Search NASA⌕ Search

SEARCH · Search NASA

Results for “Randomized methods”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,783 records · Page 99

Margin Evaluation in the Presence of Deformation, Rotation, and Translation in Prostate and Entire Seminal Vesicle Irradiation With Daily Marker-Based Setup Corrections

Purpose: To develop a method for margin evaluation accounting for all measured displacements during treatment of prostate cancer. Methods and Materials: For 21 patients treated with stereographic targeting marker-based online translation corrections, dose distributions with varying margins and gradients were created. Sets of possible cumulative delivered dose distributions were simulated by moving voxels and accumulating dose per voxel. Voxel motion was simulated consistent with measured distributions of systematic and random displacements due to stereographic targeting inaccuracies, deformation, rotation, and intrafraction motion. The method of simulation maintained measured correlation of voxel motions due to organ deformation. Results: For the clinical target volume including prostate and seminal vesicles (SV), the probability that some part receives <95% of the prescribed dose, the changes in minimum dose, and volume receiving 95% of prescription dose compared with planning were 80.5% {+-} 19.2%, 9.0 {+-} 6.8 Gy, and 3.0% {+-} 3.7%, respectively, for the smallest studied margins (3 mm prostate, 5 mm SV) and steepest dose gradients. Corresponding values for largest margins (5 mm prostate, 8 mm SV) with a clinical intensity-modulated radiotherapy dose distribution were 46.5% {+-} 34.7%, 6.7 {+-} 5.8 Gy, and 1.6% {+-} 2.3%. For prostate-only clinical target volume, the values were 51.8% {+-} 17.7%, 3.3 {+-} 1.6 Gy, and 0.6% {+-} 0.5% with the smallest margins and 5.2% {+-} 7.4%, 1.8 {+-} 0.9 Gy, and 0.1% {+-} 0.1% for the largest margins. Addition of three-dimensional rotation corrections only improved these values slightly. All rectal planning constraints were met in the actual reconstructed doses for all studied margins. Conclusion: We developed a system for margin validation in the presence of deformations. In our population, a 5-mm margin provided sufficient dosimetric coverage for the prostate. In contrast, an 8-mm SV margin was still insufficient owing to deformations. Addition of three-dimensional rotation corrections was of minor influence.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Strong converse theorems using Rényi entropies

We use a Rényi entropy method to prove strong converse theorems for certain information-theoretic tasks which involve local operations and quantum (or classical) communication between two parties. These include state redistribution, coherent state merging, quantum state splitting, measurement compression with quantum side information, randomness extraction against quantum side information, and data compression with quantum side information. The method we employ in proving these results extends ideas developed by Sharma [preprint http://arxiv.org/abs/1404.5940 [quant-ph] (2014)], which he used to give a new proof of the strong converse theorem for state merging. For state redistribution, we prove the strong converse property for the boundary of the entire achievable rate region in the (e, q)-plane, where e and q denote the entanglement cost and quantum communication cost, respectively. In the case of measurement compression with quantum side information, we prove a strong converse theorem for the classical communication cost, which is a new result extending the previously known weak converse. For the remaining tasks, we provide new proofs for strong converse theorems previously established using smooth entropies. For each task, we obtain the strong converse theorem from explicit bounds on the figure of merit of the task in terms of a Rényi generalization of the optimal rate. Hence, we identify candidates for the strong converse exponents for each task discussed in this paper. To prove our results, we establish various new entropic inequalities, which might be of independent interest. These involve conditional entropies and mutual information derived from the sandwiched Rényi divergence. In particular, we obtain novel bounds relating these quantities, as well as the Rényi conditional mutual information, to the fidelity of two quantum states.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Building and destroying symmetry in 1-D elastic systems

Locally periodic rods, which show approximate invariance with respect to translations, are constructed by joining N unit cells. The spectrum then shows a band spectrum. We then break the local periodicity by including one or more defects in the system. When the defects follow a certain definite prescription, an analog of the Wannier-Stark ladders is gotten; when the defects are random, an elastic rod showing Anderson localization is obtained. In all cases experimental values match the theoretical predictions.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Uncertainty and Intelligence in Computational Stochastic Mechanics

