AIgean : An Open Framework for Deploying Machine Learning on Heterogeneous Clusters

AIgean , pronounced like the sea, is an open framework to build and deploy machine learning (ML) algorithms on a heterogeneous cluster of devices (CPUs and FPGAs). We leverage two open source projects: Galapagos , for multi-FPGA deployment, and hls4ml , for generating ML kernels synthesizable using Vivado HLS. AIgean provides a full end-to-end multi-FPGA/CPU implementation of a neural network. The user supplies a high-level neural network description, and our tool flow is responsible for the synthesizing of the individual layers, partitioning layers across different nodes, as well as the bridging and routing required for these layers to communicate. If the user is an expert in a particular domain and would like to tinker with the implementation details of the neural network, we define a flexible implementation stack for ML that includes the layers of Algorithms, Cluster Deployment & Communication, and Hardware. This allows the user to modify specific layers of abstraction without having to worry about components outside of their area of expertise, highlighting the modularity of AIgean . We demonstrate the effectiveness of AIgean with two use cases: an autoencoder, and ResNet-50 running across 10 and 12 FPGAs. AIgean leverages the FPGA’s strength in low-latency computing, as our implementations target batch-1 implementations.

Customer Review Provision Policies with Heterogeneous Cluster Preferences

Companies often post user-generated reviews online so that potential buyers in different clusters (age, geographic region, occupation, etc.) can learn from existing customers about the quality of an experience good and cluster preferences before purchasing. In this paper, we evaluate two common user-generated review provision policies for selling experience goods to customers in different clusters with heterogeneous preferences. The first policy is called the association-based policy (AP) under which a customer in a cluster can only observe the aggregate review (i.e., average rating) generated by users within the same cluster. The second policy is called the global-based policy (GP) under which each customer is presented with the aggregate review generated by all users across clusters. We find that, in general, the firm benefits from a policy that provides a larger number of “relevant reviews” to customers. When customers are more certain about the product quality and when clusters are more diverse, AP is more profitable than GP because it provides cluster-specific reviews to customers. Otherwise, GP is more profitable as it provides a larger number of less relevant reviews. Moreover, we propose a third provision policy that imparts the union of the information by AP and GP and show that it is more profitable for the firm. Although the third policy always renders a higher consumer welfare than GP, it may generate a lower consumer welfare than AP. This paper was accepted by Martínez-de-Albéniz Victor, operations management.

Pathogenesis of Children’s Allergic Diseases: Refocusing the Role of the Gut Microbiota

Allergic diseases comprise a genetically heterogeneous cluster of immunologically mediated diseases, including asthma, food allergy (FA), allergic rhinitis (AR) and eczema, that have become major worldwide health problems. Over the past few decades, the spread of allergic diseases has displayed an increasing trend, and it has been reported that 22% of 1.39 billion people in 30 countries have a type of allergic disease. Undoubtedly, allergic diseases, which can be chronic, with significant morbidity, mortality and dynamic progression, impose major economic burdens on society and families; thus, exploring the cause of allergic diseases and reducing their prevalence is a top priority. Recently, it has been reported that the gastrointestinal (GI) microbiota can provide vital signals for the development, function, and regulation of the immune system, and the above-mentioned contributions make the GI microbiota a key player in allergic diseases. Notably, the GI microbiota is highly influenced by the mode of delivery, infant diet, environment, antibiotic use and so on. Specifically, changes in the environment can result in the dysbiosis of the GI microbiota. The proper function of the GI microbiota depends on a stable cellular composition which in the case of the human microbiota consists mainly of bacteria. Large shifts in the ratio between these phyla or the expansion of new bacterial groups lead to a disease-promoting imbalance, which is often referred to as dysbiosis. And the dysbiosis can lead to alterations of the composition of the microbiota and subsequent changes in metabolism. Further, the GI microbiota can affect the physiological characteristics of the human host and modulate the immune response of the host. The objectives of this review are to evaluate the development of the GI microbiota, the main drivers of the colonization of the GI tract, and the potential role of the GI microbiota in allergic diseases and provide a theoretical basis as well as molecular strategies for clinical practice.

