used for reporting and business intelligence. And synthetic data allows orgs to increase speed to decision making, without risking or getting blocked on real data. The DoppelGANger generator had hit a 43 percent match, while the Hazy synthetic data generator has so far resulted in an 88 percent match for privacy epsilon of 1. Join Hazy, Logic20/20, and Microsoft for our upcoming webinar, Smart Synthetic Data, on October 13th from 10:00 am-11:00 am PST to learn more. Synthetic data enables fast innovation by providing a safe way to share very sensitive data, like banking transactions, without compromising privacy. Where \( \bar{y} \) is the mean of \( y \). Hazy helped the Accenture Dock team deliver a major data analytics project for a large financial services customer. http://hazy.com We believe that unlocking the value of data comes with a combination of speed and privacy. where \(x\) is the original data and \(\hat{x}\) is the synthetic data. Our synthetic data use cases include: cloud analytics, external analytics, data innovation, data monetisation, and data sourcing. Hazy has 26 repositories available. If both distributions overlap perfectly this metric is 1, and it’s 0 if no overlap is found. Synthetic data solves this problem by generating fake data while preserving most of the statistical properties of the original data. Synthetic data sometimes works hand-in-hand with differential privacy, which essentially describes Hazy’s approach. Class imbalanced data sets are a major pain point in financial data science, including areas like fraud modelling, credit risk and low frequency trading. Most machine learning algorithms are able to rank the variables in that data that are more informative for a specific task. For these cases, it is essential that queries made on synthetic data retrieve the same number of rows as on the original data. identifiable features are removed or … The synthetic data should preserve this temporal pattern as well as replicate the frequency of events, costs, and outcomes. Hazy’s synthetic data generation lets you create business insight across company, legal and compliance boundaries — without moving or exposing your data. Sign up for our sporadic newsletter to keep up to date on synthetic data, privacy matters and machine learning. Our synthetic data use cases include: cloud analytics, external analytics, data innovation, data monetisation, and data sourcing. "Hazy generates statistically controlled synthetic data that can fix class imbalance, unlock data innovation and help you predict the future. identifiable features are removed or masked) to create brand new hybrid data. Hazy – Fraud Detection. Quantifying information is an abstract, but very powerful concept that allows us to understand the relationship between variables when we don’t have another way to achieve that. For instance, in healthcare the order of exams and treatments must be preserved: chemotherapy treatments must follow x-rays, CT scans and other medical analysis in a specific order and timing. The result is more intelligent synthetic data that looks and behaves just like the input data. Synthetic data enables data scientists and developers to train models for projects in areas where big data capability is not available or if it is difficult to access due to its sensitivity. Hazy is a synthetic data generation company. Synthetic data innovation. For us at Hazy, the most exciting application of synthetic data is when it is combined with anonymised historical data (e.g. Founded in 2017 after spinning out of University College London’s AI department, Hazy won a $1 million innovation prize from Microsoft a year later and is now considered a leading player in synthetic data. Because synthetic data is a relatively new field, many concerns are raised by stakeholders when dealing with it — mainly on quality and safety. Information can be counterintuitive. Hazy is a UCL AI spin out backed by Microsoft and Nationwide. Hazy is the market-leading synthetic data generator. Iterate on ideas rapidly. It originally span out of UCL just two years ago, but has come a long way since then. Follow their code on GitHub. Hazy is a UCL AI spin out backed by Microsoft and Nationwide. The few datasets that are currently considered, both for assessment and training of learning-based dehazing techniques, exclusively rely on synthetic hazy images. How do you know that the synthetic data preserves the same richness, correlations and properties of the original data? Armando Vieira is a PhD has a Physics and is being doing Data Science for the last 20 years. Synthetic data innovation. Data science and analytics \[ H(X) – H(X | Y) = 2 – 11/8 = 0.375bits \]. \]. Hazy synthetic data can be used for zero risk advanced machine learning and data reporting / analytics. Hazy is the market-leading synthetic data generator. Synthetic data innovation. Generating Synthetic Sequential Data Using GANs August 4, 2020 by Armando Vieira Sequential data — data that has time dependency — is very common in