Classical structural reliability assessment techniques are based on precise and crisp (sharp) definitions of failure and non-failure (survival) of a structure in meeting a set of strength, function and serviceability criteria. These definitions are provided in the form of performance functions and limit state equations. Thus, the criteria provide a dichotomous definition of what real physical situations represent, in the form of abrupt change from structural survival to failure. However, based on observing the failure and survival of real structures according to the serviceability and strength criteria, the transition from a survival state to a failure state and from serviceability criteria to strength criteria are continuous and gradual rather than crisp and abrupt. That is, an entire spectrum of damage or failure levels (grades) is observed during the transition to total collapse. In the process, serviceability criteria are gradually violated with monotonically increasing level of violation, and progressively lead into the strength criteria violation. Classical structural reliability methods correctly and adequately include the ambiguity sources of uncertainty (physical randomness, statistical and modeling uncertainty) by varying amounts. However, they are unable to adequately incorporate the presence of a damage spectrum, and do not consider in their mathematical framework any sources of uncertainty of the vagueness type. Vagueness can be attributed to sources of fuzziness, unclearness, indistinctiveness, sharplessness and grayness; whereas ambiguity can be attributed to nonspecificity, one-to-many relations, variety, generality, diversity and divergence. Using the nomenclature of structural reliability, vagueness and ambiguity can be accounted for in the form of realistic delineation of structural damage based on subjective judgment of engineers. For situations that require decisions under uncertainty with cost/benefit objectives, the risk of failure should depend on the underlying level of damage and the uncertainties associated with its definition. A mathematical model for structural reliability assessment that includes both ambiguity and vagueness types of uncertainty was suggested to result in the likelihood of failure over a damage spectrum. The resulting structural reliability estimates properly represent the continuous transition from serviceability to strength limit states over the ultimate time exposure of the structure. In this section, a structural reliability assessment method based on a fuzzy definition of failure is suggested to meet these practical needs. A failure definition can be developed to indicate the relationship between failure level and structural response. In this fuzzy model, a subjective index is introduced to represent all levels of damage (or failure). This index can be interpreted as either a measure of failure level or a measure of a degree of belief in the occurrence of some performance condition (e.g., failure). The index allows expressing the transition state between complete survival and complete failure for some structural response based on subjective evaluation and judgment.

Ayyub, Bilal M.↗

A numerical method for determining highly precise electron energy distribution functions from Langmuir probe characteristics

Electron energy distribution functions (EEDFs) were determined from probe characteristics using a numerical ac superimposed method with a distortion correction of high derivative terms by varying amplitude of a sinusoidal perturbation voltage superimposed onto the dc sweep voltage, depending on the related electron energy. Low amplitude perturbation applied around the plasma potential represented the low energy peak of the EEDF exactly, and high amplitude perturbation applied around the floating potential was effective to suppress noise or distortion of the probe characteristic, which is fatal to the tail electron distribution. When a small random noise was imposed over the stabilized prove characteristic, the numerical differentiation method was not suitable to determine the EEDF, while the numerical ac superimposed method was able to obtain a highly precise EEDF.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Nonparametric, data-based kernel interpolation for particle-tracking simulations and kernel density estimation

Traditional interpolation techniques for particle tracking include binning and convolutional formulas that use pre-determined (i.e., closed-form, parameteric) kernels. In many instances, the particles are introduced as point sources in time and space, so the cloud of particles (either in space or time) is a discrete representation of the Green’s function of an underlying PDE. As such, each particle is a sample from the Green’s function; therefore, each particle should be distributed according to the Green’s function. In short, the kernel of a convolutional interpolation of the particle sample “cloud” should be a replica of the cloud itself. This idea gives rise to an iterative method by which the form of the kernel may be discerned in the process of interpolating the Green’s function. When the Green’s function is a density, this method is broadly applicable to interpolating a kernel density estimate based on random data drawn from a single distribution. We formulate and construct the algorithm and demonstrate its ability to perform kernel density estimation of skewed and/or heavy-tailed data including breakthrough curves.