Dynamic Task Scheduling Scheme for Processing Real-Time Stream Data in Storm Environments

Owing to the recent advancements in Internet of Things technology, social media, and mobile devices, real-time stream balancing processing systems are commonly used to process vast amounts of data generated in various media. In this paper, we propose a dynamic task scheduling scheme considering task deadlines and node resources. The proposed scheme performs dynamic scheduling using a heterogeneous cluster consisting of various nodes with different performances. Additionally, the loads of the nodes considering the task deadlines are balanced by different task scheduling based on three defined load types. Based on diverse performance evaluations it is shown that the proposed scheme outperforms the conventional schemes.

Human immune globulin 10% with recombinant human Hyaluronidase- A Review

Primary immunodeficiency disorder (PID) refers to a heterogeneous cluster of over 350 syndromes that upshot from defects in the immune system development or function. PIDs are broadly classified as disorders of adaptive immunity or innate immunity. The enhanced efficacy of human immune serum globulin 10% with recombinant human Hyaluronidase with comparison to blood vessel human gamma globulin is a very prospective open-label study for PID. Treatment of primary immunological disorder diseases (PIDD) with Subcutaneous(SC) infusions of immune gamma globulin headed by an injection of hyazyme to extend SC tissue porousness was evaluated in two consecutive, prospective, non-controlled, multi-center studies. HYQVIA could be a subcutaneously mediated medication to treat the primary immunological disorder in adults. ENHANZE® drug delivery technology relies on the proprietary rHuPH20 macromolecule that facilitates the SC delivery of co administered medical specialty. Recombinant Human Hyaluronidase works by degrading the glycosaminoglycan hyaluronan, which plays a role in resistance to excessive flow of fluid within the Subcutaneous matrix, limiting massive volume SC drug delivery, dispersion, and absorption. Co-administration of recombinant Hyazyme with partner therapies can overcome administration time and volume barriers associated with existing SC therapeutic formulations.

PREVALENCE, PATHOGENESIS, CLASSIFICATION, IMAGING, SCREENING AND MANAGE-MENT OF DIABETIC RETINOPATHY- A CRITICAL REVIEW

Diabetic retinopathy is that the leading reason for sightlessness among people between twenty-five and seventyfour years older within the industrialised world. Diabetes mellitus (DM) includes a heterogeneous cluster of disorders of carbohydrate, protein, and metastasis manifesting hyperglycemia. Diabetic retinopathy could be microangiopathy ensuing from the chronic effects of the disease, and shares similarities with the microvascular alterations that occur in different tissues at risk of DM equivalent to the kidneys and also the peripheral nerves. Diabetic retinopathy is assessed into nonproliferative and proliferative stages. Nonproliferative diabetic retinopathy (NPDR) involves progressive intraretinal microvascular alterations that may result in, and a lot of advanced proliferative stages outlined by extraretinal neovascularization. Imaging modalities in common clinical use for the management of NPDR and DME embrace structure photography, fluorescein angiography (FA), and optical coherence tomography (OCT). The suggested schedule for screening and surveillance for NPDR reflects data concerning the epidemiology and natural history of the disease. Diabetic retinopathy could be a leading explanationfor vision loss in working-age Americans and a major cause of sightlessness worldwide. The International Diabetes Federation estimates that as several as 592 million individuals worldwide can have DM in 2035, a rise from or so 387 million people calculable to possess the disease in 2014. Here, we tend to present a review of the presentunderstanding and new insights into biochemical mechanisms within the pathological process in DR, classification, furthermore as clinical treatments for DR patients. Keywords: Diabetic retinopathy, diabetes mellitus, retinal degeneration, fluoresces in angiography, optical coher- ence tomography, VEGF, focal/grid laser photocoagulation.

