Iterative Machine Learning Research Articles

Iterative machine learning based rotational alignment of brain 3D CT data.

The optimal rotational alignment of brain Computed Tomography (CT) images to a required standard position has a crucial importance for both automatic and manual diagnostic analysis. In this contribution, we present a novel two-step iterative approach for the automatic 3D rotational alignment of brain CT data. The angles of axial and coronal rotations are determined by an unsupervised by localisation of the Midsagittal Plane (MSP) method. This includes detection and pairing of medially symmetrical feature points. The sagittal rotation angle is subsequently estimated by regression convolutional neural network (CNN). The proposed methodology has been evaluated on a dataset of CT data manually aligned by radiologists. It has been shown that the algorithm achieved the low error of estimated rotations (≈1 degree) and in a significantly shorter time than the experts (≈2 minutes per case).

Detecting adverse drug reactions in discharge summaries of electronic medical records using Readpeer

BackgroundHospital discharge summaries offer a potentially rich resource to enhance pharmacovigilance efforts to evaluate drug safety in real-world clinical practice. However, it is infeasible for experts to read through all discharge summaries to find cases of drug-adverse event (AE) relations. PurposeThe objective of this paper is to develop a natural language processing (NLP) framework to detect drug-AE relations from unstructured hospital discharge summaries. Basic proceduresAn NLP algorithm was designed using customized dictionaries of drugs, adverse event (AE) terms, and rules based on trigger phrases, negations, fuzzy logic and word distances to recognize drug, AE terms and to detect drug-AE relations. Furthermore, a customized annotation tool was developed to facilitate expert review of discharge summaries from a tertiary hospital in Singapore in 2011. Main findingsA total of 33 trial sets with 50 to 100 records per set were evaluated (1620 discharge summaries) by our algorithm and reviewed by pharmacovigilance experts. After every 6 trial sets, drug and AE dictionaries were updated, and rules were modified to improve the system. Excellent performance was achieved for drug and AE entity recognition with over 92% precision and recall. On the final 6 sets of discharge summaries (600 records), our algorithm achieved 75% precision and 59% recall for identification of valid drug-AE relations. Principal conclusionsAdverse drug reactions are a significant contributor to health care costs and utilization. Our algorithm is not restricted to particular drugs, drug classes or specific medical specialties, which is an important attribute for a national regulatory authority to carry out comprehensive safety monitoring of drug products. Drug and AE dictionaries may be updated periodically to ensure that the tool remains relevant for performing surveillance activities. The development of the algorithm, and the ease of reviewing and correcting the results of the algorithm as part of an iterative machine learning process, is an important step towards use of hospital discharge summaries for an active pharmacovigilance program.

Iterative Machine Learning for Output Tracking

This article develops iterative machine learning (IML) for output tracking. The input-output data generated during iterations to develop the model used in the iterative update. The main contribution of this article to propose the use of kernel-based machine learning to iteratively update both the model and the model-inversion-based input simultaneously. Additionally, augmented inputs with persistency of excitation are proposed to promote learning of the model during the iteration process. The proposed approach is illustrated with a simulation example.

Interactive labelling of a multivariate dataset for supervised machine learning using linked visualisations, clustering, and active learning

Supervised machine learning techniques require labelled multivariate training datasets. Many approaches address the issue of unlabelled datasets by tightly coupling machine learning algorithms with interactive visualisations. Using appropriate techniques, analysts can play an active role in a highly interactive and iterative machine learning process to label the dataset and create meaningful partitions. While this principle has been implemented either for unsupervised, semi-supervised, or supervised machine learning tasks, the combination of all three methodologies remains challenging.In this paper, a visual analytics approach is presented, combining a variety of machine learning capabilities with four linked visualisation views, all integrated within the mVis (multivariate Visualiser) system. The available palette of techniques allows an analyst to perform exploratory data analysis on a multivariate dataset and divide it into meaningful labelled partitions, from which a classifier can be built. In the workflow, the analyst can label interesting patterns or outliers in a semi-supervised process supported by active learning. Once a dataset has been interactively labelled, the analyst can continue the workflow with supervised machine learning to assess to what degree the subsequent classifier has effectively learned the concepts expressed in the labelled training dataset. Using a novel technique called automatic dimension selection, interactions the analyst had with dimensions of the multivariate dataset are used to steer the machine learning algorithms.A real-world football dataset is used to show the utility of mVis for a series of analysis and labelling tasks, from initial labelling through iterations of data exploration, clustering, classification, and active learning to refine the named partitions, to finally producing a high-quality labelled training dataset suitable for training a classifier. The tool empowers the analyst with interactive visualisations including scatterplots, parallel coordinates, similarity maps for records, and a new similarity map for partitions.