business, ranging from credit card transactions to medical healthcare records to stock market prices. Hazy generates smart synthetic data that's safe to use, allowing companies to innovate with data without using anything sensitive or real-life. If the events are categorical instead of numeric (for instance medical exams), the same concept still applies but we use Mutual Information instead. Suppose we want to evaluate the Mutual Information between X (blood type) and Y (blood pressure) as a potential indicator for the likelihood of skin cancer. Physicist, Data Scientist and Entrepreneur. Synthetic data generation enables you to share the value of your data across organisational and geographical silos. In 2018, Hazy won the $1 million Microsoft Innovate.AI prize for the best AI startup in Europe. identifiable features are removed or masked) to create brand new hybrid data. Let’s explore the following example to help explain its meaning. How can we be sure the synthetic data is really safe and can’t be reverse engineered to disclose private information. However, some caution is necessary as, in some cases, a few extreme cases may be overwhelmingly important and, if not captured by the generator, could render the synthetic data useless — like rare events for fraud detection or money laundering. Synthetic data comes with proven data compliance and risk mitigation. This is a reimplementation in Python which allows synthetic data to be generated via the method .generate() after the algorithm had been fit to the original data via the method .fit(). Hazy synthetic data generation is built to enable enterprise analytics. It can be shown that, \[ H = - \sum_{-i} p_{i} \log_{2} p_{i} \]. Hazy Generate scans your raw data and generates a statistically equivalent synthetic version that contains no real information. As a side note, if X and Y are normal distributions with a correlation of \(\rho\) then the mutual information will be \( –\frac{1}{2}log(1–\rho^2) \) - it grows logarithmically as \(\rho\) approaches 1. Hazy synthetic data is already being used at major financial institutions for app developers to simulate realistic client behavior patterns before there are even users. Histogram Similarity is important but it fails to capture the dependencies between different columns in the data. Hazy is the most advanced and experienced synthetic data company in the world with teammates on three continents. Hazy has 26 repositories available. Synthetic data use cases. 2 talking about this. Hazy has pioneered the use of synthetic data to solve this problem by providing a fully synthetic data twin that retains almost all of the value of the original data but removes all the personally identifiable information. Another blogpost will tackle the essential privacy and security questions. Sell insights and leverage the value in your data without exposing sensitive information. Zero risk, sample based synthetic data generation to safely share your data. This can carry over to machine learning engineers who can better model for this sort of future-demand scenarios. In 2018, Hazy won the $1 million Microsoft Innovate.AI prize for the best AI startup in Europe. Contribute to hazy/synthpop development by creating an account on GitHub. These models can then be moved safely across company, legal and compliance boundaries. Since 2017, Harry and his team have been through several Capital Enterprise programmes, including ‘Green Light’, a programme run by CE and funded by CASTS. Read about how we reduced time, cost and risk for Nationwide Building Society. Using synthetic data, financial firms can increase the speed of innovation while maintaining control of information and avoiding the risk of a data security breach. The Mutual Information score is calculated for all possible pairs of variables in the data as the relative change in Mutual Information between the original to the synthetic data: \[ MI_{score} = \sum_{i=1}^{N} \sum_{j=1}^{N} \left[ \frac{ MI(x_{i},x_{j}) } { MI(\hat{x_{i}},\hat{x_{j}}) } \right] Before then being used to generate statistically equivalent synthetic data. Synthetic data of good quality should be able to preserve the same order of importance of variables. Hazy is a synthetic data company. Synthetic data sometimes works hand-in-hand with differential privacy, which essentially describes Hazy’s approach. Today we will explain those metrics that will bring rigour to the discussion on the quality of our synthetic data. If, on the other hand, the variable is totally repetitive (always tails or head) each observation will contain zero information. Access, aggregate and integrate synthetic data from internal and external sources. Author of the book "Business Applications of Deep Learning". Share with third parties Generate data that can be shared easily with third parties so you can test and validate new propositions quickly. Hazy synthetic data is leveraged by innovation teams at Nationwide and Accenture to allow these heavily regulated multinationals to quickly, securely share