From stock exchange matching engines to autonomous vehicles, modern safety-critical systems may be in a virtually infinite number of possible states. A rigorous approach to their design and governance requires the ability to reason symbolically about all such states. With Region Decomposition, Imandra automatically decomposes the state-space of a system and describes its possible future behaviors in a systematic manner. Each region represents a subset of all possible states and precisely describes both constraints on the inputs that will lead the system to that region, and constraints on the output which correspond to how the system will behave within such scenarios. Each region is equipped with a scenario generator, which will solve its input constraints in order to create concrete testing scenarios which push the system into that region. Combined, these regions fully describe the possible symbolic behaviors of the system. This collection of regions may be explored, analyzed, used to generate tests, documentation and as a basis for rigorous approaches to explainability.
Denis Ignatovich, co-founder and co-CEO of Imandra, explained it as follows, “When you’re playing chess, you try to think N-steps ahead – what your opponent might do and how the game may evolve. All of the possible configurations of the chess board comprise its so-called ‘state-space’. Complex systems like autonomous vehicles work similarly, except the set of possible states is virtually infinite. Region Decomposition allows you to explore, analyse, visualize and understand it.”
Grant Passmore, co-founder and co-CEO of Imandra said, “We built Imandra’s Region Decomposition with the goal of giving algorithm stakeholders automatic tools for ‘mapping’ and understanding the possible behaviors of a complex system, with an API that allows them to easily tap into this data for increased rigor in design, testing, documentation and explainability. Region Decomposition is inspired by Cylindrical Algebraic Decomposition (CAD), a powerful technique for reasoning about continuous systems. To lift this idea to algorithms at large, we’ve leveraged many recent advances in automated reasoning, including automated induction and nonlinear decision procedures. We’re grateful to our industrial partners for helping us ensure these techniques scale to large real-world problems.”
Imandra’s initial focus will be on applying Region Decomposition to safety of autonomous vehicles, financial trading systems and reinforcement learning algorithms. For more information, please visit Imandra’s blog at https://www.imandra.ai/media
Imandra Inc. (www.imandra.ai) is the world-leader in cloud-scale automated reasoning, democratizing deep advances in algorithm analysis and symbolic AI for making algorithms safe, explainable and fair. Imandra has been deep in R&D and industrial pilots over the past 5 years and has recently closed its $5mm Seed round led by several top deep-tech investors in US and UK. Imandra is headquartered in Austin, TX, and has offices in the UK and continental Europe.
SOURCE Imandra Inc.