MATRIX-VECTOR ALGORITHMS FOR NORMALIZING FACTORS IN ALGEBRAIC BAYESIAN NETWORKS LOCAL POSTERIORI INFERENCE
Annotation
We consider a task of local posteriori inference description by means of matrix-vector equations in algebraical Bayesian networks that represent a class of probabilistic graphical models. Such equations were generally presented in previous publications, however containing normalizing factors that were provided with algorithmic descriptions of their calculations instead of the desired matrix-vector interpretation. To eliminate this gap, the normalized factors were firstly represented as scalar products. Then, it was successfully shown that one of the components in each scalar product can be expressed as a Kronecker degree of a constant two-dimensional vector. Later on, non-normalized posteriori inference matrixoperator transplantation and further transfer within each scalar product yielded a representation of one of the scalar product components as a sequence of tensor products of two-dimensional vectors. The latter vectors have only two possible values in one case and three values in the other. The choice among those values is determined by the structure of input evidence. The second component of each scalar products is the vector with original data. The calculations performed gave the possibility for constructing corresponding vectors; the paper contains a table with proper examples for some of them. Local posteriori inference representation for matrix-vector equations simplify the development of local posteriori inference algorithms, their verification and further implementation based on available libraries. These equations also give the possibility for application of classical mathematical techniques to the obtained results analysis. Finally, the results obtained make it possible to apply the method of postponed calculations. This method helps avoiding construction of big-size vectors; instead, the vectors components can be calculated just in time they are needed by means of bitwise operations.
Keywords
Постоянный URL
Articles in current issue
- OPTICAL PULLING FORCES IN “NANOPARTICLES DIMER IN THE STRUCTURED FIELD” SYSTEM
- ADVANTAGES OF DIFFRACTIVE OPTICAL ELEMENTS APPLICATION IN SIMPLE OPTICAL IMAGING SYSTEMS
- COMPARISON OF HOLOGRAPHIC AND ITERATIVE METHODS FOR AMPLITUDE OBJECT RECONSTRUCTION
- STARS IDENTIFICATION AT THE ASTRONOMICAL COORDINATES DETERMINATION BY MEANS OF AN AUTOMATED ZENITH TELESCOPE
- TRANSFORMATION ALGORITHM FOR IMAGES OBTAINED BY OMNIDIRECTIONAL CAMERAS
- ADAPTIVE FLUX OBSERVER FOR PERMANENT MAGNET SYNCHRONOUS MOTORS
- STUDY OF MECHANISMS RESPONSIBLE FOR THE EFFICIENCY DEGRADATION OF THE III-NITRIDES LIGHT EMITTING DIODES
- FORMATION OF LUMINESCENT OPTICAL WAVEGUIDES IN SILICATE GLASS MATRIX BY THE ION-EXCHANGE TECHNIQUE
- INVESTIGATION OF HETEROSTRUCTURES 3C-SIC/15R-SIC
- INFLUENCE OF THE ORTHOGONALLY POLARIZED BACK REFLECTIONS ON THE POWER AND RADIATION SPECTRUM OF SUPERLUMINESCENT DIODES
- SPEAKERS' IDENTIFICATION METHOD BASED ON COMPARISON OF PHONEME LENGTHS STATISTICS
- STUDY OF BLOCKING EFFECT ELIMINATION METHODS BY MEANS OF INTRAFRAME VIDEO SEQUENCE INTERPOLATION
- ALGORITHM OF RATIONAL PROCESSOR ARCHITECTURE
- RUNTIME BALANCING EFFECT IN DISTRIBUTED SIMULATION MODEL
- STUDY OF SOLUTION REPRESENTATION LANGUAGE INFLUENCE ON EFFICIENCY OF INTEGER SEQUENCES PREDICTION
- NUCLEAR-MAGNETIC MINI-RELAXOMETER FOR LIQUID AND VISCOUS MEDIA CONTROL
- COMPARISON OF VARIOUS APPROACHES TO MULTI-CHANNEL
INFORMATION FUSION IN C-OTDR SYSTEMS FOR REMOTE MONITORING OF EXTENDED OBJECTS - COORDINATION IN MULTILEVEL NETWORK-CENTRIC CONTROL SYSTEMS OF REGIONAL SECURITY: APPROACH AND FORMAL MODEL
- ANALYSIS OF FINITE-DIFFERENCE SCHEMES BASED ON EXACT AND APPROXIMATE SOLUTION OF RIEMANN PROBLEM
- APPLICATION OF MODIFIED CONVERSION METHOD TO A NONLINEAR DYNAMICAL SYSTEM
- SERVICES OF FULL-TEXT SEARCHING IN A DISTRIBUTED INFORMATION ENVIRONMENT (PROJECT HUMANITARIANA)
- ANALYSIS OF THE RELATIONSHIP BETWEEN DEGREE OF BLOOD OXYGENATION AND BACKSCATTERED RADIATION WITH THE USE OF NUMERICAL MODELING
- USING PRECEDENTS FOR REDUCTION OF DECISION TREE BY GRAPH SEARCH
- TEXTS SENTIMENT-ANALYSIS APPLICATION FOR PUBLIC OPINION ASSESSMENT