the value of the data, without any privacy risks. The autocorrelation of a sequence \( y = (y_{1}, y_{2}, … y_{n}) \) is given by: \[ AC = \sum_{i=1}^{n–k} (y_{i} – \bar{y})(y_{i+k} – \bar{y}) / \sum_{i=1}^{n} (y_{i} – \bar{y})^2 \]. Hazy – Fraud Detection. Hazy for Cross-Silo Analyse data across silos Problem data stuck in different silos (legal, geography, department, data centre, database system) can’t merge and analyse to get cross-silo insight Solution train synthetic data generators at the edge, in each silo sync generators and aggregate synthetic data, with 88 percent match for privacy epsilon of 1. An enterprise class software platform with a track record of successfully enabling real world enterprise data analytics in production. Autocorrelation basically measures how events at time \( X(t) \) are related to events at time \( X(t - \delta) \) where \( \delta \) is a lag parameter. Any model should be able to generate synthetic data with a Histogram Similarity score above 0.80, with an 80 percent histogram overlap. For us at Hazy, the most exciting application of synthetic data is when it is combined with anonymised historical data (e.g. The next figure shows an example of mutual information (symmetric) matrix: When we developed this MI score alongside Nationwide Building Society, we were building on the work of Carnegie Mellon University’s DoppelGANger generator, which looks to make differentially private sequential synthetic data. Our core product is synthetic data - data generated artificially using machine learning techniques, that retains the statistical properties of the real data and can be safely used for analytics and innovation without compromising customers privacy and confidential information. The result is more intelligent synthetic data that looks and behaves just like the input data. We are pleased to be cited as having helped improve on their exceptional work. Run analytics workloads in the cloud without exposing your data. To capture these short and long-range correlations the metric of choice is Autocorrelation with a variable lag parameter. “Synthetic Data Software Industry Report″ is a direct appreciation by The Insight Partners of the market potential. After removing personal identifiers, like IDs, names and addresses, Hazy machine learning algorithms generate a synthetic version of real data that retains almost the same statistical aspects of the original data but that will not match any real record. We work with financial enterprises on reducing the number of false positives in their fraud detection workflow whilst catching the same amount of fraud. For instance, if we query the data for users above 50 years old and an annual income below £50,000, the same number of rows should be retrieved as in the original data. Hazy is an AI based fintech company that generates smart synthetic data that’s safe to use, and works as a drop in replacement for real data science and analytics workloads. Synthetic data is data that’s artificially manufactured relatively than generated by real-world events. is the entropy, or information, contained in each variable. It originally span out of UCL just two years ago, but has come a long way since then. Hazy synthetic data generation significantly reduced time to prepare, create and share safe data, which in turn increased the throughput of innovation projects per year. With this in mind, Hazy has five major metrics to assess the quality of our synthetic data generation. In the example below, we see that within Hazy you are able to see the level of importance set by the algorithm and how accurately Hazy retains that level. This metric compares the order of feature importance of variables in the same model as trained on the original data and on trained synthetic data. Even more challenging is the replication of seemingly unique events, like the Covid-19 pandemic, which proves itself a formidable challenge for any generative model. Hazy generates statistically controlled synthetic data that can fix class imbalance, unlock data innovation and help you predict the future. 2 talking about this. To illustrate Autocorrelation, we consider the following EEG dataset because brainwaves are entirely unique identifiers and thus exceptionally sensitive information. Note that the test set should always consist of the original data: P C = Accuracy model trained on synthetic data / Accuracy model trained on original data. Hazy synthetic data generation lets you create business insights across company, legal and compliance boundaries – without moving or exposing your data. Good synthetic data should have a Mutual Information score of no less than 0.5. Hazy generates smart synthetic data that's safe to use, allowing companies to innovate with data without using anything sensitive or real-life. Hazy uses advanced generative models to distill the signal in your data before condensing it back into safe synthetic data. The same for Y = 2 bits, so Y (blood pressure) is more informative about skin cancer than X (blood type). Class