ADAPTIVE PARAMETER TRANSFER FOR LEARNING MODELS

1 . A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computer system, effectuate operations comprising:

obtaining, with a computer system, a directed graph encoding a symbolic artificial intelligence (AI) model used by a first entity, the directed graph comprising a first set of vertices and a first set of edges associating pairs of vertices of the first set of vertices, wherein:

each respective vertex of the first set of vertices is associated with a vertex status and is labeled with a category selected from a set of mutually-exclusive categories, and

a vertex of the first set of vertices is associated with a conditional statement that is indicated as triggerable by the first entity;

determining, with the computer system, a set of features based on the directed graph, the set of features comprising an identifier of a graph portion template, wherein each respective vertex of the graph portion template of the graph portion template is labeled with a same category from the set of mutually-exclusive categories as a corresponding respective vertex of a graph portion of the directed graph and is associated with a same count of edges;

obtaining, with the computer system, a set of model parameter values for a machine learning model based on the graph portion template; and

providing, with the computer system, the set of model parameter values and the graph portion templates to the first entity, wherein the set of model parameter values are used by to determine an outcome score based on the directed graph.

2 . The medium of claim 1 , the operations further comprising:

obtaining a conditional statement parameter, wherein the conditional statement parameter is used by the conditional statement;

wherein obtaining the set of model parameter values comprises selecting a parameter of the set of model parameter values based on the conditional statement parameter.

3 . The medium of claim 1 , wherein the outcome score is a first outcome score, the operations further comprising:

determining a map indicating a first graph portion of the directed graph, wherein the map comprises identifiers for vertices of the directed graph; and

determining a second outcome score using the set of model parameters based on the map.

4 . The medium of claim 3 , wherein the directed graph is a first directed graph, and wherein determining the first graph portion comprises:

using a first neural network based on the first directed graph and the set of model parameter values to determine the first outcome score, wherein the machine learning model comprises the first neural network;

updating the map multiple times to generate a plurality of simulated directed graphs, wherein a first simulated directed graph matches the first graph portion with respect to vertex categories, and wherein each of the simulated directed graphs is a subgraph of the first directed graph;

determining a set of simulated outcome scores with the neural network based on the plurality of simulated directed graphs and the set of model parameter values, wherein the plurality of simulated outcome scores comprises the second outcome score, and wherein using the neural network based on the first simulated directed graph provides the second outcome score;

selecting the second outcome score of the set of simulated outcome scores based a difference between the second outcome score and the first outcome score;

selecting the first graph portion based on the selection of the second outcome score; and

sending an indicator associated with the first graph portion to the first entity.

5 . The medium of claim 3 , wherein the directed graph is a first directed graph, and wherein the set of features is a first set of features, and wherein determining the first graph portion comprises:

using a first neural network based on the first directed graph and the set of model parameter values to determine the first outcome score, wherein the machine learning model comprises the first neural network;

updating the map multiple times to generate a plurality of sets of features, wherein each respective set of features is different from the first set of features;

determining a set of simulated outcome scores based on the plurality of sets of features using the set of model parameter values and the neural network, wherein the plurality of simulated outcome scores comprises the second outcome score, and wherein using the neural network based on a first set of features provides the second outcome score;

selecting the second outcome score of the set of simulated outcome scores based a difference between the second outcome score and the first outcome score;

selecting the first graph portion based on the first graph portion being associated with first set of features; and

sending an indicator associated with the first graph portion to the first entity.

6 . The medium of claim 1 , the operations further comprising:

obtaining a plurality of directed graphs;

determining multiple sets of features, wherein each respective set of features is determined based on a respective directed graph of the plurality of directed graphs;

determining the set of model parameter values by training a version of the machine learning model based on the multiple sets of features and a set of training outputs;

storing the set of model parameter values in a record of a database, wherein the record is associated with a set of labels, and wherein a search through the database using the set of labels provides an identifier of the record.

7 . The medium of claim 6 , wherein determining the multiple sets of features comprises determining a set of values in a non-Euclidean domain based on the multiple sets of directed graphs, and wherein determining the set of values in the non-Euclidean domain comprises determining a matrix inversion of a diagonal of an adjacency matrix of the directed graph.

8 . The medium of claim 6 , the operations further comprising:

obtaining a first transaction score, wherein the first transaction score is associated with a first transaction indicated to have triggered or activated a first vertex of a first directed graph of the plurality of directed graphs;

obtaining a second transaction score, wherein the second transaction score is associated with a second transaction indicated to have triggered or activated a second vertex of a second directed graph of the plurality of directed graphs;

aggregating the first score and the second score to form an aggregated score, the aggregating comprising a summation operation; and

determining the set of model parameter values during a training operation based on the aggregated score.