42 ENGINEERING↗

Hydrogel Spacer Prospective Multicenter Randomized Controlled Pivotal Trial: Dosimetric and Clinical Effects of Perirectal Spacer Application in Men Undergoing Prostate Image Guided Intensity Modulated Radiation Therapy

Purpose: Perirectal spacing, whereby biomaterials are placed between the prostate and rectum, shows promise in reducing rectal dose during prostate cancer radiation therapy. A prospective multicenter randomized controlled pivotal trial was performed to assess outcomes following absorbable spacer (SpaceOAR system) implantation. Methods and Materials: Overall, 222 patients with clinical stage T1 or T2 prostate cancer underwent computed tomography (CT) and magnetic resonance imaging (MRI) scans for treatment planning, followed with fiducial marker placement, and were randomized to receive spacer injection or no injection (control). Patients received postprocedure CT and MRI planning scans and underwent image guided intensity modulated radiation therapy (79.2 Gy in 1.8-Gy fractions). Spacer safety and impact on rectal irradiation, toxicity, and quality of life were assessed throughout 15 months. Results: Spacer application was rated as “easy” or “very easy” 98.7% of the time, with a 99% hydrogel placement success rate. Perirectal spaces were 12.6 ± 3.9 mm and 1.6 ± 2.0 mm in the spacer and control groups, respectively. There were no device-related adverse events, rectal perforations, serious bleeding, or infections within either group. Pre-to postspacer plans had a significant reduction in mean rectal V70 (12.4% to 3.3%, P<.0001). Overall acute rectal adverse event rates were similar between groups, with fewer spacer patients experiencing rectal pain (P=.02). A significant reduction in late (3-15 months) rectal toxicity severity in the spacer group was observed (P=.04), with a 2.0% and 7.0% late rectal toxicity incidence in the spacer and control groups, respectively. There was no late rectal toxicity greater than grade 1 in the spacer group. At 15 months 11.6% and 21.4% of spacer and control patients, respectively, experienced 10-point declines in bowel quality of life. MRI scans at 12 months verified spacer absorption. Conclusions: Spacer application was well tolerated. Increased perirectal space reduced rectal irradiation, reduced rectal toxicity severity, and decreased rates of patients experiencing declines in bowel quality of life. The spacer appears to be an effective tool, potentially enabling advanced prostate RT protocols.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

A cross-sectional study of the temporal evolution of electricity consumption of six commercial buildings

Current approaches to building efficiency diagnoses include conventional energy audit techniques that can be expensive and time consuming. In contrast, virtual energy audits of readily available 15-minute-interval building electricity consumption are being explored to provide quick, inexpensive, and useful insights into building operation characteristics. A cross sectional analysis of six buildings in two different climate zones provides methods for data cleaning, population-based building comparisons, and relationships (correlations) of weather and electricity consumption. Data cleaning methods have been developed to categorize and appropriately filter or correct anomalous data including outliers, missing data, and erroneous values (resulting in < 0.5% anomalies). The utility of a cross-sectional analysis of a sample set of building's electricity consumption is found through comparisons of baseload, daily consumption variance, and energy use intensity. Correlations of weather and electricity consumption 15-minute interval datasets show important relationships for the heating and cooling seasons using computed correlations of a Time-Specific-Averaged- Ordered Variable (exterior temperature) and corresponding averaged variables (electricity consumption)(TSAOV method). The TSAOV method is unique as it introduces time of day as a third variable while also minimizing randomness in both correlated variables through averaging. This study found that many of the pair-wise linear correlation analyses lacked strong relationships, prompting the development of the new TSAOV method to uncover the causal relationship between electricity and weather. We conclude that a combination of varied HVAC system operations, building thermal mass, plug load use, and building set point temperatures are likely responsible for the poor correlations in the prior studies, while the correlation of time-specific-averaged-ordered temperature and corresponding averaged variables method developed herein adequately accounts for these issues and enables discovery of strong linear pair-wise correlation R values. TSAOV correlations lay the foundation for a new approach to building studies, that mitigates plug load interferences and identifies more accurate insights into weather-energy relationship for all building types. Over all six buildings analyzed the TSAOV method reported very significant average correlations per building of 0.94 to 0.82 in magnitude. Our rigorous statistics-based methods applied to 15- minute-interval electricity data further enables virtual energy audits of buildings to quickly and inexpensively inform energy savings measures.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Controlling radioisotope proportions when randomly sampling from Dirichlet distributions in PyRIID

As machine learning models for radioisotope quantification become more powerful, likewise the need for high-quality synthetic training data grows as well. For problem spaces that involve estimating the relative isotopic proportions of various sources in gamma spectra it is necessary to generate training data that accurately represents the variance of proportions encountered. In this report, we aim to provide guidance on how to target a desired variance of proportions which are randomly when using the PyRIID Seed Mixer, which samples from a Dirichlet distribution. We provide a method for properly parameterizing the Dirichlet distribution in order to maintain a constant variance across an arbitrary number of dimensions, where each dimension represents a distinct source template being mixed. We demonstrate that our method successfully parameterizes the Dirichlet distribution to target a specific variance of proportions, provided that several conditions are met. This allows us to follow a principled technique for controlling how random mixture proportions are generated which are then used downstream in the synthesis process to produce the final, noisy gamma spectra.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

