No abstract

One of the key factors for the acceptance of expert systems in real-world domains is the ability to explain their reasoning (Buchanan & Shortliffe, 1984; Henrion & Druzdzel, 1990). This paper describes the basic properties that characterise explanation methods and reviews the methods developed to date for explanation in Bayesian networks.

Spiegelhalter and Lauritzen [15] studied se quential learning in Bayesian networks and proposed three models for the representation of conditional probabilities. A forth model, shown here, assumes that the parameter dis tribution is given by a product of Gaussian functions and updates them from the >. and 1r messages of evidence propagation. We also generalize the noisy OR-gate for multival ued variables, develop the algorithm to com pute probability in time proportional to the number of parents (even in networks with loops) and apply the learning model to this gate.

Hydrodynamic signals from turbulence and waves may provide marine invertebrate larvae with behavioral cues that affect the pathways and energetic costs of larval delivery to adult habitats. Oysters (Crassostrea virginica) live in sheltered estuaries with strong turbulence and small waves, but their larvae can be transported into coastal waters with large waves. These contrasting environments have different ranges of hydrodynamic signals, because turbulence generally produces higher spatial velocity gradients, whereas waves can produce higher temporal velocity gradients. To understand how physical processes affect oyster larval behavior, transport and energetics, we exposed larvae to different combinations of turbulence and waves in flow tanks with (1) wavy turbulence, (2) a seiche and (3) rectilinear accelerations. We quantified behavioral responses of individual larvae to local instantaneous flows using twophase, infrared particle-image velocimetry. Both high dissipation rates and high wave-generated accelerations induced most larvae to swim faster upward. High dissipation rates also induced some rapid, active dives, whereas high accelerations induced only weak active dives. In both turbulence and waves, faster swimming and active diving were achieved through an increase in propulsive force and power output that would carry a high energetic cost. Swimming costs could be offset if larvae reaching surface waters had a higher probability of being transported shoreward by Stokes drift, whereas diving costs could be offset by enhanced settlement or predator avoidance. These complex behaviors suggest that larvae integrate multiple hydrodynamic signals to manage dispersal tradeoffs, spending more energy to raise the probability of successful transport to suitable locations.

An experimental investigation of the deformation and breakup properties of turbulent round liquid jets in uniform gaseous crossflows is described. Pulsed shadowgraph and holograph observations were obtained for turbulent round liquid jets injected normal to air crossflow in a shock tube. Crossflow velocities of the air behind the shock wave relative to the liquid jet were subsonic (36-90 m=s) and the air in this region was at normal temperature and pressure. Liquid injection was done by a pressure feed system through round tubes having inside diameters of 1 and 2 mm and length-to-diameter ratios greater than 100 to provide fully developed turbulent pipe flow at the jet exit. Test conditions were as follows: water and ethyl alcohol as test liquids, crossflow Weber numbers based on gas properties of 0-282, streamwise Weber numbers based on liquid properties of 1400-32,200, liquid/gas density ratios of 683 and 845, and jet exit Reynolds numbers based on liquid properties of 7100-48,200, all at conditions in which direct effects of liquid viscosity were small (Ohnesorge numbers were less than 0.12). Measurements were carried out to determine conditions required for the onset of breakup, ligament and drop sizes along the liquid surface, drop velocities after breakup, liquid column breakup as whole, rates of turbulent primary breakup, and liquid column trajectories. Phenomenological theories proved to be quite successful in interpreting and correlating the measurements.

The influence of a high electric field applied on both fluid flow and particle velocities is quantified at large Peclet numbers. The experiments involved simultaneous particle image velocimetry and flow rate measurements. These are conducted in polydimethylsiloxane channels with spherical nonconducting polystyrene particles and DI water as the background flow. The high electric field tests produced up to three orders of magnitude higher electrokinetic velocities than any previous reports. The maximum electroosmotic velocity and electrophoretic velocity measured were 3.55 and 2.3 m/s. Electrophoretic velocities are measured over the range of 100 V/cm < E < 250 000 V/cm. The results are separated according to the different nonlinear theoretical models, including low and high Peclet numbers, and weak and strong concentration polarization. They show good agreement with the models. Such fast velocities could be used for flow separation, mixing, transport, control, and manipulation of suspended particles as well as microthrust generation among other applications.

Díez's algorithm for the noisy MAX is very efficient for polytrees, but when the network has loops, it has to be combined with local conditioning, a suboptimal propagation algorithm. Other algorithms, based on several factorizations of the conditional probability of the noisy MAX, are not as efficient for polytrees but can be combined with general propagation algorithms such as clustering or variable elimination, which are more efficient for networks with loops. In this article we propose a new factorization of the noisy MAX that amounts to Díez's algorithm in the case of polytrees and at the same time is more efficient than previous factorizations when combined with either variable elimination or clustering.

No abstract