Alife machine learning system and method
Roughly described, a problem solving platform distributes the solving of the problem over a evolvable individuals, each of which also evolves its own pool of actors. The actors have the ability to contribute collaboratively to a solution at the level of the individual, instead of each actor being a candidate for the full solution. Populations evolve both at the level of the individual and at the level of actors within an individual. In an embodiment, an individual defines parameters according to which its population of actors can evolve. The individual is fixed prior to deployment to a production environment, but its actors can continue to evolve and adapt while operating in the production environment. Thus a goal of the evolutionary process at the level of individuals is to find populations of actors that can sustain themselves and survive, solving a dynamic problem for a given domain as a consequence.
1. A computer-implemented data mining system, for use with a data mining training database containing samples of training data, each of the samples having at least one data entry, comprising:
a training system including at least one evolutionary unit with an associated processor subsystem having stored accessibly thereto a candidate database having a pool of candidate individuals, each candidate individual having a respective pool of actors and a respective set of parameters which influence evolution of the actors in the respective pool of actors, the at least one evolutionary unit with associated processor subsystem, for each given one of at least a subset of the candidate individuals:
evolving the parameters of the given candidate individual in dependence upon the samples of training data, and
evolving each of the actors of the given candidate individual in dependence upon the samples of training data and further in dependence upon the parameters of the given candidate individual; and
a production system including an associated processor subsystem having stored accessibly thereto a production population database of deployable individuals selected from the candidate individuals wherein each of the deployable individuals includes a fixed set of evolved parameters:
deploying a deployable individual from the production population database to assert actions in dependence upon production data, the actions asserted by the deployable individual being dependent on a response of the actors of the deployable individual to the production data, and
an evolutionary engine for further evolving the actors of the deployable individual after deployment in dependence upon the fixed set of evolved parameters of the given individual and the production data.
2. A computer-implemented data mining system, for use with a data mining training database containing a plurality of training samples, each of the training samples having a set of at least one data entry, comprising:
a memory storing a candidate database having a pool of candidate individuals, each candidate individual in at least a non-null subset of the candidate individuals having a respective pool of actors and a respective set of parameters which influence evolution of actors in the respective pool of actors, each candidate individual further having associated therewith an indication of a respective fitness estimate;
a candidate pool processor which:
evolves the set of parameters of each individual in dependence upon the training samples;
applies the training samples to individuals from the candidate pool, each individual being tested processing its actors through N>1 actor-level cycles of activity in response to each of the data entries, the actors of the individual collaborating to assert an action for the individual in response to the data entry, and
updates the fitness estimate associated with each of the individuals being tested in dependence upon both the training data and actions asserted by the respective individual in a battery of trials;
a competition processor including an associated processor subsystem which selects individuals for discarding from the candidate pool in dependence upon predefined criteria;
a candidate harvesting processor including an associated processor subsystem providing for deployment of selected ones of the individuals from the candidate pool;
a production system that deploys selected ones of the individuals from the candidate pool to assert actions in dependence upon production data, the actions asserted by the deployed individual being dependent on a response of the actors of the deployed individual, wherein each of the deployed individuals includes a fixed set of evolved parameters;
an evolutionary engine for further evolving the actors of the deployed individual after deployment in dependence upon the fixed set of evolved parameters of the given individual and the production data.
3. The system of claim 2 , wherein, applying the training samples to a particular one of the individuals:
each of the actors of the particular individual asserts an actor-level action at each of the actor-level cycles in response to the training data; and
the action asserted for the individual in response to the data entry is responsive to the actor-level actions asserted by the actors at the actor-level cycles.
4. The system of claim 3 , wherein each of the individuals has associated therewith a respective library of actor-level actions that are available to each of the individual's actors for assertion during the actor-level cycles,
wherein each of the individuals has associated therewith a respective library of indicators in response to which the individual's actors can select an actor-level action to assert during the actor-level cycles,
one of the individuals in the candidate pool differs from a second one of the individuals in the candidate pool in their respective libraries of actor-level actions and/or indicators.
5. The system of claim 4 , further comprising a procreation processor including an associated processor subsystem which evolves new individuals for the candidate pool at least in part by crossover and/or mutation of the indicators and actor-level actions in the libraries of parent ones of the new individuals.
6. The system of claim 3 , wherein the actor-level actions are selected by each of the actors of the particular individual from a library of available actor-level actions,
wherein one of the available actor-level actions comprises asserting a signal readable by actors of the particular individual,
and wherein one of the actors of the particular individual asserts an actor-level action in response to signals asserted by other actors of the particular individual.
7. The system of claim 2 , wherein the actors of the individual collaborating to assert an action for the individual in response to the data entry includes the actors of the individual communicating among themselves in response to the data entry in one of the actor-level cycles.
8. The system of claim 2 , wherein each of the individuals in the subset of candidate individuals further has associated therewith an energy treasury for the individual as a whole,
wherein each of the actors in the pool of actors for the individual has associated therewith an energy level which affects the actor's ability to survive,
and wherein the candidate pool processor further:
changes the energy treasury of each of the individuals being tested in dependence upon changes to the fitness estimate of the individual; and
allocates the changes in the energy treasury among one or more of the actors of the respective individual.
9. The system of claim 8 ,
wherein each of the actors of the particular individual asserts an actor-level action at each of the actor-level cycles in response to the training data,
wherein the actor-level actions are selected by each of the actors of the particular individual from a library of available actor-level actions,
and wherein allocating the changes in the energy treasury among one or more of the actors of the respective individual comprises assertion by the actor of an actor-level action which entails an energy cost to the actor,
and wherein the candidate pool processor tombstones actors whose energy level falls below a predetermined minimum.
10. The system of claim 9 , wherein allocating the changes in the energy treasury among one or more of the actors of the respective individual further comprises assertion by the actor of an action which transfers energy from the individual's energy treasury to the actor.
11. A computer-implemented data mining process, for use with a database of deployable individuals, each of the deployable individuals having a respective pool of actors and a respective fixed set of evolved parameters which influence evolution of the actors in the respective pool of actors, the fixed set of evolved parameters of each individual having been evolved based on training data, the process comprising:
deploying one of the individuals from the database of deployable individuals to assert actions in dependence upon production data, the actions asserted by the deployable individual being dependent on a response of the actors of the deployable individual to the production data, and
further evolving by an evolutionary engine the actors of the deployable individual after deployment in dependence upon the fixed set of evolved parameters of the given individual and the production data.