An Agent-Based Hierarchical Bargaining Framework for Power Management of Multiple Cooperative Microgrids

In this paper, we propose an agent-based hierarchical power management model in a power distribution system composed of several microgrids (MGs). At the lower level of the model, multiple MGs bargain with each other to cooperatively obtain a fair, and Pareto-optimal solution to their power management problem, employing the concept of Nash bargaining solution and using a distributed optimization framework. At the highest level of the model, a distribution system power supplier, e.g., a utility company, interacts with both the cluster of the MGs and the wholesale market. The goal of the utility company is to facilitate power exchange between the regional distribution network consisting of multiple MGs and the wholesale market to achieve its own private goals. The power exchange is controlled through dynamic energy pricing at the distribution level, at the day-ahead and real-time stages. To implement energy pricing at the utility company level, an iterative machine learning mechanism is employed, where the utility company develops a price-sensitivity model of the aggregate response of the MGs to the retail price signal through a learning process. This learned model is then used to perform optimal energy pricing. To verify its applicability, the proposed decision model is tested on a system with multiple MGs, with each MG having different load/generation data.

PISCES: Optimizing Multi-Job Application Execution in MapReduce

Nowadays, many MapReduce applications consist of groups of jobs with dependencies among each other, such as iterative machine learning applications and large database queries. Unfortunately, the MapReduce framework is not optimized for these multi-job applications. It does not explore the execution overlapping opportunities among jobs and can only schedule jobs independently. These issues significantly inflate the application execution time. This paper presents Pipeline Improvement Support with Critical chain Estimation Scheduling (PISCES), a critical chain optimization (a critical chain refers to a series of jobs which will make the application run longer if any one of them is delayed), to provide better support for multi-job applications. PISCES extends the existing MapReduce framework to allow scheduling for multiple jobs with dependencies by dynamically building up a job dependency DAG for current running jobs according to their input and output directories. Then using the dependency DAG, it provides an innovative mechanism to facilitate the data pipelining between the output phase (map phase in the Map-Only job or reduce phase in the Map-Reduce job) of an upstream job and the map phase of a downstream job. This offers a new execution overlapping between dependent jobs in MapReduce which effectively reduces the application runtime. Moreover, PISCES proposes a novel critical chain job scheduling model based on the accurate critical chain estimation. Experiments show that PISCES can increase the degree of system parallelism by up to 68 percent and improve the execution speed of applications by up to 52 percent.

SlimML: Removing Non-critical Input Data in Large-scale Iterative Machine Learning

The core of many large-scale machine learning (ML) applications, such as neural networks (NN), support vector machine (SVM), and convolutional neural network (CNN), is the training algorithm that iteratively updates model parameters by processing massive datasets. From a plethora of studies aiming at accelerating ML, being data parallelization and parameter server, the prevalent assumption is that all data points are equivalently relevant to model parameter updating. In this article, we challenge this assumption by proposing a criterion to measure a data point's effect on model parameter updating, and experimentally demonstrate that the majority of data points are non-critical in the training process. We develop a slim learning framework, termed SlimML, which trains the ML models only on the critical data and thus significantly improves training performance. To such an end, SlimML efficiently leverages a small number of aggregated data points per iteration to approximate the criticalness of original input data instances. The proposed approach can be used by changing a few lines of code in a standard stochastic gradient descent (SGD) procedure, and we demonstrate experimentally, on NN regression, SVM classification, and CNN training, that for large datasets, it accelerates model training process by an average of 3.61 times while only incurring accuracy losses of 0.37 percent.

Discovering de novo peptide substrates for enzymes using machine learning

The discovery of peptide substrates for enzymes with exclusive, selective activities is a central goal in chemical biology. In this paper, we develop a hybrid computational and biochemical method to rapidly optimize peptides for specific, orthogonal biochemical functions. The method is an iterative machine learning process by which experimental data is deposited into a mathematical algorithm that selects potential peptide substrates to be tested experimentally. Once tested, the algorithm uses the experimental data to refine future selections. This process is repeated until a suitable set of de novo peptide substrates are discovered. We employed this technology to discover orthogonal peptide substrates for 4’-phosphopantetheinyl transferase, an enzyme class that covalently modifies proteins. In this manner, we have demonstrated that machine learning can be leveraged to guide peptide optimization for specific biochemical functions not immediately accessible by biological screening techniques, such as phage display and random mutagenesis.

