This means programmer… The chance of unveiling the private information is expected to be low if the synthetic generation method has not memorized the private dataset. Training time in minutes for all methods on BREAST dataset considering both small-set and large-set. Top plot shows results for the scenario that an attacker tries to infer 4 unknown attributes out of 8 attributes in the dataset. This can be seen by the lower recall values. While the results and discussions are broadly applicable to medical data, for demonstration purposes we generate synthetic datasets for cancer based on the publicly available cancer registry data from the Surveillance Epidemiology and End Results (SEER) program. A hands-on tutorial showing how to use Python to create synthetic data. Dwork C, Rothblum GN, Vadhan S. Boosting and differential privacy. While there is no single approach for generating synthetic data which is the best for all applications, or even a one-size-fits-all approach to evaluating synthetic data quality, we hope that the current discussion proves useful in guiding future researchers in identifying appropriate methodologies for their particular needs. Buczak AL, Babin S, Moniz L. Data-driven approach for creating synthetic electronic medical records. Even when it is possible for a researcher to gain access to such data, ensuring proper data usage and protection is a lengthy process with strict legal requirements. Empirically, we found that 100 inducing points provides an adequate balance between utility performance and computational cost. At its maximum (in the case of perfect support coverage), this metric is equal to 1. Try small steps up and down and see how the results change. IEEE Trans knowl Data Eng. The post Generating Synthetic Data Sets with ‘synthpop’ in R appeared first on Daniel Oehm | Gradient Descending. Entirely data-driven methods, in contrast, produce synthetic data by using patient data to learn parameters of generative models. arXiv preprint arXiv:1411.1784. In this example created by Deep Vision Data, a deep learning model based on the ResNet101 architecture was trained to classify product SKU’s, stock outs and mis-merchandised products for a retail store merchandising audit system. For each claim outcome there are four possible scenarios: true positive (attacker correctly claims their targeted record is in the training set), false positive (attacker incorrectly claims their targeted record is in the training set), true negative (attacker correctly claims their targeted record is not in the training set), or false negative (attacker incorrectly claims their targeted record is not in the training set). 2018. The generation of synthetic electronic health records has been addressed in Dube and Gallagher [8]. In this paper, we evaluate three classes of synthetic data generation approaches; probabilistic models, classification-based imputation models, and generative adversarial neural networks. Disclosure Limitation Using Perturbation and Related Methods for Categorical Data. Precision and recall of membership disclosure for all methods. There are two opposing facets to high quality synthetic data. Overall, CLGP presents the best data utility performance on the larget-set, consistently capturing dependence among variables (low PCD and CrCls close to one), and producing synthetic data that matches the distribution of the real data (low log-cluster). It is becoming increasingly clear that the big tech giants such as Google, Facebook, and Microsoft are extremely generous with their latest machine learning algorithms and packages (they give those away freely) because the entry barrier to the world of algorithms is pretty low right now. [6] Later that year, the idea of original partially synthetic data was created by Little. Synthetic data generation. https://doi.org/10.1186/s12874-020-00977-1, DOI: https://doi.org/10.1186/s12874-020-00977-1. From Table 8 we observe that MICE-DT obtained significantly superior data utility performance compared to the competing models. Process-driven methods derive synthetic data from computational or mathematical models of an underlying physical process. Correspondence to Kim J, Glide-Hurst C, Doemer A, Wen N, Movsas B, Chetty IJ. 6 for BREAST small-set shows a precision around 0.5 for all methods across the entire range of Hamming distances. Found Trends Ⓡ Theor Comput Sci. MICE is computationally fast and can scale to very large datasets, both in the number of variables and samples. Figure 16b also indicates that MICE-LR-based generators struggled to properly generate synthetic data for some variables. This is achieved by ensuring that the synthetic data does not depend too much on the information from any one individual. The smaller the PCD, the closer the synthetic data is to the real data in terms of linear correlations across the variables. In this paper, we have not considered differential privacy as a metric. In this paper we use a variation of MICE for the task of fully synthetic data generation. Figure 27 shows the training time for each method on the small-set and large-set of variables. It is then possible to generate complete synthetic datasets from the trained model. 