• Home
• Services
• Products
• About Us
• Contact

Its unique data model is optimized for handling complex numbers and provides a significant increase in performance over other solvers.  LinCoS includes a rich library of partitioners, preconditioners and iterative methods for both serial and parallel computing environments.

 LinCoS Overview (pdf)

 LinCoS Whitepaper (pdf)

Machine Learning Algorithm Library

Machine learning is a broad field of computational science, strongly grounded in modern mathematics, that encompasses systems that learn and evolve based on data from their environment.  Machine learning methodologies create descriptive and predictive models representing the relationships between inputs and outputs of a system.  Both simulated and real data from multiple sources can be used to create models that “teach” themselves from existing examples of a system.  The design process is similar to how humans learn from experience.

Machine learning methods can be applied in numerous situations but are most useful in applications where formal, complete algorithms for the solution either are not yet developed, or are too costly and complicated for practical use.  “Costly and complicated” can mean not only the time and cost to develop the algorithm, but also the time and cost to compute the results.  Machine learning-based solutions often can by-pass time consuming calculations required by traditional, computational-based methods producing results much more quickly.

We have more than a decade of experience developing and implementing machine learning-based solutions in real-world applications including:

• Industrial process control

• Defect recognition and sizing within in-line pipeline inspection data

• Anomaly discovery and classification in multivariate time series data

This experience is captured in a library of algorithms and tools that can be called upon to create custom solutions for a wide variety of customer needs.  To learn more about how machine learning methodologies can be applied to challenges in the energy sector, please see our whitepaper:

Machine Learning Opportunities in the Energy Sector (pdf)

Data Optimizers

Optimus™ is a data optimization tool that increases the performance of linear programming (LP) and mixed-integer linear programming (MIP) solvers by reordering and restructuring a system’s data before passing the data on to the solver.  Optimus can significantly reduce solution times for many very large optimization problems – in some cases from days down to hours.  Optimus can easily be interfaced to the leading commercial LP and MIP solvers or used in conjunction with Sparsix's own custom solvers.

Parallel Partitioning and Preconditioning Tools

Despite the increasing power of multi-core processors, many computationally demanding analysis applications are not easily modified to perform certain tasks in parallel.  Traditional numerical analysis techniques are often based on algorithms and methodologies that predate such architectures.  To truly reap the benefits of these modern architectures often requires more than simply rewriting some software - it requires a fundamentally new approach.

Sparsix has extensive experience implementing parallel computing architectures, particularly in the area of linear solvers.  By their very nature, linear solvers are difficult to parallelize and Sparsix had to re-think some of the standard methods that have been used for years to solve these problems. 

Based on new parallel processing algorithms and methodologies, Sparsix has created a library of partitioning and preconditioning tools that provide significant performance increases and highly efficient scaling for our family of linear solvers.  These same methods can be applied to other solvers or analysis applications to dramatically improve their parallel processing capabilities.

Home | Services | Products | About Us | Contact

© 2010 Sparsix Corporation. All rights reserved.