Search

As a technique that can compactly represent complex patterns, machine learning has significant potential for predictive inference. K-fold cross-validation (CV) is the most common approach to ascertaining the likelihood that a machine learning outcome is generated by chance, and it frequently outperforms conventional hypothesis testing. This improvement uses measures directly obtained from machine learning classifications, such as accuracy, that do not have a parametric description. To approach a frequentist analysis within machine learning pipelines, a permutation test or simple statistics from data partitions (i.e., folds) can be added to estimate confidence intervals. Unfortunately, neither parametric nor non-parametric tests solve the inherent problems of partitioning small sample-size datasets and learning from heterogeneous data sources. The fact that machine learning strongly depends on the learning parameters and the distribution of data across folds recapitulates familiar difficulties around excess false positives and replication. A novel statistical test based on K-fold CV and the Upper Bound of the actual risk (K-fold CUBV) is proposed, where uncertain predictions of machine learning with CV are bounded by the worst case through the evaluation of concentration inequalities. Probably Approximately Correct-Bayesian upper bounds for linear classifiers in combination with K-fold CV are derived and used to estimate the actual risk. The performance with simulated and neuroimaging datasets suggests that K-fold CUBV is a robust criterion for detecting effects and validating accuracy values obtained from machine learning and classical CV schemes, while avoiding excess false positives., Comment: 40 pages, 24 figures

Throughout the Next STEP Phase A and Phase B, the Trash Compaction Processing System (TCPS) is being developed for a technology demonstration (TD) on the International Space Station (ISS) in 2025. For Phase A, two contractors built proof-of-concept hardware. One contractor was chosen to build the TD hardware for Phase B. Both Phase A lessons learned and risk reduction activities at Ames Research Center (ARC) were used to write the TD science objectives, scope, and requirements. The work at ARC aims to retire technical risks and provide design data to TCPS developers and the ISS system integrators. This paper will summarize the lessons learned from the proof-of-concept hardware, the risk reduction activities, and how these lessons learned form the basis of the TD requirement matrix.

This report is an analysis of the Teacher and Student Advancement (TAP) System on student achievement and teacher engagement in Louisiana between 2010-11 and 2011-12. It analyzed: (1) student achievement data; (2) web-survey self-report questionnaire data for each TAP role; and (3) on-site interview and observation data; and data from (d) random-interval work-sampling. The data are from 17 TAP elementary, middle and high schools and from a group of propensity-score matched comparison schools. At the initiation of this analysis in 2010, the treatment schools had all completed at least two years with TAP; that is, they had accomplished the planning and initiation activities. The following study question were examined: (1) Do students perform better in TAP schools? (2) Do TAP teachers teach differently? and (3) Is the engagement of teachers with the TAP System and its components related to results for students, teachers and schools? [The report was prepared by Interactive, Inc.]

KNET is a general-purpose shell for constructing expert systems based on belief networks and decision networks. Such networks serve as graphical representations for decision models, in which the knowledge engineer must define clearly the alternatives, states, preferences, and relationships that constitute a decision basis. KNET contains a knowledge-engineering core written in Object Pascal and an interface that tightly integrates HyperCard, a hypertext authoring tool for the Apple Macintosh computer, into a novel expert-system architecture. Hypertext and hypermedia have become increasingly important in the storage management, and retrieval of information. In broad terms, hypermedia deliver heterogeneous bits of information in dynamic, extensively cross-referenced packages. The resulting KNET system features a coherent probabilistic scheme for managing uncertainty, an objectoriented graphics editor for drawing and manipulating decision networks, and HyperCard's potential for quickly constructing flexible and friendly user interfaces. We envision KNET as a useful prototyping tool for our ongoing research on a variety of Bayesian reasoning problems, including tractable representation, inference, and explanation., Comment: Appears in Proceedings of the Fourth Conference on Uncertainty in Artificial Intelligence (UAI1988)

In recent years, researchers in decision analysis and artificial intelligence (Al) have used Bayesian belief networks to build models of expert opinion. Using standard methods drawn from the theory of computational complexity, workers in the field have shown that the problem of probabilistic inference in belief networks is difficult and almost certainly intractable. K N ET, a software environment for constructing knowledge-based systems within the axiomatic framework of decision theory, contains a randomized approximation scheme for probabilistic inference. The algorithm can, in many circumstances, perform efficient approximate inference in large and richly interconnected models of medical diagnosis. Unlike previously described stochastic algorithms for probabilistic inference, the randomized approximation scheme computes a priori bounds on running time by analyzing the structure and contents of the belief network. In this article, we describe a randomized algorithm for probabilistic inference and analyze its performance mathematically. Then, we devote the major portion of the paper to a discussion of the algorithm's empirical behavior. The results indicate that the generation of good trials (that is, trials whose distribution closely matches the true distribution), rather than the computation of numerous mediocre trials, dominates the performance of stochastic simulation. Key words: probabilistic inference, belief networks, stochastic simulation, computational complexity theory, randomized algorithms., Comment: Appears in Proceedings of the Fifth Conference on Uncertainty in Artificial Intelligence (UAI1989)