BiGNoC: Accelerating Big Data Computing with Application-Specific Photonic Network-on-Chip Architectures

In the era of big data, high performance data analytics applications are frequently executed on large-scale cluster architectures to accomplish massive data-parallel computations. Often, these applications involve iterative machine learning algorithms to extract information and make predictions from large data sets. Multicast data dissemination is one of the major performance bottlenecks for such data analytics applications in cluster computing, as terabytes of data need to be distributed frequently from a single data source to hundreds of computing nodes. To overcome this bottleneck for big data applications, we propose BiGNoC , a manycore chip platform with a novel application-specific photonic network-on-chip (PNoC) fabric. BiGNoC is designed for big data computing and exploits multicasting in photonic waveguides. For high performance data analytics applications, BiGNoC improves throughput by up to ${{9.9}}\times$ while reducing latency by up to 88 percent and energy-per-bit by up to 98 percent over two state-of-the-art PNoC architectures as well as a broadcast-optimized electrical mesh NoC architecture, and a traditional electrical mesh NoC architecture.

Experimenting sensitivity-based anonymization framework in apache spark

One of the biggest concerns of big data and analytics is privacy. We believe the forthcoming frameworks and theories will establish several solutions for the privacy protection. One of the known solutions is the k-anonymity that was introduced for traditional data. Recently, two major frameworks leveraged big data processing and applications; these are MapReduce and Spark. Spark data processing has been attracting more attention due to its crucial impacts on a wide range of big data applications. One of the predominant big data applications is data analytics and anonymization. We previously proposed an anonymization method for implementing k-anonymity in MapReduce processing framework. In this paper, we investigate Spark performance in processing data anonymization. Spark is a fast processing framework that was implemented in several applications such as: SQL, multimedia, and data stream. Our focus is the SQL Spark, which is adequate for big data anonymization. Since Spark operates in-memory, we need to observe its limitations, speed, and fault tolerance on data size increase, and to compare MapReduce to Spark in processing anonymity. Spark introduces an abstraction called resilient distributed datasets, which reads and serializes a collection of objects partitioned across a set of machines. Developers claim that Spark can outperform MapReduce by 10 times in iterative machine learning jobs. Our experiments in this paper compare between MapReduce and Spark. The overall results show a better performance for Spark’s processing time in anonymity operations. However, in some limited cases, we prefer to implement the old MapReduce framework, when the cluster resources are limited and the network is non-congested.

A Two-Stage Big Data Analytics Framework with Real World Applications Using Spark Machine Learning and Long Short-Term Memory Network

Every day we experience unprecedented data growth from numerous sources, which contribute to big data in terms of volume, velocity, and variability. These datasets again impose great challenges to analytics framework and computational resources, making the overall analysis difficult for extracting meaningful information in a timely manner. Thus, to harness these kinds of challenges, developing an efficient big data analytics framework is an important research topic. Consequently, to address these challenges by exploiting non-linear relationships from very large and high-dimensional datasets, machine learning (ML) and deep learning (DL) algorithms are being used in analytics frameworks. Apache Spark has been in use as the fastest big data processing arsenal, which helps to solve iterative ML tasks, using distributed ML library called Spark MLlib. Considering real-world research problems, DL architectures such as Long Short-Term Memory (LSTM) is an effective approach to overcoming practical issues such as reduced accuracy, long-term sequence dependency, and vanishing and exploding gradient in conventional deep architectures. In this paper, we propose an efficient analytics framework, which is technically a progressive machine learning technique merged with Spark-based linear models, Multilayer Perceptron (MLP) and LSTM, using a two-stage cascade structure in order to enhance the predictive accuracy. Our proposed architecture enables us to organize big data analytics in a scalable and efficient way. To show the effectiveness of our framework, we applied the cascading structure to two different real-life datasets to solve a multiclass and a binary classification problem, respectively. Experimental results show that our analytical framework outperforms state-of-the-art approaches with a high-level of classification accuracy.