SU-E-J-83: Ion Imaging to Better Estimate In-Vivo Relative Stopping Powers Using X-Ray CT Prior-Knowledge Information

Purpose: To reduce uncertainties in relative stopping power (RSP) estimates for ions (alpha and carbon) by using Ion radiographic-imaging and X-ray CT prior-knowledge. Methods: A 36×36 phantom matrix composed of 9 materials with different thicknesses and randomly placed is generated. Theoretical RSPs are calculated using stopping power (SP) data from three references (Janni, ICRU49 and Bischel). We introduced an artificial systematic error (1.5%, 2.5% or 3.5%) and a random error (<0.5%) to the SP to simulated patient ion-range errors present in clinic environment. Carbon/alpha final energy for each RSPs set (theoretical and from CT images) is obtained with a ray-tracing algorithm. A gradient descent (GD) method is used to minimize the difference in exit particle energy, between theory and X-ray CT RSP maps, by iteratively correcting the RSP map from X-ray CT. Once a new set of RSPs is obtained for a direction a new optimization is done over other direction using the RSPs from the previous optimization. Theoretical RSPs are compared with experimental RSPs obtained with Gammex Phantom. Results: Preliminary results show that optimized RSP values can be obtained with smaller uncertainties (<1%) than clinical RSPs (1.5% to 3.5%). Theoretical values from three different references show uncertainties, up to 3% from experimental values. Further investigation will consider prior-knowledge from RSP obtained with CT images and ion radiographies from Monte Carlo Simulations. Conclusion: GD and ray-tracing methods have been implemented to reduce RSP uncertainties from values obtained for clinical treatment. Experimental RSPs will be obtained using carbon/alpha beams to consider the existence of material dependent systematic errors. Based on the results, it is hoped to show that using ray-tracing optimization with ion radiography and prior knowledge on RPSs, treatment planning accuracy and cost-effectiveness can be improved.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

On the apparent insignificance of the randomness of flexible joints on large space truss dynamics

Deployable periodic large space structures have been shown to exhibit high dynamic sensitivity to period-breaking imperfections and uncertainties. These can be brought on by manufacturing or assembly errors, structural imperfections, as well as nonlinear and/or nonconservative joint behavior. In addition, the necessity of precise pointing and position capability can require the consideration of these usually negligible and unknown parametric uncertainties and their effect on the overall dynamic response of large space structures. This work describes the use of a new design approach for the global dynamic solution of beam-like periodic space structures possessing parametric uncertainties. Specifically, the effect of random flexible joints on the free vibrations of simply-supported periodic large space trusses is considered. The formulation is a hybrid approach in terms of an extended Timoshenko beam continuum model, Monte Carlo simulation scheme, and first-order perturbation methods. The mean and mean-square response statistics for a variety of free random vibration problems are derived for various input random joint stiffness probability distributions. The results of this effort show that, although joint flexibility has a substantial effect on the modal dynamic response of periodic large space trusses, the effect of any reasonable uncertainty or randomness associated with these joint flexibilities is insignificant.

Koch, R. M.↗

Device-scale CFD study for mass transfer coefficient and effective mass transfer area in packed column