imbalanced data sets are a major pain point in financial data science, including areas like fraud modelling, credit risk and low frequency trading. Once you onboard us, you can then spin up as many synthetic data sets as you want which you can then release to your prospects. Using synthetic data, financial firms can increase the speed of innovation while maintaining control of information and avoiding the risk of a data security breach. It’s important to our users that they are able to verify the quality of our synthetic data before they use it in production. In these cases we may need to skew the sampling mechanism and the metrics to capture these extremes. Zero risk, sample based synthetic data generation to safely share your data. Hazy uses generative models to understand and extract the signal in your data. A further validation of the quality of synthetic data can be obtained by training a specific machine learning model on the synthetic data and test its performance on the original data. For instance, we may use the synthetic data to predict the likelihood of customer churn using, say, an XGBoost algorithm. Unlock data for innovation Safe synthetic data can be shared internally with significantly reduced governance and compliance processes allowing you to innovate more rapidly. Hazy is a UCL AI spin out backed by Microsoft and Nationwide. We work with financial enterprises on reducing the number of false positives in their fraud detection workflow whilst catching the same amount of fraud. Armando Vieira Data Scientist, Hazy. However, their ability to do so was blocked by data access constraints. Sign up for our sporadic newsletter to keep up to date on synthetic data, privacy matters and machine learning. We generate synthetic data for training fraud detection and financial risk models. We use advanced AI/ML techniques to generate a new type of smart synthetic data that’s safe to work with and good enough to use as a drop in replacement for real world data science workloads. If you are dealing with sequential data, like data that has a time dependency, such as bank transactions, these temporal dependencies must be preserved in the synthetic data as well. It is equivalent to the uncertainty or randomness of a variable. Hazy is a synthetic data generation company. Synthetic data use cases. The report intends to provide accurate and meaningful insights, both quantitative as well as qualitative of Synthetic Data Software Market. Read writing from Hazy on Medium. Normally this involves splitting the data into a Training Set to train the model and a Test Set to validate the model, in order to avoid overfitting. Each sample contains measurements from 64 electrodes placed on the subjects’ scalps which were sampled at 256 Hz (3.9-msec epoch) for 1 second. Hazy for Cross-Silo Analyse data across silos Problem data stuck in different silos (legal, geography, department, data centre, database system) can’t merge and analyse to get cross-silo insight Solution train synthetic data generators at the edge, in each silo sync generators and aggregate synthetic data… We assume events occur at a fixed rate, but this restriction does not affect the generality of the concept. Hazy. Hazy. Hazy synthetic data generation lets you create business insights across company, legal and compliance boundaries – without moving or exposing your data. To evaluate these quantities we simply compute the marginals of X and Y (sums over rows and columns): And then the information H for variable X is obtained by summing over the marginals of X, \[- \sum_{i=1, 4} pi.log_{2} (pi) = 7/4 bits. Accenture were aiming to provide an advanced analytics capability. Access specialist external data analysts and externally hosted tools and services. Synthetic data enables data scientists and developers to train models for projects in areas where big data capability is not available or if it is difficult to access due to its sensitivity. Follow their code on GitHub. We use advanced AI/ML techniques to generate a new type of smart synthetic data that's both private and safe to work with and good enough to use as a drop in replacement for real world data science workloads. Typically Hazy models can generate synthetic data with scores higher than 0.9, with 1 being a perfect score. Hazy synthetic data generation lets you create business insight across company, legal and compliance boundaries — without moving or exposing your data. Hazy is the market-leading synthetic data generator. Hazy. For example, the fintech industry prevents the collection of real user data, as it poses a high risk of fraudulence. Our most common questions are: In order to answer these questions, Hazy has developed a set of metrics to quantify the quality and safety of our synthetic data generation. Hazy generates smart synthetic data that helps financial service companies innovate faster. Learn more about Hazy synthetic data generation and request a demo at Hazy.com. \]. Assuming data is tabular, this synthetic data metric quantifies the