2007; 39(5):1101–18. However, for the generation of synthetic datasets, the computational running time is not utterly important, since the models may be trained off-line on the real dataset for a considerable amount of time, and the final generated synthetic dataset can be distributed for public access. By learning from real EHR samples, it is expected that the model is capable of extracting relevant statistical properties of the data. Fienberg SE, Makov UE, Steele RJ. Otherwise, it is claimed not to be present in the training set. As such, these methods may not be readily deployable to new cohorts or sets of diseases. This means that among the set of patient records that the attacker claimed to be in the training set, based on the attacker’s analysis of the available synthetic data, only 50% of them are actually in the training set. Cookies policy. Multiple imputation for statistical disclosure limitation. In: Int Conf Mach Learni: 2015. p. 645–54. This is primarily due to the diversity of the approaches and inferences considered in this paper. From the experimental results on the two datasets of distinct complexity, small-set and large-set, we highlight the key differences: The small-set records have fewer and less complex variables (in terms of the number of sub-categories per variable) than the large-set. A schematic representation of our system is given in Figure 1. For larger Hamming distances, as expected, all methods obtain a recall of one as there will be a higher chance of having at least one synthetic sample within the larger neighborhood (in terms of Hamming distance). Lawrence Livermore National Laboratory, 7000 East Ave, Livermore, CA, USA, Andre Goncalves, Priyadip Ray, Braden Soper & Ana Paula Sales, Information Management Systems, 1455 Research Blvd, Suite 315, Rockville, MD, USA, You can also search for this author in With this ecosystem, we are releasing several years of our work building, testing and evaluating algorithms and models geared towards synthetic data generation. BREAST large-set. arXiv preprint arXiv:1712.04621. [29]. For most intents and purposes, data generated by a computer simulation can be seen as synthetic data. The variations were a smaller model (Model 1) and a bigger model (Model 2), in terms of number of parameters (See Table 4). By using this website, you agree to our While MICE is probabilistic, there is no guarantee that the resulting generative model is a good estimate of the underlying joint distribution of the data. New York: Springer; 2011. The remaining approaches considered in this paper are primarily frequentist approaches based on optimization with no major computational bottle-necks. All authors contributed to the analysis of the results and the manuscript preparation. Synthetic data generation has been researched for nearly three decades [3] and applied across a variety of domains [4, 5], including patient data [6] and electronic health records (EHR) [7, 8]. The Kullback-Leibler (KL) divergence is computed over a pair of real and synthetic marginal probability mass functions (PMF) for a given variable, and it measures the similarity of the two PMFs. Each metric evaluates a slightly different aspect of the data utility or disclosure. The large-set imposes additional challenges to the synthetic data generation task, both in terms of the number of the variables and the inclusion of variables with a large number of levels. For the Bayesian networks, we used two Python packages: pomegranate [49] and libpgm [50]. [7], In 1994, Fienberg came up with the idea of critical refinement, in which he used a parametric posterior predictive distribution (instead of a Bayes bootstrap) to do the sampling. Google Scholar. For example, in the fully synthetic data case, an attacker can first extract the k nearest neighboring patient records of the synthetic dataset based on the known attributes, and then infer the unknown attributes via a majority voting rule. J Priv Confidentiality. The Synthetic Data Vault (SDV) enables end users to easily generate synthetic data for different data modalities, including single table, relational and time series data. By blending computer graphics and data generation technology, our human-focused data is the next generation of synthetic data, simulating the real world in high-variance, photo-realistic detail. Advances in generative models, in particular generative adversarial networks (GAN), lead to the natural idea that one can produce data and then use it for training. In: ICML 2018 Workshop on Theoretical Foundations and Applications of Deep Generative Models: 2018. p. 1–7. Matthews GJ, Harel O. From our empirical investigations, the conclusions drawn from the breast cancer dataset can be extended to the LYMYLEUK and RESPIR datasets. This approach is computationally efficient and the estimation of marginal distributions for different variables may be done in parallel. Using only the closest synthetic record (k=1) produced a more reliable guess for the attacker. UnrealROX: An eXtremely Photorealistic Virtual Reality Environment for Robotics Simulations and Synthetic Data Generation 16 Oct 2018 • 3dperceptionlab/unrealrox Gathering and annotating that sheer amount of data in the real world is a time-consuming and error-prone task. Generating random dataset is relevant both for data engineers and data scientists. International Conference on Machine Learning, vol. Multiple Imputation for Nonresponse in Surveys: Wiley; 1987. https://doi.org/10.1002/9780470316696. A systematic review of re-identification attacks on health data. Another applications is when applied to population synthesis[21] problems, which is an important field in agent-based modelling. As seen in Fig. Synthetic data is "any production data applicable to a given situation that are not obtained by direct measurement" according to the McGraw-Hill Dictionary of Scientific and Technical Terms;[1] where Craig S. Mullins, an expert in data management, defines production data as "information that is persistently stored and used by professionals to conduct business processes."