Carbon dioxide (CO 2 ) is a significant contributor of greenhouse gases emissions responsible for global warming. The fossil-fueled power plant is one of the major sources of CO 2 emission worldwide and needs to be mitigated. Post-combustion carbon capture by chemical absorption can be a promising technology to reduce CO 2 emission from power plants. The absorption is carried out via countercurrent flow in the packed columns. Although monoethanolamine (MEA) is a popular solvent, it has also limitations such as high volatility of solvent, corrosivity that leads to product degradation and damage to apparatuses, low CO 2 loading capacity, etc. Therefore, alternate solvents are being explored as prospective materials for carbon capture. A water-lean solvent, CO 2 binding organic liquids (CO2BOL), is currently being developed as a prospective material in carbon capture at the Pacific Northwest National Laboratory. Since the development of the material is at the primitive stage, it needs extensive investigation before industrial deployments. Hydrodynamics and mass transfer phenomena in the packed column using CO 2 BOL is unknown. Computational fluid dynamics (CFD) can be a useful tool to understand these phenomena in the packed column. Before employing the extensive CFD investigations, one needs to carefully validate CFD predictions for accurate flow predictions. This report demonstrates the validation of multiphase flow simulations against the experimental results for both structured and random packed columns. The reactive multiphase flow simulations using volume of fluid method were conducted at a wide range of liquid loads and contacts angles. The effective area is one of the critical factors dictating interphase mass transfer due to chemical reactions. Accordingly, CFD-predicted effective areas were compared with the experimental results of Tsai et al. [1] for Mellapak 250.Y structured packing. Both results matched well at different liquid loads in the presence of fast chemical kinetics using caustic solvents. Further, reactive multiphase flow simulations were also conducted in pall ring random packing to compute effective areas in the presence of fast chemical reactions. The CFD-predicted effective area compared well with correlation developed by Song et al. [2]. With successful validations, the current simulation framework can be further used for prediction of hydrodynamics and mass transfer phenomena using CO 2 BOL solvents in the packed column. Subsequently, it would be helpful in design optimization of a packed column. In addition to validation of the CFD model, this report also explains the mechanism for enhanced liquid side mass transfer for a structured packed column as compared to the wetted wall column.

20 FOSSIL-FUELED POWER PLANTS↗

Long-Term Results of a Randomized Trial in Locally Advanced Rectal Cancer: No Benefit From Adding a Brachytherapy Boost

Purpose/Objective(s): Mature data on tumor control and survival are presented from a randomized trial of the addition of a brachytherapy boost to long-course neoadjuvant chemoradiation therapy (CRT) for locally advanced rectal cancer. Methods and Materials: Between March 2005 and November 2008, 248 patients with T3-4N0-2M0 rectal cancer were prospectively randomized to either long-course preoperative CRT (50.4 Gy in 28 fractions, per oral tegafur-uracil and L-leucovorin) alone or the same CRT schedule plus a brachytherapy boost (10 Gy in 2 fractions). The primary trial endpoint was pathologic complete response (pCR) at the time of surgery; secondary endpoints included overall survival (OS), progression-free survival (PFS), and freedom from locoregional failure. Results: Results for the primary endpoint have previously been reported. This analysis presents survival data for the 224 patients in the Danish part of the trial. In all, 221 patients (111 control arm, 110 brachytherapy boost arm) had data available for analysis, with a median follow-up time of 5.4 years. Despite a significant increase in tumor response at the time of surgery, no differences in 5-year OS (70.6% vs 63.6%, hazard ratio [HR] = 1.24, P=.34) and PFS (63.9% vs 52.0%, HR=1.22, P=.32) were observed. Freedom from locoregional failure at 5 years were 93.9% and 85.7% (HR=2.60, P=.06) in the standard and in the brachytherapy arms, respectively. There was no difference in the prevalence of stoma. Explorative analysis based on stratification for tumor regression grade and resection margin status indicated the presence of response migration. Conclusions: Despite increased pathologic tumor regression at the time of surgery, we observed no benefit on late outcome. Improved tumor regression does not necessarily lead to a relevant clinical benefit when the neoadjuvant treatment is followed by high-quality surgery.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Universality in dissipative Landau-Zener transitions

We introduce a random-variable approach to investigate the dynamics of a dissipative two-state system. Based on an exact functional integral description, our method reformulates the problem as that of the time evolution of a quantum state vector subject to a Hamiltonian containing random noise fields. This numerically exact, nonperturbative formalism is particularly well suited in the context of time-dependent Hamiltonians, at both zero and finite temperature. As an important example, we consider the renowned Landau-Zener problem in the presence of an Ohmic environment with a large cutoff frequency at finite temperature. We investigate the ''scaling'' limit of the problem at intermediate times, where the decay of the upper-spin-state population is universal. Such a dissipative situation may be implemented using a cold-atom bosonic setup.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Graph Sampling for Visual Analytics

Effectively visualizing large graphs and capturing the statistical properties are two challenging tasks. To aid in these two tasks, many sampling approaches for graph simplification have been proposed, falling into three categories: node sampling, edge sampling, and traversal-based sampling. It is still unknown which approach is the best. We evaluate commonly used graph sampling methods through a combined visual and statistical comparison of graphs sampled at various rates. We conduct our evaluation on three graph models: random graphs, small-world graphs, and scale-free graphs. Initial results indicate that the effectiveness of a sampling method is dependent on the graph model, the size of the graph, and the desired statistical property. This benchmark study can be used as a guideline in choosing the appropriate method for a particular graph sampling task, and the results presented can be incorporated into graph visualization and analysis tools.

