Getting Started

Oracle Engine 1.0

ORACLE ENGINE is a scheduling framework and service for machine learning models referred to as nodes. Nodes represent the machine learning models and their relationships with each other (or the model's local sub network).

These relationships are depicted as weights within a subnet.

There are two main types of nodes

• alpha nodes

• beta nodes.

Here are the main rules

• An oracle engine has only one alpha node

• Beta nodes can be pre-trained models, custom models, or a set of instructions within a function. Pre-trained models are referred to as bagels. Openai’s general pre-trained transformer is an example of a bagel that can be used in oracle’s engine

• Weights can be mapped to unique data sources as long as these data sources are from the original training and testing dataset

Example

A patient has just completed their health assessment on Penrose Care. A set of questions in this HA were designed to get a better sense of the patients current mental state. These questions will be sent to pre-trained transformer called Maeva and it will be asked to perform semantic analysis on the patients response. Ideally, Maeva's response will exist on a scale from -1 to 1 where 1 means very positive (or stress free) and -1 means very stressful. However, Maeva's response is only as good as the data it was trained on. Penrose Care is aware of this risk and decides to improve the probability of providing a better response by seeking answers from other well trained models in an oracle engine that also includes Maeva. The question then becomes, how do we know which "nodes" response to use? One way to address this issue is to take an average of all the responses provided by each node.

$f(x) =(B_0x_0+ B_1x_1+B_2x_2+B_3x_3.....+B_nx_n)/(n-1)$

where

• $f(x)$ represents the final output of the alpha node

• $x_k$represents weights

• $B_k$represents nodes

Engine Architechture Designs

Example 1.0: Oracle sample with Open AI gpt model, and oracle custom models