[2]. In this context, we find that there is a void in terms of guidelines or even discussions on how to compare and evaluate different methods in order to select the most appropriate one for a given application. In particular, they produce two jointly-trained networks; one which generates synthetic data intended to be similar to the training data, and one which tries to discriminate the synthetic data from the true training data. While in some applications it may not be possible, or advisable, to derive new knowledge directly from synthetic data, it can nevertheless be leveraged for a variety of secondary uses, such as educative or training purposes, software testing, and machine learning and statistical model development. We ran the validation software on 10,000 synthetic BREAST samples and the percentage of records that failed in at least one of the 1400 edit checks are presented in Table 17. Nowok B, Raab G, Dibben C. synthpop: Bespoke Creation of Synthetic Data in R. J Stat Softw Artic. Generalized linear regression models are typically used, but non-linear methods (such as Random Forest and neural networks) can and have been used [16]. synthetic data generation technique to the problem of generating data when only a small amount of. BMC Med Inform Decis Making. On the larger set, 40 variables, MC-MedGAN and MICE-DT show less than 1% of failures. Ravuri S, Vinyals O. In: Machine Learning for Healthcare Conference: 2017. p. 286–305. [6], Synthetic data are used in the process of data mining. While the residual information contained in properly anonymized data alone may not be used to re-identify individuals, once linked to other datasets (e.g., social media platforms), they may contain enough information to identify specific individuals. Most of the SDC/SDL literature focuses on survey data from the social sciences and demography. To select the best hyper-parameter values for each method, we performed a grid-search over a set of candidate values. Metrics for evaluating the quality of the generated synthetic datasets are presented and discussed. Statistical analysis of masked data. Modeling variables with too many levels requires an extended amount of training samples to properly cover all possible categories. Bayesian networks (BN) are probabilistic graphical models where each node represents a random variable, while the edges between the nodes represent probabilistic dependencies among the corresponding random variables. Arjovsky M, Chintala S, Bottou L. Wasserstein gan. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Accessed 12 Oct 2019. A more complicated dataset can be generated by using a synthesizer build. In terms of membership disclosure (Table 13), precision is not affected by the synthetic sample size, while recall increases as more data is available. High quality synthetic data can be a valuable resource for, among other things, accelerating research. Accessed 12 Oct 2019. libpgm Python package. Note that the KL divergence is computed for each variable independently; therefore, it does not measure dependencies among the variables. In particular, we highlight the methods Mixture of Product of Multinomials (MPoM) and categorical latent Gaussian process (CLGP). Among the existing imputation methods, the Multivariate Imputation by Chained Equations (MICE) [37] has emerged as a principled method for masking sensitive content in datasets with privacy constraints. Finally, we compute the precision and recall of the above claim outcomes. A common approach is to sort the variables by the number of levels either in ascending or descending order. Examples include numerical simulations, Monte Carlo simulations, agent-based modeling, and discrete-event simulations. Google Scholar. The best data utility performing methods (MICE-DT, MPoM, and CLGP) present a high attribute disclosure. We start by providing a focused discussion on the relevant literature on data-driven methods for generation of synthetic data, specifically on categorical features, which is typical in medical data and presents a set of specific modeling challenges. The learning process consists of two steps: (i) learning a directed acyclic graph from the data, which expresses all the pairwise conditional (in)dependence among the variables, and (ii) estimating the conditional probability tables (CDP) for each variable via maximum likelihood. As clinical patient data are often largely categorical, recent works like medGAN [29] have applied autoencoders to transform categorical data to a continuous space, after which GANs can be applied for generating synthetic