97 MATHEMATICS AND COMPUTING↗

Amino Acid Encoding for Deep Learning Applications

Background: The number of applications of deep learning algorithms in bioinformatics is increasing as they usually achieve superior performance over classical approaches, especially, when bigger training datasets are available. In deep learning applications, discrete data, e.g. words or n-grams in language, or amino acids or nucleotides in bioinformatics, are generally represented as a continuous vector through an embedding matrix. Recently, learning this embedding matrix directly from the data as part of the continuous iteration of the model to optimize the target prediction – a process called ‘end-to-end learning’ – has led to state-of-the-art results in many fields. Although usage of embeddings is well described in the bioinformatics literature, the potential of end-to-end learning for single amino acids, as compared to more classical manually-curated encoding strategies, has not been systematically addressed. To this end, we compared classical encoding matrices, namely one-hot, VHSE8 and BLOSUM62, to end-to-end learning of amino acid embeddings for two different prediction tasks using three widely used architectures, namely recurrent neural networks (RNN), convolutional neural networks (CNN), and the hybrid CNN-RNN. Results: By using different deep learning architectures, we show that end-to-end learning is on par with classical encodings for embeddings of the same dimension even when limited training data is available, and might allow for a reduction in the embedding dimension without performance loss, which is critical when deploying the models to devices with limited computational capacities. We found that the embedding dimension is a major factor in controlling the model performance. Surprisingly, we observed that deep learning models are capable of learning from random vectors of appropriate dimension. Conclusion: Our study shows that end-to-end learning is a flexible and powerful method for amino acid encoding. Further, due to the flexibility of deep learning systems, amino acid encoding schemes should be benchmarked against random vectors of the same dimension to disentangle the information content provided by the encoding scheme from the distinguishability effect provided by the scheme.

Deep-learning↗

Amino Acid Encoding for Deep Learning Applications

Background: The number of applications of deep learning algorithms in bioinformatics is increasing as they usually achieve superior performance over classical approaches, especially, when bigger training datasets are available. In deep learning applications, discrete data, e.g. words or n-grams in language, or amino acids or nucleotides in bioinformatics, are generally represented as a continuous vector through an embedding matrix. Recently, learning this embedding matrix directly from the data as part of the continuous iteration of the model to optimize the target prediction – a process called ‘end-to-end learning’ – has led to state-ofthe-art results in many fields. Although usage of embeddings is well described in the bioinformatics literature, the potential of end-to-end learning for single amino acids, as compared to more classical manually-curated encoding strategies, has not been systematically addressed. To this end, we compared classical encoding matrices, namely one-hot, VHSE8 and BLOSUM62, to end-to-end learning of amino acid embeddings for two different prediction tasks using three widely used architectures, namely recurrent neural networks (RNN), convolutional neural networks (CNN), and the hybrid CNN-RNN. Results: By using different deep learning architectures, we show that end-to-end learning is on par with classical encodings for embeddings of the same dimension even when limited training data is available, and might allow for a reduction in the embedding dimension without performance loss, which is critical when deploying the models to devices with limited computational capacities. We found that the embedding dimension is a major factor in controlling the model performance. Surprisingly, we observed that deep learning models are capable of learning from random vectors of appropriate dimension. Conclusion: Our study shows that end-to-end learning is a flexible and powerful method for amino acid encoding. Further, due to the flexibility of deep learning systems, amino acid encoding schemes should be benchmarked against random vectors of the same dimension to disentangle the information content provided by the encoding scheme from the distinguishability effect provided by the scheme.

Hesham ElAbd↗

Discriminant forest classification method and system

A hybrid machine learning methodology and system for classification that combines classical random forest (RF) methodology with discriminant analysis (DA) techniques to provide enhanced classification capability. A DA technique which uses feature measurements of an object to predict its class membership, such as linear discriminant analysis (LDA) or Andersen-Bahadur linear discriminant technique (AB), is used to split the data at each node in each of its classification trees to train and grow the trees and the forest. When training is finished, a set of n DA-based decision trees of a discriminant forest is produced for use in predicting the classification of new samples of unknown class.

54 ENVIRONMENTAL SCIENCES↗