Smart Sampling Machine Learning
An AI-based solution to the “too heavy to compute” challenge
What is MoCaX Intelligence?
MoCaX Intelligence is an AI-driven accelerator that dramatically speeds up complex computations, reducing subtantially the need for costly IT infrastructure.
Key Benefits
- 10x-1000x Faster: Accelerates heavy computations by orders of magnitude.
- 10x-1000x Cost Reduction: Reduces infrastructure expenses significantly.
Which Calculations Benefit?
MoCaX is ideal for any computation requiring repeated evaluations of the same routine. Strong candidates include:
- Monte Carlo-based methods
- Optimization problems
- Historical simulations
How Does It Work?
In an “off-line” step, MoCaX’s proprietary Machine Learning (ML) technology intelligently samples the input space of computationally heavy routines. By learning from a minimal number of sample points, it replicates these routines with high accuracy while being orders of magnitude faster to evaluate.
Then, in an “on-line” step, this highly accurate and ultra-fast version of the heavy routine is used in the calculation instead of the original one, provoking a substantial gain in speed and compute demand with no practical impact to the final result.
How Is MoCaX Different from Standard Machine Learning?
Unlike traditional ML techniques like Deep Neural Networks (DNNs), which rely on random or manual sampling, MoCaX leverages exponential convergence properties of Chebyshev Tensors to intelligently sample the space. On this way, MoCaX is able to replicate complex routines from a very low number of sampling points.
Because of those exponential convergence properties, MoCaX ML techniques can replicate a vast range of functions and compute routines with very high accuracy, and from minimal compute effort.
MoCaX results are unparalleled in the ML world.
Example: In this peer-reviewed paper, MoCaX technology trained our Machine Learning models for sophisticated stochastic pricing models in just 15 CPU-hours, compared to 5,000+ CPU-hours required by optimised and independently calibrated standard DNNs. A 200x efficiency gain.
Are MoCaX Models a “Black Box”?
Unlike traditional ML frameworks, MoCaX models are fully analytical and trackable, ensuring no “black boxes” are used at any point in the enhanced computation.
Is My Computation a Good Fit for MoCaX?
If your calculation is slow, expensive, and has a heavy routine that is evaluated repeatedly (e.g. a Monte Carlo simulation), MoCaX can almost certainly significantly optimize it.
How does MoCaX integrate into my existing compute engine?
MoCaX integrates into existing computational frameworks. Once trained on a small data set, the MoCaX model replaces the original routine, delivering ultra-fast and highly accurate results with no disruption to the existing workflow.
Who Should Try MoCaX Solutions?
Any computation that is slow or expensive to run is a strong candidate to benefit from MoCaX. Contact us for a free online assessment.
Next Steps After Assessment
If your calculation is a candidate, we will conduct a proof-of-concept to demonstrate MoCaX’s benefits. Our solutions aim to accelerate calculations and reduce costs by at least 10x, often achieving gains of 100x or 1,000x. You can then license our solution at a fraction of the savings it produces, ensuring a win-win for everyone.