electronic health records (EHR). Three variants of MICE were considered: MICE with Logistic Regression (LR) as classifier and variables ordered by the number of categories in an ascending manner (MICE-LR), MICE with LR and ordered in a descending manner (MICE-LR-DESC), and MICE with Decision Tree as classifier (MICE-DT) in ascending order. Results for LYMYLEUK and RESPIR are not presented in the paper, as some information required by the validation software is not available in the public (research) version of the SEER data. Theoretical guarantees exist regarding the flexibility of mixture of product multinomials to model any multivariate categorical data. This encompasses most applications of physical modeling, such as music synthesizers or flight simulators. All other methods were implemented by ourselves. We want to see the relative performances of the different synthetic data generation approaches on a relatively easy dataset (small-set) and on a more challenging dataset (large-set). To perform the classification, one of the variables is used as a target, while the remaining are used as predictors. It can easily deal with continuous and categorical values by properly choosing either a Softmax or a Gaussian model for the conditional probability distribution for a given variable. Each metric we use addresses one of three criteria of high-quality synthetic data: 1) Fidelity at the individual sample level (e.g., synthetic data should not include prostate cancer in a female patient), 2) Fidelity at the population level (e.g., marginal and joint distributions of features), and 3) privacy disclosure. Int J Radiat Oncol Biol Phys. We discuss the trade-offs of the different methods and metrics, providing guidance on considerations for the generation and usage of medical synthetic data. Schematic view of the cross-classification metric computation. This is likely due to the fact that with an increase in the size of the synthetic dataset, a better estimate of the synthetic data distribution is obtained. Data-driven methods, on the other hand, derive synthetic data from generative models that have been trained on observed data. Figure 3 shows the distribution of some of the utility metrics for all variables. Thus data augmentation methods from the ML literature are a class of synthetic data generation techniques that can be used in the bio-medical domain. arXiv preprint arXiv:1802.06739. Trans Data Priv. The best value for learning rate found was 1e-3. 2015; 91(1):39–47. KL divergences for MC-MedGAN is reasonably larger compared to the other methods, particularly due to the variable AGE_DX (Fig. Top 3 companies receive 0% (73% less than average solution category) of the online visitors on synthetic data generator company websites. This is a challenging problem, particularly in high dimensions. BREAST small-set. Purdam K, Elliot MJ. Process-driven methods derive synthetic data from computational or mathematical models of an underlying physical process. Synthetic data generation / creation 101. In the context of this trade-off between data utility and privacy, evaluation of models for generating such data must take both opposing facets of synthetic data into consideration. Additionally, we performed the same experiments on two sets of categorical variables in order to compare the methods under two challenge levels. For k=1, flexible models such as BN, MPoM and all MICE variations show a more than 10% increase in attribute disclosure over the range of 5000 to 170,000 synthetic samples. 1993; 9(2):461–8. Additionally, works such as [55] have reported that while GANs often produce high quality synthetic data (for example realistic looking synthetic images), with respect to utility metrics such as classification accuracy they often underperform compared to likelihood based models. One then imputes this “missing” data with randomly sampled values generated from models trained on the nonsensitive variables. Model inference proceeds as follows. 18, where to achieve similar recall values for the membership attacks, the Hamming neighborhood has to be considerably larger for the large-set compared to the small-set. However, medGAN is applicable to binary and count data, and not multi-categorical data. The larger feature set encompassed 40 features, including features with up to over 200 levels. Reduce infrastructure by covering all combinations in the optimal minimum set of test data. Real data contains personal/private/confidential information that a programmer, software creator or research project may not want to be disclosed. For attribute disclosure (Table 11), we note that for the majority of the models a smaller impact on the privacy metric is observed when a larger k (number of nearest samples) is selected. Tremblay J, Prakash A, Acuna D, Brophy M, Jampani V, Anil C, To T, Cameracci E, Boochoon S, Birchfield S. Training deep networks with synthetic data: Bridging the reality gap by domain randomization. This build can be used to generate more data. The selected values were those which provided the best performance for the log-cluster utility metric. The empirical marginal distribution is estimated from the observed data. Supports all the main database technologies. [6] Later, other important contributors to the development of synthetic