Twincher
error-free AI beyond neural networks
Twincher is our original machine learning agent that, unlike neural networks, learns not patterns but geometries or topologies intrinsic to continuous data dependencies. In its elemental form twincher can study a transformation from X to Y to become able to quickly and accurately solve the inverse problem: find an X that yields a given Y. As distinct from neural networks being subject to a never-ending trade-off between errors and resources, twinchers provide a mechanism for completely eliminating errors (within certain formulation) under finite resources. In such a way they can serve as a new basis for applications in many areas such as robotics, medicial diagnostics and computer vision. Here are a few other points that are special about this technology.
Twinchers can be used to identify elemental objects, ascertain their parameters and decompose complex structures into sets of such elemental objects. Crucially, twinchers provide an inherent means for making their assessments robust or even indifferent with respect to general or any specific types of distortions, noise or features. Moreover, they can disentangle and quantify such features enabling possibilities for further analysis or single-shot learning.
When training twinchers it is possible to compute and reduce to any desired level the sensitivity to noise or any specific deviations in the input. This gives robustness and reliability necessary for dealing with real-world data. For example, twinchers can be used to read visual and other inputs to determine shapes, structures, motion or mechanical parameters of various objects. This enables their use in such areas as industrial diagnostics, tracking body pose and motion for sports and movie production, virtual reality and 3d scene perception.
Twinchers are smart. Instead of fitting examples of a training set, twinchers actively study given models of reality, identify subtle cases, and learn them with greater thoroughness. In such a way twinchers masterfully achieve the state of “exact knowledge”. This makes them extremely efficient in comprehending abstract models under resource-constrained conditions. This can be further exploited to automate the development of ensembles of abstract models (mathematical or computational), their logical interaction, and cumulative aggregation for a given context.
Currently, the project is at the stage of a working prototype that demonstrates the outlined properties. We are forming the team of the project, looking for investors and partners from industry as well as other areas of potential applications. Interested? Please let us know: contact@twincher.ai