Probabilistic Logic Networks

Patterns of Cognition: Cognitive Algorithms as Galois Connections Fulfilled by Chronomorphisms On Probabilistically Typed Metagraphs

“…Forward chaining inference in Probabilistic Logic Networks [GIGH08] or other similar frameworks may be based on a quality metric measuring the interestingness or surprisingness of an inference conclusion (see [Goe21b] for a formalization of surprisingness in an OpenCog context). The general process involved is then, qualitatively:…”

“…A more efficient alternative to trying to explicitly solve the stochastic dynamic programming functional equation, in this case, is to build a large model of which series of choices ( , , ) seem to be effective in which contexts. This is what PLN "adaptive inference control" aims to do [GIGH08]. Doing explicit stochastic dynamic programming but using the (2ai) "inference based inference control" option is one way of implementing this, as each instance of the underlying inference can make use of knowledge generated and saved when applying similar inference elsewhere in the dynamic programming process.…”

“…To quantify the amount of reward achieved in expanding the BID and thus moving from a prior truth value estimate ( 1 , 1 ) to a new truth value estimate ( 2 , 2 ) for the root statement, we introduce the CWI (Confidence Weighted Information Gain) measure. Referring to the fundamental theory of STVs as outlined in [GIGH08], let ( , ) denote the second-order distribution corresponding to the pair ( , ). We then define the CWIG as the information gain in the distribution ( 2 , 2 ) conditioned on ( 1 , 1 ) plus background knowledge; we can also make similar calculations using graphtropy, yielding a Confidence Weighted Graphtropy Gain (CWGG).…”

“…In the latter case, multiple different algorithmic approaches are assumed to have foundational value for handling different sorts of cognitive problem, and a cognitive architecture is introduced that has capability of effectively integrating these different approaches, generally including some sort of common knowledge meta-representation and some cross-algorithm methodology for controlling resource allocation. The OpenCog AGI architecture pursued by the author and colleagues [GPG13a] [GPG13b] is an example of the latter, with key component algorithms being probabilistic logical inference [GIGH08], evolutionary program learning [Loo06], pattern mining, concept formation and nonlinear-dynamical attention allocation. In recent years the hybridization of OpenCog with neural-net learning mechanisms has been explored and in some cases practically utilized as well.…”

Info-Evo: Using Information Geometry to Guide Evolutionary Program Learning

“…The new guesses are evaluated and inserted into the population; unless a sufficiently good answer has been found or the allotted time has run out, we return to Step 1 and keep going This high-level strategy could be instantiated in many different ways; e.g. in [GPG13] it has been proposed to hybridize the MOSES program learning algorithm [Loo06] and the Probabilistic Logic Networks inference system [GIGH08] in such a fashion. Here we propose a particular instantiation called Info-Evo, in which the role of the probabilistic inference algorithm is filled by information-geometric ("natural gradient") search on the space of "promise landscapes" -a particular space of probability distributions over genotype space.…”