Introduction (WIP)

Cyberium is a platform and collection of tools for making structured knowledge graphs from any type of document for any kind of purpose, but focused on "actor", "asset" and "action" type of entities. We are focused on machine execution through knowledge graphs and that is why every other type of knowledge elements are abstract from this point of view.

Document ==> Graph(Actor, Asset, Action)

The document parser, a trained and pluggable PoS + NER language model, identifies all the actors, assets and actions as well as all the relationships among them.

Applications

Knowledge graphs(K-graph in short from hereon) are primarily used for organizing data from multiple sources, and capture information about entities of interest. Although we go beyond that for applying K-graphs in various scenarios of content generation, content control and content exploration.

Applications in Content Generation

A fully connected K-graph provides complete information about the nature of a data element.

Derived graph for A => D(A) =  δG/δA 
      ; where A can be any element of graph. In this context, it is either actor, asset or action.

A D(A) based prompt can be an effective way to create a better prompt for prompt generation. As such a good prompt for generative AI system should have all the blocks and channels for a holistic content generation. Therefore a derived graph D(A) can be used to make a completely good prompt. Using this mechanism, the prompt creation can be automated as well.

Here the prompt can be used for any type of generative tasks like text generation, or code generation, or image generation, or any other type of generation objective.

Applications in Content Control

Another significant use of K-graph is in content control for generative tasks of artificial intelligence. Since AI models learn by masking or predicting the tokens in vector space, they learn to extrapolate information for generative tasks. This leads to both greatly good or greatly bad content generation. When applied to such cases, K-graph acts as a guardrail for controlled content generation. It can differentiate between good generation and bad generation. Although the disparity in the both the cases from ground truth will be equivalent, but a bad generation

Example:

Prompt: 

Output:  

Exception:

Remedy:

Applications in Content Exploration

There are always things we don't know. But you don't know what you don't know.

Moving mountains with K-graphs

Somethings which are practically impossible is becoming possible with K-graphs.

Applications by industry