International Business Machines Corporation

Fearghal O'Donncha

Emanuele Ragnoli

Sean A McKenna

Ernesto Arandia

Eli M Dow

Ying Liu

Sergiy Zhuk

Francesco Fusco

Frank Suits

Campbell D Watson

Charles Arthur Jochim

Seshu Tirupathi

Noreen O'Brien

Bei Chen

Mathieu Sinn

Julien Monteil

Jakub Marecek

Ambrish Rawat

Harry Roger Kolar

Frank Liu

Albert Akhriev

Mark Purcell

Michael Barry

Callum Sutherland

Massimiliano Meneghin

Philipp Haehnel

Techniques of facilitating deep learning model rescaling by computing devices. In one example, a system can comprise a processor that executes computer executable components stored in memory. The computer executable components can comprise: a rescaling component; and a forecasting component. The rescaling component can determine a scaling ratio that maps low mesh resolution predictive data output by a partial differential equation (PDE)-based model for a sub-domain to high-resolution observational or ground-truth data for a domain comprising the sub-domain. The forecasting component can generate high mesh resolution predictive data for the domain with a machine-learning model using input data of the PDE-based model and the scaling ratio.

Superresolution and consistency constraints to scale up deep learning models

Embodiments for managing a wave energy converter (WEC) device by one or more processors are described. At least one environmental characteristic associated with a WEC device in a body of water is received. A prediction of wave conditions on the body of water is calculated based on the at least one environmental characteristic. A signal representative of the prediction of wave conditions is generated.

Methods and systems for wave energy generation prediction and optimization

Embodiments for deep learning for partial differential equation (PDE)-based models by a processor. A trained forecasting model and consistency constraints may be generated using a PDE-based model, a discretization of the PDE-based model, historical inputs the of the PDE-based model, and a representation of consistency constraints to generate a predictive output.

Deep learning for partial differential equation (PDE) based models

A method and system for outputting a state of a physical system using a calibrated model of the physical system, where the calibrated model is used to generate a model prediction. The system includes a plurality of sensors connected to a routing node are used to monitor measured data of the physical system. A first sensor of the plurality of sensors includes a logic module configured to determine an uncertainty quantification, and to combine the uncertainty quantification with the model prediction to output the state of the physical system.

Integration of physical sensors in a data assimilation framework

Embodiments for dynamically distributing loads in computational rendering in a computing environment. A computational rendering model on a computational rendering to exploit nested recursive parallelism within a heterogenous computing architecture to enable communication overlap, memory transfer, and data and task management, wherein the computational rendering model is developed for the heterogenous computing architecture.

Dynamic distribution of loads across heterogeneous computing structures in computational rendering

The protection of water providing entities from loss due to environmental events is disclosed including receiving structured and unstructured data related to a water resource, generating time series for a plurality of index items, generating training data and testing data based on the time series for the plurality of index items, determining correlations between the time series of the index items using the generated training data, determining a classification boundary as a strike level based on the determined correlation, generating a protection strategy based on the strike level, applying the protection strategy to the testing data to generate a resulting data time series, and presenting the resulting data time series via a display of a computing device to at least one user of the computing device.

Protection of water providing entities from loss due to environmental events

Embodiments for estimating ice formation and depth by a processor. Ice formation and ice depth of a selected region of water may be cognitively forecasted using a prediction model such that the prediction model identifies similar characteristics between the selected region of water and one or more alternative regions of water by analyzing content of one or more data sources associated with an ontology of concepts representing a domain knowledge related to the selected region of water and the one or more alternative regions.

Ice formation and ice depth estimation system

Embodiments for implementing intelligent vessel trajectory planning and adapting by a processor. A trajectory of a vessel may be dynamically determined according to forecasted wave conditions using a surrogate wave model, a wave forecasting model, one or more user defined constraints, or a combination thereof.