A Resource-aware Container Management Scheme in a Heterogeneous Cluster

Cloud computing nowadays is not an emerging topic, and virtualization is an indispensable technology to expedite cloud computing to become the next sign of the coming Internet revolution. In real life, scientists never stop at exploring the possibilities from such technology by investigating millions of experiments and applications to enhance the quality of virtual services. However, isolated construction for the virtual machine doesn’t save the technology from unwanted data volumes or insensitive processing time. Containers are created to address such problems, by distributing applications without initiating the entire virtual machine. Docker, as an important player in this game, is an open-source application of the container family. The management tool from Docker containers, Swamskit, does not take heterogeneities in either virtualized containers or physical nodes. There are different nodes in the cluster, and each node is different in configurations, resource availability, or concerning resource, etc. Furthermore, the requirements initiated by different services change all the time. The demand might be CPU-intensive (e.g. Clustering services) and also memory-intensive (e.g. Web services), or completely at the opposite. In this paper, we focus on exploring the Docker container cluster and designing, DRAPS, a resource-aware placement scheme, to improve the system performance in a heterogeneous cluster.

Building alternative consensus trees and supertrees using k-means and Robinson and Foulds distance

Each gene has its own evolutionary history which can substantially differ from the evolutionary histories of other genes. For example, some individual genes or operons can be affected by specific horizontal gene transfer and recombination events. Thus, the evolutionary history of each gene should be represented by its own phylogenetic tree which may display different evolutionary patterns from the species tree that accounts for the main patterns of vertical descent. The output of traditional consensus tree or supertree inference methods is a unique consensus tree or supertree. Here, we describe a new efficient method for inferring multiple alternative consensus trees and supertrees to best represent the most important evolutionary patterns of a given set of phylogenetic trees (i.e. additive trees or X-trees). We show how a specific version of the popular k-means clustering algorithm, based on some interesting properties of the Robinson and Foulds topological distance, can be used to partition a given set of trees into one (when the data are homogeneous) or multiple (when the data are heterogeneous) cluster(s) of trees. We adapt the popular Caliński-Harabasz, Silhouette, Ball and Hall, and Gap cluster validity indices to tree clustering with k-means. A special attention is paid to the relevant but very challenging problem of inferring alternative supertrees, built from phylogenies constructed for different, but mutually overlapping, sets of taxa. The use of the Euclidean approximation in the objective function of the method makes it faster than the existing tree clustering techniques, and thus perfectly suitable for the analysis of large genomic datasets. In this study, we apply it to discover alternative supertrees characterizing the main patterns of evolution of SARS-CoV-2 and the related betacoronaviruses.

Collaborative Filtering Recommendation Using Nonnegative Matrix Factorization in GPU-Accelerated Spark Platform

Nonnegative matrix factorization (NMF) has been introduced as an efficient way to reduce the complexity of data compression and its capability of extracting highly interpretable parts from data sets, and it has also been applied to various fields, such as recommendations, image analysis, and text clustering. However, as the size of the matrix increases, the processing speed of nonnegative matrix factorization is very slow. To solve this problem, this paper proposes a parallel algorithm based on GPU for NMF in Spark platform, which makes full use of the advantages of in-memory computation mode and GPU acceleration. The new GPU-accelerated NMF on Spark platform is evaluated in a 4-node Spark heterogeneous cluster using Google Compute Engine by configuring each node a NVIDIA K80 CUDA device, and experimental results indicate that it is competitive in terms of computational time against the existing solutions on a variety of matrix orders. Furthermore, a GPU-accelerated NMF-based parallel collaborative filtering (CF) algorithm is also proposed, utilizing the advantages of data dimensionality reduction and feature extraction of NMF, as well as the multicore parallel computing mode of CUDA. Using real MovieLens data sets, experimental results have shown that the parallelization of NMF-based collaborative filtering on Spark platform effectively outperforms traditional user-based and item-based CF with a higher processing speed and higher recommendation accuracy.