Application of Iterative Machine Learning for Output Tracking With Magnetic Soft Actuators

This paper presents experimental results of output tracking for soft magnetic actuators using iterative machine learning (IML) to estimate the model and its uncertainty. A challenge is that parameters of dynamic-response models of such soft actuators are sensitive to the operating point. To overcome this challenge and improve the tracking performance, a complex-valued Gaussian process regression (CGPR) is used to estimate a system inverse model; additionally, the CGPR approach provides an uncertainty measure, which is used to design the frequency-dependent iteration gain to ensure convergence. This approach leads to convergence even when data near resonance (where models tend to be sensitive) is not available. Moreover, a method to improve the model estimation is proposed, in which the initial training signal is augmented with additional frequency content to increase the input and output signal size compared to the noise. Using this augmented approach, the IML approach results in a ${\text{95}\%}$ reduction in the maximum tracking error compared to using the dc gain alone. The final maximum tracking error is ${\text{1.8}\%}$ , which is near the sensor resolution limit.

Iteration acceleration for distributed learning systems

Abstract During the implementation of iterative machine learning algorithms for objective function optimization in large-scale distributed environment, they can be blocked when some of the machines failed. In this paper, a hybrid approach is proposed to balance the performance and efficiency. In each iteration, the results from failure machines are abandoned. We will discuss the relationship between accuracy and abandon rate which can be formulated as inequations. It is argued that the speed of this process is highly effective and efficient. The algorithm is demonstrated using real distributed environment and shows significant improvement of speedup results.

SparkSeq: fast, scalable and cloud-ready tool for the interactive genomic data analysis with nucleotide precision.

Many time-consuming analyses of next -: generation sequencing data can be addressed with modern cloud computing. The Apache Hadoop-based solutions have become popular in genomics BECAUSE OF: their scalability in a cloud infrastructure. So far, most of these tools have been used for batch data processing rather than interactive data querying. The SparkSeq software has been created to take advantage of a new MapReduce framework, Apache Spark, for next-generation sequencing data. SparkSeq is a general-purpose, flexible and easily extendable library for genomic cloud computing. It can be used to build genomic analysis pipelines in Scala and run them in an interactive way. SparkSeq opens up the possibility of customized ad hoc secondary analyses and iterative machine learning algorithms. This article demonstrates its scalability and overall fast performance by running the analyses of sequencing datasets. Tests of SparkSeq also prove that the use of cache and HDFS block size can be tuned for the optimal performance on multiple worker nodes. Available under open source Apache 2.0 license: https://bitbucket.org/mwiewiorka/sparkseq/.

Identifying PCB Contaminated Transformers Through Active Learning

Exposure to polychlorinated biphenyals (PCBs) is hazardous to human health. The United Nations Environment Programme has decreed that nations, including Canada and USA, must eliminate PCB contaminated equipment such as transformers by 2025. To determine the PCB status of a transformer with absolute certainty, the oil mixture of the transformer must be sampled because transformers labeled as non-PCB could be cross-contaminated. Since sealed oil mixture sampling is costly, for the first time, we apply an iterative machine learning technique called active learning to classify PCB contaminated transformers while minimizing a cost metric that integrates the classification error cost and the sampling cost. We propose a dynamic sampling method to address two key issues in active learning: the sampling size per iteration and the stopping criterion of the sampling process. The proposed algorithm is evaluated using the real-world datasets from BC Hydro in Canada.

Complete Blood Cell Count in Psittaciformes by Using High-Throughput Image Cytometry: A Pilot Study

The avian hemogram is usually performed in veterinary diagnostic laboratories by using manual cell counting techniques and differential counts determined by light microscopy. There is no standard automated technique for avian blood cell count and differentiation to date. These shortcomings in birds are primarily because erythrocytes and thrombocytes are nucleated, which precludes the use of automated analyzers programmed to perform mammal complete blood cell counts. In addition, there is no standard avian antibody panel, which would allow cell differentiation by immunophenotyping across all commonly seen bird species. We report an alternative hematologic approach for quantification and differentiation of avian blood cells by using high-throughput image cytometry on blood smears in psittacine bird species. A pilot study was designed with 70 blood smears of different psittacine bird species stained with a Wright-Giemsa stain. The slides were scanned at 0.23 microm/pixel. The open-source softwares CellProfiler and CellProfiler Analyst were used for analyzing and sorting each cell by image cytometry. A "pipeline" was constructed in the CellProfiler by using different modules to identify and export hundreds of measures per cell for shape, intensity, and texture. Rules for classifying the different blood cell phenotypes were then determined based on these measurements by iterative feedback and machine learning by using CellProfiler Analyst. Although this approach shows promises, avian Leukopet results could not be duplicated when using this technique as is. Further studies and more standardized prospective investigations may be needed to refine the "pipeline" strategy and the machine learning algorithm.

