Causal Embedding
Causal embedding is the process of mapping outcome causes/symptoms to vectors of real numbers.
The illustration below depicts how causal embedding can be used in model training and prediction processing.
Causal Vector
The causal vector contains an element for each observed cause. Each vector element represents a linear equation dimension for the cause in the result/cause model multidimensional space.
Embedding Vectors Creation
Embedding vectors are created by assembling individual elements that represent linear equation dimensions in the result/cause model multidimensional space.
Observed Causes
Observed causes are measured against a specific scale, such as 0-100 or .00-1.00. Examples are medical causes/symptoms such as chest pain and headache.
Predicted Result
A predicted result is the output of the result/cause model prediction processing. An example predicted result in medical diagnosis is influenza.
Result/Cause Associations
Result/cause associations are the pairing of a result with a potential cause or symptom. An example is headache paired with influenza.
Result/Cause Embedding Vectors
Causal embedding vectors are created by assembling individual elements that represent linear equation dimensions in the result/cause model multidimensional space. they are paired with their corresponding result scalar values for use in model training.
Result/Cause Model
A result/cause model is trained and used to predict results. There are a number of model possibilities for use with causal embedding, such as Artificial Neural Networks, Probabilistic Graphical Models, and Support Vector Machines.
Result/Cause Model Training
Result/cause model is trained using causal embedding vectors and result scalar values.
Result/Cause Model Prediction Processing
A causal embedding vector us used as input to prediction processing and a predicted result scalar is the output.
Examples
An Embedding-Based Approach for Oral Disease Diagnosis Prediction from Electronic Medical Records
Disentangling User Interest and Popularity Bias for Recommendation with Causal Embedding
Medical knowledge embedding based on recursive neural network for multi-disease diagnosis