9 . The medium of claim 6 , wherein obtaining the plurality of directed graphs comprises obtaining the plurality of directed graphs from a tamper-evident, distributed ledger.

10 . The medium of claim 1 , further comprising selecting the machine learning model based on the set of features, wherein obtaining the set of model parameter values comprises selecting a model parameter value based on the selected learning model.

11 . The medium of claim 10 , wherein the set of features is a first set of features, the operations further comprising determining a second set of features based on the selected learning model and the directed graph, wherein obtaining the set of model parameter values comprises selecting a model parameter value based on the second set of features.

12 . The medium of claim 11 , wherein the first set of features and the second set of features are orthogonal to each other.

13 . The medium of claim 1 , wherein the set of model parameter values is received at a computer device controlled by the first entity, the operations further comprising:

determining the outcome score using the set of model parameter values; and

in response to a determination that the outcome score satisfies a warning threshold, sending a message to the first entity associated with the warning threshold.

14 . The medium of claim 1 , where obtaining the set of model parameter values comprises obtaining the set of model parameter values from a record stored on a cloud computing resource.

15 . The medium of claim 1 , wherein obtaining the set of model parameter values comprises:

determining an entity role associated with the entity; and

selecting a model parameter value of the set of model parameter values based on the entity role.

16 . The medium of claim 1 , wherein the machine learning model comprises a neural network, and wherein the model parameter values comprise weights for neurons of a neural network.

17 . The medium of claim 1 , wherein the set of model parameter values is a first set of model parameter values, the operations further comprising:

obtaining a second set of model parameter values associated with an entity identifier;

determining a second outcome score based on the second set of model parameter values; and

based on a comparison between the first outcome score and the second outcome score, causing a transaction that updates a score associated with an entity identified by the entity identifier, wherein the score is stored in tamper-evident, distributed ledger.

18 . The medium of claim 1 , wherein determining the set of model parameter values comprises steps for determining the set of model parameter values.

19 . The medium of claim 1 , wherein determining the outcome score comprises steps for determining the outcome score.

20 . A method comprising:

obtaining, with a computer system, a directed graph encoding a symbolic artificial intelligence (AI) model used by a first entity, the directed graph comprising a first set of vertices and a first set of edges associating pairs of vertices of the first set of vertices, wherein:

each respective vertex of the first set of vertices is associated with a vertex status and is labeled with a category selected from a set of mutually-exclusive categories, and

a vertex of the first set of vertices is associated with a conditional statement that is indicated as triggerable by the first entity;

determining, with the computer system, a set of features based on the directed graph, the set of features comprising an identifier of a graph portion template, wherein each respective vertex of the graph portion template of the graph portion template is labeled with a same category from the set of mutually-exclusive categories as a corresponding respective vertex of a graph portion of the directed graph and is associated with a same count of edges;

obtaining, with the computer system, a set of model parameter values for a machine learning model based on the graph portion template; and

providing, with the computer system, the set of model parameter values and the graph portion templates to the first entity, wherein the set of model parameter values are used by to determine an outcome score based on the directed graph.

21 . A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computer system, effectuate operations comprising:

obtaining, with a first computer system controlled by a first entity, a first training set, the first training set comprising a plurality of pairs of symbolic AI models and logs of state of the symbolic AI models;

training, with the first computer system, a first machine learning model on the first training set, wherein training comprises iteratively adjusting parameters of the first machine learning model based on a first objective function;

after training the first machine learning model, providing the first machine learning model to a second computer system controlled by a second entity, wherein the second entity does not have access to at least some of the first training set;

obtaining, with the second computer system, a second training set, the second training set comprising a plurality of pairs of symbolic AI models and logs of state of the symbolic AI models, the second training set being different at least in part from the first training set;

training, with the second computer system, a second machine learning model that includes the first machine learning model on the second training set, wherein training comprises iteratively adjusting parameters of the second machine learning model based on a second objective function; and

after training the second machine learning model, storing, with the second computer system, the second machine learning model in memory.

22 . The medium of claim 21 , wherein the second machine learning model comprises:

the first machine learning model as a sub-model having an output;

an error-correcting machine learning model as a sub-model having an input based on the output of the first machine learning model, wherein training the second machine learning model comprises iteratively adjusting parameters of the error-correcting machine learning model without changing parameters of the of the first machine learning model while adjusting parameters of the error-correcting machine learning model.

23 . The medium of claim 21 , wherein:

parameters of the second machine learning model are initialized to values of parameters of the first machine learning model before being iteratively adjusted during training of the second machine learning model.

24 . The medium of claim 21 , wherein the first and second machine learning models are non-symbolic artificial intelligence models comprising steps for machine learning.