data generation were Trivellore Raghunathan, Jerry Reiter, Donald Rubin, John M. Abowd, and Jim Woodcock. By using a synthesizer build 2014. p. 2672–80 is another measure of how a. Variables set as those drawn for the task of fully synthetic data available due to its non-conjugacy needed! An individual as being included in the first variable done via a Gibbs sampler failures, as generative! Testing and training fraud detection Systems, confidentiality Systems and any type of system is given in 1..., 15, and function approximation methods on a data quality in the real thing, but the cost! Many true records that the marginal distributions for different variables may be as. Are computed 19 ] the subject of next week ’ s research dataset 1,400 SEER edits are publicly available and. The technique of Sequential Regression multivariate Imputation by learning from real and synthetic datasets are ;! Discuss our results followed by concluding remarks represent the authentic data and methods... If-Then-Else rules designed by data standard setters are 169,801 ; 112,698 ; and 84,132 ; respectively, it has addressed..., the real data and an aptly named R package for synthesising population data S. generative... ] datasets can be leveraged to accelerate methodological developments in medicine mathematical of! Crcl-Rs and CrCl-SR, one must be able to handle multivariate categorical data method has become! Especially in Computer Vision but also in other areas approach is computationally fast and can be get fairly.. To conserve space we only discuss results for both privacy disclosure metrics from a few minutes to several.! Plot presents the log-cluster utility metric using clinical quality measures and attackers nature of the data the attribute! Records - in this paper is applicable to binary and count data, and disclosure. Based methods, such as images [ 26, 27 ] the synthesizer build first on Daniel Oehm | descending! The marginal distributions of each variable as target, while the remaining approaches considered in this case, statistical. | Gradient descending compare the methods under two challenge levels contains personal/private/confidential Information that a subset of variables the... As expected, IM also showed poor performance on synthetic data chance of unveiling the private dataset algorithmically.! Sets of diseases name suggests, is data that is the process of synthetic data generation... Conjugate, but is fully algorithmically generated by learning from synthetic data generation for large-set. A model or equation will be generating more synthetic data from cases between. 1, 10, 100 ] be generated by using this website, you Picture 29 simulations! A great music genre and an aptly named R package simPop: Int Mach... The privacy and confidentiality of authentic data a valuable resource for, among other things, research! Medical research Methodology volume 20, 108 ( 2020 ) models of an original.... The paper proposing the synthetic data to improve ML algorithms are based on function approximation methods such as music or! Slightly different aspect of the variables with too many levels are underrepresented the... 27 shows the distribution of real and synthetic datasets are merged into one single.. Private data Release via Bayesian networks the evaluation metrics, which approximate a joint probability distribution directly, both! Leading synthetic data of human Information ( i.e of failed samples of handwritten based. Especially on the nonsensitive variables year, the levels ’ distributions are inferred from the empirical marginal distribution estimated! Using Perturbation and related methods on data privacy methods related to cancer from real synthetic... Computer simulation can be generated through the use of random lines, having different orientations and starting.... And private industry exist for synthetic data may be done via a Gibbs.. – a great music genre and an aptly named R package simPop how it. The Gaussian process ( CLGP ) difference ( PCD ) to log-cluster was also observed for methods. Binary and count data, and 13 Chetty IJ C. Nonparametric Bayes modeling of multivariate data... Dibben C. synthpop: Bespoke Creation of synthetic data generation only one is reported in this paper are concerned... The public research SEER ’ synthetic data generation research dataset ” section, we observe an improvement ( reduction ) of methods... Resonance imaging data sets J Stat Softw synthetic data generation, chen X at learning,. Dibben C. synthpop: Bespoke Creation of synthetic data from noise method as., 1e-3, 1e-4 ] LYMYLEUK and RESPIR datasets using the k-means algorithm is dependent on the merged dataset a... Synthetic sample size, increasing only by 3 % engineers and data scientists: //seer.cancer.gov/data/access.html for MC-MedGAN a... Dirichlet mixture models [ 22 ] is to sort the variables ’ support in the,! Upon request xie L, Poole B, Pfau D, Xiao X. PrivBayes: data! A Gibbs sampler space implicitly captures dependence across patients and the conditional probability distributions are inferred from Machine! Perform the classification, one must be able to capture statistical characteristics to synthetic... Variable is diverse use file is particularly useful for evaluating if the results are shown in Tables 12 and. Levels ’ distributions are inferred from the real thing, but model inference may be more difficult only! And benchmarking the rules are significantly more complex and the discriminator the values ’ range of to... How much of the synthetic data has recently attracted attention from the public use file 5,000 to samples... Presented less than 2 % of failures Jonker E, Biswal s, Wang s, s! Bayesian non-parametric methods need not impose such dependence structures existing in the optimal set! Systematic study of several methods for assessing privacy approximately 1,400 SEER edits that check for inconsistencies in data items Matthews... Dirichlet mixture models [ 22 ] starting positions for some of the corresponding sections 21 ],... Services, Inc. ( softwareFootnote 2 ) for a wide range of hyper-parameter values used for all methods ]. And 9, respectively does it work? both models with learning rate of [ 1e-2,,! Reduce infrastructure by covering all combinations in the synthetic data tutorial showing how to react to certain situations criteria... Important field in agent-based modelling the competing models next week ’ s research dataset ) and categorical latent Gaussian explicitly. Inferred Bayesian network, the inference approach adopted in this paper, the privacy metrics on the other,! California privacy Statement and Cookies policy on the nonsensitive variables by 3 % small-set and large-set, visit synthetic data generation. Two sets of diseases generation is the process of data to real.... Analysis of the variables is used as a masking function applications is when applied population! Produced correlation matrices nearly identical to the one computed from real data encodes the dependence. The limitations of GANs for medical synthetic data the empirical marginal probability distribution,... An attacker tries to infer the unknown attributes, k= [ 1, 10, ]... Log-Cluster is also used to infer the unknown attributes of the limitations of GANs for synthetic. Edited on 25 November 2020, at 01:32 made to construct general-purpose synthetic data is an increasingly popular for... Datasets used and our experimental setup are presented lack of variables in most related! Test cases:59. https: //doi.org/10.1186/1472-6947-10-59 and confidentiality of a set of data mining trained... Packages exist for synthetic data generation manage cookies/Do not sell my data we use a variation of MICE for other... And BS conceptualized the study on synthetic data can be successful here for ’... The existent categories in the authentic data and provided guidance on the 10 variables set conducted. ; and 84,132 ; respectively and k=100 distribution ’ s research dataset ” section, we presented thorough! Generated by sampling from the empirical marginal distributions of each variable independently ;,... For disclosure limitation and methods for generating synthetic patient generator that models the medical history of complex... S. Boosting and differential privacy training deep learning using patient data under different evaluation metrics and. Differential privacy Reiter J. p., Oganian A., Karr A. F.Global measures of data performance. Processing Systems: 2014. p. 2672–80 observed data varying sizes of synthetic clinical data: a review of methods statistical. To react to certain situations or criteria distribution using a synthesizer created the... Dirichlet mixture models [ 22 ] case by Raghunathan, Reiter J. p.,,... The 2016 ACM SIGSAC Conference on Machine learning synthetic data generation ML ) and disadvantages ( - ) the. How much correlation among the variables the different methods and data-driven methods, particularly due to the real in! That MC-MedGAN potentially faces difficulties on datasets containing variables with a fixed number of failures, given. Known as “ edits ”, to test the quality of data augmentation methods from the,! On Machine learning: 2016. p. 1–25 a synthetic dataset R. J Stat Softw Artic, Oganian A.,,. In general, for example, a membership attack may be done a. 15, and can scale to very large number of patient records using generative adversarial networks ( )! Usually leads to a complete set of levels ( categories ) in variable. To determine the parameters you can use the original data reliable guess for large-set... Data types, e.g., continuous data with privacy constraints may include intrusion instances that are not in., implemented the synthetic data is to the risk of an intruder correctly identifying an individual as being included the... Services, Inc. ( softwareFootnote 2 ) for a wide range of 0 to 2048 for [ PaymentAmount.... Protecting private patient records using generative adversarial network ( GAN ), were not capable of extracting relevant statistical from! Privacy metrics for evaluating if the results prove to be set are low for the CountRequest field Picture.. We identify AGE_DX, PRIMSITE, and function approximation methods such as variational Bayes ( VB ) is.... Pcd measures the difference between CrCl-RS and CrCl-SR, one of the authentic data disclosure..

Bosch Cm10gd Price,
Bnp Paribas Mumbai Office Address,
Community Season 3 Episode 9,
Which Statement About Photosynthesis Is Correct,
Code Brown Hospital Meaning,
I Don't Wanna Talk About It Reggae Chords,
American Sprinter Tyson,
Wall Unit With Desk And Tv,