Dynamic adaption of vessel trajectory using machine learning models

A mechanism is provided in a data processing system for minimizing uncertainty envelopes in trajectories of evolving ensemble members. The mechanism generates a trajectory forecast of each member object of an ensemble based on an initial state-space and a model for predicting trajectories of the member objects to generate a plurality of trajectory forecasts. Each of the plurality of trajectory forecasts has an individual uncertainty envelope. The mechanism applies a classification algorithm on the plurality of trajectory forecasts to identify at least one group of member objects having similar trajectory forecasts, generates a reduced ensemble of member objects including the identified group of member objects, and reconfigures the state-space and the model for predicting trajectories. The mechanism generates an updated trajectory forecast of each member object of the reduced ensemble based on the reconfigured state-space and the reconfigured model for predicting trajectories of the member objects.

Minimizing uncertainty envelopes in trajectories of evolving ensemble members

Embodiments for intelligent forecasting of material concentrations in a fluid by a processor in a computing environment. A material concentration of a material in a fluid may be predicted according to one or more continuous stirred tank reactor (CSTR) surrogate models on statistical flow trajectories of the fluid defined by a principle component analysis (PCA) operation of a system.

Combining feature selection and surrogate models for the forecast of material concentration in fluids

A fluid is modeled as a set of discrete particles. Each of the particles is associated with one or more properties, and a spatial distance comprising a smoothing length over which the one or more properties are to be smoothed. A corresponding trajectory is simulated for each of the particles. The corresponding trajectory is used to formulate a first solution for simulating transport within the fluid. A first predicted error is determined for the first solution. An iterative adjustment is performed to at least one of: a quantity of particles, the smoothing length, or the one or more corresponding properties, to formulate a second solution for simulating transport with the fluid, and a second predicted error is determined for the second solution, until the second predicted error is within a predefined boundary.

Performing Lagrangian particle tracking with adaptive sampling to provide a user-defined level of performance

A computer system, method, and computer readable product are provided for cognitive initialization of large-scale advection-diffusion models. In various embodiments, parameters are identified for configuring an advection-diffusion model to be run. Then a natural language processing search is conducted for these parameters in relevant literature to find corresponding numerical values. These search results are compared, and values for the parameters are set for the advection-diffusion model, and the model is run.

Cognitive initialization of large-scale advection-diffusion models

A mechanism is provided for targeted, iterative refinement of advection-diffusion model grid resolution. A simulation is executed of advection-diffusion processes for an area associated with a user-defined location. Responsive to detecting an event within a subset of cells of a first set of cells, the subset of cells is refined such that each of the subset of cells is gridded to comprise a second set of cells with dimensions less than the first set of cells. A number of run-time floating-point operations per second (FLOPS) of the model is computed for each of the first set of cells and the second set of cells. Based on a collective time associated with a computed number of FLOPS failing to exceed the user-defined constraint, the simulation continues to execute with the first set of cells and the second set of cells. The process is repeated until a predetermined resolution value is met.

Forecast-based refinement and load balancing for prediction of advection-diffusion processes

A distributed computer system includes a distributed processor, a distributed memory, and a simulation engine (SE). The SE includes a simulation I/O coupler that links a first variable of a first simulation model I/O data structure to a second variable of a second simulation model I/O data structure. The SE includes a natural language processing system that extracts a first variable description associated with the first variable, determines similar character strings to the first variable description from an information corpus, and ranks the determined character strings based upon similarity to the first variable description. The SE links the first variable to the second variable if the rank of an equal character string to the second variable description is greater than a rank threshold. The SE may augment the simulation model I/O data structures by writing a value of the first variable to a value of the second variable, or visa versa.

Data analysis for automated coupling of simulation models

Embodiments for managing aquatic infrastructure by one or more processors are described. A vehicle is detected within a vicinity of an aquatic infrastructure device. A wireless capability level of the vehicle is determined. Based on the determined wireless capability level of the vehicle, at least one of a wireless power transfer between the vehicle and the aquatic infrastructure device or a wireless data transfer between the vehicle and the aquatic infrastructure device is enabled.

Growing community of inventors

Galway, Ireland

Fearghal O'Donncha