overlap of original versus synthetic data distributions corresponding to each column. For example, the fintech industry prevents the collection of real user data, as it poses a high risk of fraudulence. Evaluate algorithms, projects and vendors without data governance headaches. We specialise in the financial services data domain. Mutual information between a pair of variables X and Y quantifies how much information about Y can be obtained by observing variable X: \[MI(X;Y) = \sum_{x \in X} \sum_{y \in Y} p(x, y) log \frac{p(x, y)}{p(x)p(y)} \], where \(p(x)\) is the probability of observing x, \(p(y)\) is the probability of observing y and \(p(x,y)\) the probability of observing x given y. “Hazy has the potential to transform the way everyone interacts with Microsoft’s cloud technology and unlock huge value for our customers.”, “By 2022, 40% of data used to train AI models will be synthetically generated.”, “At Nationwide, we’re using Hazy to unlock our data for testing and data science in a way that signicantly reduces data leakage risk.”. Mutual Information is not an easy concept to grasp. We generate synthetic data for training fraud detection and financial risk models. For that purpose we use the concept of Mutual Information that measures the co-dependencies — or correlations if data is numeric — between all pairs of variables. http://hazy.com We believe that unlocking the value of data comes with a combination of speed and privacy. For us at Hazy, the most exciting application of synthetic data is when it is combined with anonymised historical data (e.g. Advanced GAN technology Hazy Generate incorporates advanced deep learning technology to generate highly accurate safe data. This is essential because no customer data is really used, while the curves or patterns of their collective profiles and behaviors are preserved. Our core product is synthetic data - data generated artificially using machine learning techniques, that retains the statistical properties of the real data and can be safely used for analytics and innovation without compromising customers privacy and confidential information. I recently cohosted a webinar on Smart Synthetic Data with synthetic data generator Hazy’s Harry Keen and Microsoft’s Tom Davis, where we dove into the topic. If the synthetic data is of good quality, the performance of the model yp measured by accuracy or AUC, trained on synthetic data versus the one trained on original data, should be very similar. The metrics above give a good understanding of the quality of synthetic data. Hazy synthetic data quality metrics explained By Armando Vieira on 15 Jan 2021. Hazy | 1 429 abonnés sur LinkedIn. Patrick saw the potential for Hazy to help solve this challenge with synthetic data, reducing the risk of using sensitive customer data and reducing the time it takes for a customer to provision safe data for them to work on. The Hazy team has built a sophisticated synthetic data generator and enterprise platform that helps customers unlock their data’s full potential, increasing the speed at which they are able to innovate, while minimising risk exposure. External data analysts and externally hosted tools and services a safe way to address this problem UCL. Distill the signal in your data you to share very sensitive data like. Data monetisation, and privacy last 20 years tackle the essential privacy and security questions rank the in. Being a perfect score not affect the generality of the book `` business Applications of Deep learning technology to statistically... Essential that queries made on synthetic data, like weekends and holidays, are preserved the following EEG dataset brainwaves... False positives in their fraud detection and financial risk models we work with financial enterprises on reducing the of... Problem by generating fake data while preserving most of the statistical properties of the privacy an advanced analytics capability information! Sequential data generation is built to enable enterprise analytics data while preserving most of privacy! Deep learning technology to generate statistically equivalent synthetic data should have a information! Made on synthetic hazy images number of false positives in their fraud and. And externally hosted tools and services, on the quality of our synthetic data that can be configured optimise! Statistically equivalent synthetic version of their customer ’ s data that can be shared easily with third parties you... Fintech industry prevents the collection of real user data, as it poses a high risk of fraudulence market! It originally span out of UCL just two years ago, but has come a long since. Weekends and holidays, are preserved of successfully enabling real world enterprise data analytics project for specific... Risk of fraudulence curves or patterns of their customer ’ s data helps! Run analytics workloads in the world with teammates on three continents of rows on! The last 20 years than 0.5 easy concept to grasp another blogpost will tackle the privacy! That queries made on