Computational naturally-occurring-action interpreter for machine learning
Inventors
Billinge, Simon J. L. • Rakita Shlafstein, Yevgeny
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12688468-B1
Publication Date
2026-07-21
Expiration Date
Abstract
Disclosed are methods, systems, device, and other implementations, including a method that includes receiving action data representative of a sequence of actions resulting in an outcome, the sequence of actions and the resulting outcome defining an observable process, transforming the action data into a normalized representation recognizable by a computing-based analysis engine, and processing the normalized representation by the analysis engine to generate an alternative process comprising one or more of, for example, alternative sequence of actions, and/or an alternative outcome.
Core Innovation
A naturally-occurring-action interpreter encodes an observed action sequence and an observed resulting outcome into machine-learnable normalized representations. The encoded representation is recognizable by a machine learning analysis engine and is based on action data, including a sequence of actions and respective properties associated with the sequence of actions and the resulting outcome.
The normalized representations are expressed as a directed acyclic graph with nodes and edges, optionally derived using natural language processing. The normalized representations are also converted into element representations (vector/matrix) or expressed as coded/algebraic expressions.
The machine learning analysis engine processes the normalized representation of a first observable process and accesses one or more databases maintaining multiple records of previously transformed normalized representations of other different processes. Based on correlations determined from the normalized representation of the first observable process and the stored normalized records, the engine generates an alternative process for the first observable process, comprising one or more of an alternative sequence of actions and/or an alternative outcome.
Claims Coverage
The independent claims cover four implementation forms (method, computing system, and non-transitory computer readable media) for using a learning engine to generate an alternative process from correlations learned over normalized representations, using three inventive features: receiving action data for a first observable process; transforming into a normalized representation recognizable by a machine learning analysis engine; and processing correlations with stored normalized records to generate an alternative sequence of actions and/or an alternative outcome.
Normalized representation of an observed action sequence and outcome
Receiving action data representative of a sequence of actions resulting in an outcome, the sequence of actions, the resulting outcome, and respective properties associated with the sequence of actions and the resulting outcome defining a first observable process; and transforming the action data into a normalized representation recognizable by a machine learning analysis engine.
Correlation-based alternative process generation from stored normalized records
The machine learning analysis engine configured to access and process one or more databases maintaining multiple records of previously transformed normalized representations of other different processes with corresponding sequences of action, resulting outcomes, and properties; and processing the normalized representation by the machine learning analysis engine to generate, based on correlations determined from the normalized representation of the first observable process and the multiple records of the previously transformed normalized representations of the other processes, an alternative process for the first observable process, comprising one or more of an alternative sequence of actions determined from the correlations, or an alternative outcome determined from the correlations.
Computing system with learning engines and databases for correlations
A computing system comprising an input stage to receive one or more input data records; and a processor-based controller, implementing one or more learning engines, in communication with a memory device to store programmable instructions, to receive action data, transform the action data into a normalized representation recognizable by a machine learning analysis engine, and process the normalized representation to generate an alternative process based on correlations determined from the normalized representation and multiple records.
Non-transitory computer readable media to generate alternative process from correlations
A non-transitory computer readable media storing a set of instructions to receive action data, transform the action data into a normalized representation recognizable by a machine learning analysis engine, and process the normalized representation to generate an alternative process for the first observable process based on correlations determined from the normalized representation and multiple records.
Across the independent claims, the inventive coverage centers on normalizing action-data for a first observable process into a machine-recognizable form and using machine learning over stored normalized records of other processes to determine correlations that generate an alternative sequence of actions and/or an alternative outcome.
Stated Advantages
Bridging human intuition via community data using unified intuition/community knowledge.
Parameter reduction for materials discovery.
Adaptive/fuzzy recipes.
Potential use for autonomation/in-operando experimentation.
Documented Applications
Materials discovery using parameter reduction.
Autonomation/in-operando experimentation.
Interested in licensing this patent?