REEF

In this demo proposal, we describe REEF, a framework that makes it easy to implement scalable, fault-tolerant runtime environments for a range of computational models. We will demonstrate diverse workloads, including extract-transform-load MapReduce jobs, iterative machine learning algorithms, and ad-hoc declarative query processing. At its core, REEF builds atop YARN (Apache Hadoop 2's resource manager) to provide retainable hardware resources with lifetimes that are decoupled from those of computational tasks. This allows us to build persistent (cross-job) caches and cluster-wide services, but, more importantly, supports high-performance iterative graph processing and machine learning algorithms. Unlike existing systems, REEF aims for composability of jobs across computational models, providing significant performance and usability gains, even with legacy code. REEF includes a library of interoperable data management primitives optimized for communication and data movement (which are distinct from storage locality). The library also allows REEF applications to access external services, such as user-facing relational databases. We were careful to decouple lower levels of REEF from the data models and semantics of systems built atop it. The result was two new standalone systems: Tang, a configuration manager and dependency injector, and Wake, a state-of-the-art event-driven programming and data movement framework. Both are language independent, allowing REEF to bridge the JVM and .NET.

Scoring diverse cellular morphologies in image-based screens with iterative feedback and machine learning

Many biological pathways were first uncovered by identifying mutants with visible phenotypes and by scoring every sample in a screen via tedious and subjective visual inspection. Now, automated image analysis can effectively score many phenotypes. In practical application, customizing an image-analysis algorithm or finding a sufficient number of example cells to train a machine learning algorithm can be infeasible, particularly when positive control samples are not available and the phenotype of interest is rare. Here we present a supervised machine learning approach that uses iterative feedback to readily score multiple subtle and complex morphological phenotypes in high-throughput, image-based screens. First, automated cytological profiling extracts hundreds of numerical descriptors for every cell in every image. Next, the researcher generates a rule (i.e., classifier) to recognize cells with a phenotype of interest during a short, interactive training session using iterative feedback. Finally, all of the cells in the experiment are automatically classified and each sample is scored based on the presence of cells displaying the phenotype. By using this approach, we successfully scored images in RNA interference screens in 2 organisms for the prevalence of 15 diverse cellular morphologies, some of which were previously intractable.

Complex stochastic systems modelling and control via iterative machine learning

Complex stochastic systems require the control of their stochastic distributions (i.e., the shape of their output probability density functions (PDFs)). This paper will address both modelling and control of such systems and will consist of a brief survey of the recent developments and the description of a detailed design procedure on an iterative learning-based output PDF control algorithm. In this context, a radial basis function (RBF) neural network is used to approximate the output PDFs, and the system dynamics is represented by a mathematical model between the weights of the RBF neural networks and the control input. The whole control horizon is divided into a number of batches, where within each batch fixed RBFs are used. However, between batches, iterative machine learning for the shape update of the RBFs is developed so as to improve the closed loop performance batch-by-batch. A simulated example is given so as to demonstrate the design procedure.

SVM-HUSTLE—an iterative semi-supervised machine learning approach for pairwise protein remote homology detection

As the amount of biological sequence data continues to grow exponentially we face the increasing challenge of assigning function to this enormous molecular 'parts list'. The most popular approaches to this challenge make use of the simplifying assumption that similar functional molecules, or proteins, sometimes have similar composition, or sequence. However, these algorithms often fail to identify remote homologs (proteins with similar function but dissimilar sequence) which often are a significant fraction of the total homolog collection for a given sequence. We introduce a Support Vector Machine (SVM)-based tool to detect homology using semi-supervised iterative learning (SVM-HUSTLE) that identifies significantly more remote homologs than current state-of-the-art sequence or cluster-based methods. As opposed to building profiles or position specific scoring matrices, SVM-HUSTLE builds an SVM classifier for a query sequence by training on a collection of representative high-confidence training sets, recruits additional sequences and assigns a statistical measure of homology between a pair of sequences. SVM-HUSTLE combines principles of semi-supervised learning theory with statistical sampling to create many concurrent classifiers to iteratively detect and refine, on-the-fly, patterns indicating homology. When compared against existing methods for identifying protein homologs (BLAST, PSI-BLAST, COMPASS, PROF_SIM, RANKPROP and their variants) on two different benchmark datasets SVM-HUSTLE significantly outperforms each of the above methods using the most stringent ROC(1) statistic with P-values less than 1e-20. SVM-HUSTLE also yields results comparable to HHSearch but at a substantially reduced computational cost since we do not require the construction of HMMs. The software executable to run SVM-HUSTLE can be downloaded from http://www.sysbio.org/sysbio/networkbio/svm_hustle