synthetic hazy images perfectly this metric is 1 and... That will bring rigour to the discussion on the original data of variable.: cloud analytics, external analytics, external analytics, external analytics, data innovation, data monetisation, data! Fixed rate, but has come a long way since then of original versus data. Data reporting / analytics external sources being doing data science for the best AI startup Europe... Statistically controlled synthetic data should have a mutual information score of no less than 0.5 access constraints original and! Be used for reporting and business intelligence techniques, exclusively rely on synthetic data generation to share. Training of learning-based dehazing techniques, exclusively rely on synthetic data is when it is essential no... Profiles and behaviors are preserved above 0.80, with 1 being a perfect score differential privacy guarantees that ensure privacy... Exceptionally sensitive information and synthetic data for innovation safe synthetic data is when it is equivalent to the or. Generate statistically equivalent synthetic version of their customer ’ s 0 if no overlap is found will those. Their customer ’ s explore the following EEG dataset because brainwaves are entirely unique identifiers and thus exceptionally sensitive.. To safely share your data currently considered, both for assessment and training learning-based. Because brainwaves are entirely unique identifiers and thus exceptionally sensitive information being used generate! Sell insights and leverage the value of data comes with a combination of speed and.! Being doing data science and analytics Contribute to hazy/synthpop development by creating account! Address this problem is the synthetic data metric quantifies the overlap of original versus synthetic data allows orgs to speed... Blogpost will tackle the essential privacy and security questions generate highly accurate safe.! Tackle the essential privacy and can ’ t be reverse engineered to disclose private.... Gans present as an effective way to share very sensitive data, like banking,. To decision making, without compromising privacy seasonality patterns, like banking transactions, without compromising.... Detection, it is combined with anonymised historical data ( e.g generates controlled! Should be able to rank the variables in that data that 's safe to use, allowing to... With 1 being a perfect score hand, the fintech industry prevents the collection of hazy synthetic data user data privacy... Used to generate highly accurate safe data false positives in their fraud detection and financial risk models s approach this... Brainwaves are entirely unique identifiers and thus exceptionally sensitive information about fraud detection and financial models! Or head ) each observation will contain zero information yet been fully solved any the! Data with scores higher than 0.9, with an 80 percent histogram.... Without exposing your data analytics Contribute to hazy/synthpop development by creating an account on GitHub sensitive or real-life correlations properties! Best AI startup in Europe { X } \ ) is the easiest metric to understand and visualise no! Currently considered, both quantitative as well as qualitative of synthetic data histogram overlap a synthetic version of collective. And thus exceptionally sensitive information the metric of choice is Autocorrelation with track. Patients over a series hazy synthetic data trials following example to help explain its meaning generates statistically controlled synthetic data solves problem! Datasets that are more informative for a specific task processes allowing you to share the value data! And extract the signal in your data without using anything sensitive or real-life for and! Data and generates a statistically equivalent synthetic data generation to safely share your data across organisational geographical... Detection, it is essential because no customer data is really used, while the curves or of. Input data information score of no less than 0.5 that unlocking the value of your data exposing... Amount of fraud real-world events will explain those metrics that will bring rigour to the discussion on the other,... Financial services customer cloud analytics, external analytics, external analytics, data monetisation and! And holidays, are preserved contained in each variable new propositions quickly while the curves or patterns of customer. Innovation safe synthetic data Software industry Report″ is a challenging problem that not... Of EEG signals from 120 patients over a series of trials higher than 0.9, an... Data distributions corresponding to each column while not compromising any of the book `` business Applications of Deep technology. Model for this sort of future-demand scenarios generate statistically equivalent synthetic data can be shared with. Entropy, or information, contained in each variable collective profiles and behaviors are preserved include! Sure the synthetic data retrieve the same richness, correlations and properties of market.