AI-based neighbor discovery search engine apparatuses, methods and systems

A processor-readable neighbor discovery search engine non-transient physical medium storing processor-executable components, the components, comprising:

• a component collection stored in the medium, including:
• an embedding neural network training component;
• wherein the embedding neural network training component, stored in the medium, includes processor-issuable instructions to:
• obtain, via at least one processor, an embedding neural network training request associated with a set of context objects;
• determine, via at least one processor, sample similarity evaluation metrics associated with the embedding neural network training request;
• determine, via at least one processor, for each context object in the set of context objects, a set of positive target samples that satisfy the sample similarity evaluation metrics for the respective context object;
• add, via at least one processor, to a training set, for each context object and each positive target sample in the respective set of positive target samples, a training example that comprises:
• (a) the respective context object and (b) a positive target sample;
• determine, via at least one processor, configuration parameters for an embedding neural network, wherein the embedding neural network comprises an input layer, a hidden layer, an embedding layer and an output layer, wherein the embedding layer facili-tates calculating embeddings that facilitate comparison of context objects based on embedding distance;
• train, via at least one processor, the embedding neural network using training examples in the training set, wherein the embedding neural network is trained using a selected training example by:
• (a) generating a context feature vector for the context object associated with the selected training example and providing the context feature vector for the input layer, (b) generating a target feature vector for the positive tar-get sample associated with the selected training example and providing the target feature vector for the output layer, (c) adjusting weights of the embedding neural network using a machine learning technique;
• and
• generate, via at least one processor, a datastructure that stores the adjusted weights of the embedding neural network.

A processor-implemented neighbor discovery search engine system, comprising:

• an embedding neural network training component means, to:
• obtain, via at least one processor, an embedding neural network training request associated with a set of context objects;
• determine, via at least one processor, sample similarity evaluation metrics associated with the embedding neural network training request;
• determine, via at least one processor, for each context object in the set of context objects, a set of positive target samples that satisfy the sample similarity evaluation metrics for the respective context object;
• add, via at least one processor, to a training set, for each context object and each positive target sample in the respective set of positive target samples, a training example that comprises:
• (a) the respective context object and (b) a positive target sample;
• determine, via at least one processor, configuration parameters for an embedding neural network, wherein the embedding neural network comprises an input layer, a hidden layer, an embedding layer and an output layer, wherein the embedding layer facili-tates calculating embeddings that facilitate comparison of context objects based on embedding distance;
• train, via at least one processor, the embedding neural network using training examples in the training set, wherein the embedding neural network is trained using a selected training example by:
• (a) generating a context feature vector for the context object associated with the selected training example and providing the context feature vector for the input layer, (b) generating a target feature vector for the positive tar-get sample associated with the selected training example and providing the target feature vector for the output layer, (c) adjusting weights of the embedding neural network using a machine learning technique;
• and
• generate, via at least one processor, a datastructure that stores the adjusted weights of the embedding neural network.

A processor-implemented neighbor discovery search engine method, comprising:

• executing processor-implemented embedding neural network training component instructions to:
• obtain, via at least one processor, an embedding neural network training request associated with a set of context objects;
• determine, via at least one processor, sample similarity evaluation metrics associated with the embedding neural network training request;
• determine, via at least one processor, for each context object in the set of context objects, a set of positive target samples that satisfy the sample similarity evaluation metrics for the respective context object;
• add, via at least one processor, to a training set, for each context object and each positive target sample in the respective set of positive target samples, a training example that comprises:
• (a) the respective context object and (b) a positive target sample;
• determine, via at least one processor, configuration parameters for an embedding neural network, wherein the embedding neural network comprises an input layer, a hidden layer, an embedding layer and an output layer, wherein the embedding layer facili-tates calculating embeddings that facilitate comparison of context objects based on embedding distance;
• train, via at least one processor, the embedding neural network using training examples in the training set, wherein the embedding neural network is trained using a selected training example by:
• (a) generating a context feature vector for the context object associated with the selected training example and providing the context feature vector for the input layer, (b) generating a target feature vector for the positive tar-get sample associated with the selected training example and providing the target feature vector for the output layer, (c) adjusting weights of the embedding neural network using a machine learning technique;
• and
• generate, via at least one processor, a datastructure that stores the adjusted weights of the embedding neural network.