method of interagent communication in probabilistic agents implementing factor graph document databases
A method of interagent communication in probabilistic agents using factor graph document databases wherein multiple agents reach a consensus on the probabilistic structure of the environment in which they perform their tasks using factor graph document databases implemented by said agents, and updating a conditional probability matrix for each agent using an update method, and sharing the updates using a sharing method involving matrix multiplication of the conditional probability matrices.
1 . A method of interagent communication in probabilistic agents, comprising,
defining factor graph document databases for two or more probabilistic agents encoding one or more conditional probabilities, wherein the conditional probabilities are encoded in conditional probability vectors, matrices or tensors,
increasing counts in one or more cells of the vectors, matrices or tensors of the factor graph document database of one or more of the probabilistic agents when said one or more probabilistic agents receives new information,
renormalizing the conditional probability distributions in one or more of the factor graph document databases of the probabilistic agents,
combining the updated conditional probability vectors, matrices, or tensors of all of the factor graph document databases of the probabilistic agents by means of vector, matrix, or tensor multiplication operations,
replacing the conditional probability vectors, matrices, or tensors of each probabilistic agent with the new vectors, matrices, or tensors resulting from the multiplication operations.
2 . A method of claim 1 , wherein the new information corresponds to the observation of the co-occurrence of events mapped by the dimensions of a vector, matrix, or tensor of the factor graph document database by means of inputs coming from one or more physical or virtual sensors of one or more probabilistic agents.
3 . A method of claim 1 , wherein the renormalization of a probability distribution is achieved by summing over all the values of the cells of the vector, matrix or tensor forming the probability distribution, and by dividing each cell value by the sum of all the cells, or by an equivalent renormalization method.