In recent years, researchers in decision analysis and artificial intelligence (AI) have used Bayesian belief networks to build models of expert opinion. Using standard methods drawn from the theory of computational complexity, workers in the field have shown that the problem of exact probabilistic inference on belief networks almost certainly requires exponential computation in the worst ease [3]. We have previously described a randomized approximation scheme, called BN-RAS, for computation on belief networks [ 1, 2, 4]. We gave precise analytic bounds on the convergence of BN-RAS and showed how to trade running time for accuracy in the evaluation of posterior marginal probabilities. We now extend our previous results and demonstrate the generality of our framework by applying similar mathematical techniques to the analysis of convergence for logic sampling [7], an alternative simulation algorithm for probabilistic inference., Comment: Appears in Proceedings of the Sixth Conference on Uncertainty in Artificial Intelligence (UAI1990)

As part of the NASA Next Space Technologies for Exploration Partnerships (NextSTEP) Trash Compaction and Processing System (TCPS) program, the Generation 2 (Gen 2) Heat Melt Compactor (HMC) tests a variety of trash models (nominal, high liquid, and high cloth) in order to evaluate the technical risk associated with processing spacecraft trash. The trash components used in the trash models not only impacts the final solid trash disk quality, but also the water and gas effluent compositions. In order to better design an auxiliary system to treat the effluent streams, both the input trash components and the effluent must be characterized. Evolved gas analysis (EGA) methods are used to characterize the input trash components. The EGA results will be used to determine various HMC operational parameters, as well as to determine the optimal operation for the TCPS system.

The processing of trash and waste is a welcome and valuable addition to humans living and working in space. Besides the obvious desire to have a pleasant and productive habitation environment, trash management has many practicalbenefits for crew health, resource recovery, and volume reclamation through garbage compaction. The Trash Compaction and Processing System (TCPS), which is a NASA project to develop a trash processing system for long-duration spaceflight, is currently undergoing concept development with engineering prototype validation through two contracted efforts. The development efforts are being supported with activities associated with the NASA Generation 2 Heat Melt Compactor (HMC). The HMC is a facility that compacts trash, recovers water, heats the trash to eliminate biological activity, and manages gas and vapor effluents. The resulting residual processed trash is a compact tile that is free of biological growth and that can be used for augmenting radiation shields. The work being conducted with the HMC focuses on high risk technical areas with respect to operations, subsystem performance, and ISS effluent management interface requirements. This paper gives an overview of the technical risks and the current use of the HMC as a facility for reducing risk.

The Heat Melt Compactor (HMC) system has been developed to manage the waste generated on board spacecraft during long-duration missions. The quantity and composition of spacecraft trash depends upon the mission and can have a very high daily variability. This requires a flexible system able to manage extreme waste scenarios. Most missions will generate on average about one kilograms of trash per astronaut per day, derived mainly from the spacecraft logistics supplies and consisting of clothing, food & beverage residues, packaging, paper, plastic, hygiene wipes, and many other personal or scientific items used and discarded by the crew. Uncontained and unprocessed waste is a health hazard and a habitat storage problem. However, trash also contains valuable resources such as water. HMC is designed to provide volume reduction, microbial safening and stabilization, water recovery, and radiation shielding material. The final byproduct generated by HMC is a sterilized tile with the consistency of hard plastic that can be safely handled, easily stored, and used for radiation protection. This paper provides the summary of an extensive campaign of testing performed using the HMC system to simulate different nominal and extreme operational scenarios and to generate the data necessary to finalize requirements for proto-flight hardware to be deployed to an International Space Station (ISS) EXPRESS Rack.

Greenhouse gas (GHG) emissions estimation methods that use atmospheric trace gas observations, including inverse modeling techniques, perform better when carbon dioxide (CO2) fluxes are more accurately transported and dispersed in the atmosphere by a numerical model. In urban areas, transport and dispersion is particularly difficult to simulate using current mesoscale meteorological models due, in part, to added complexity from surface heterogeneity and fine spatial/temporal scales. It is generally assumed that the errors in GHG estimation methods in urban areas are dominated by errors in transport and dispersion. Other significant errors include, but are not limited to, those from assumed emissions magnitude and spatial distribution. To assess the predictability of simulated trace gas mole fractions in urban observing systems using a numerical weather prediction model, we employ an Eulerian model that combines traditional meteorological variables with multiple passive tracers of atmospheric CO2 from anthropogenic inventories and a biospheric model. The predictability of the Eulerian model is assessed by comparing simulated atmospheric CO2 mole fractions to observations from four in situ tower sites (three urban and one rural) in the Washington DC/Baltimore, MD area for February 2016. Four different gridded fossil fuel emissions inventories along with a biospheric flux model are used to create an ensemble of simulated atmospheric CO2 observations within the model. These ensembles help to evaluate whether the modeled observations are impacted more by the underlying emissions or transport. The spread of modeled observations using the four emission fields indicates the model's ability to distinguish between the different inventories under various meteorological conditions. Overall, the Eulerian model performs well; simulated and observed average CO2 mole fractions agree within 1% when averaged at the three urban sites across the month. However, there can be differences greater than 10% at any given hour, which are attributed to complex meteorological conditions rather than differences in the inventories themselves. On average, the mean absolute error of the simulated compared to actual observations is generally twice as large as the standard deviation of the modeled mole fractions across the four emission inventories. This result supports the assumption, in urban domains, that the predicted mole fraction error relative to observations is dominated by errors in model meteorology rather than errors in the underlying fluxes in winter months. As such, minimizing errors associated with atmospheric transport and dispersion may help improve the performance of GHG estimation models more so than improving flux priors in the winter months. We also find that the errors associated with atmospheric transport in urban domains are not restricted to certain times of day. This suggests that atmospheric inversions should use CO2 observations that have been filtered using meteorological observations rather than assuming that meteorological modeling is most accurate at certain times of day (such as using only mid-afternoon observations).