Theofrastos Mantadelis

Dedicated tabling for a probabilistic setting

ProbLog is a probabilistic framework that extends Prolog with probabilistic facts. To compute the probability of a query, the complete SLD proof tree of the query is collected as a sum of products. ProbLog applies advanced techniques to make this feasible and to assess the correct probability. Tabling is a well-known technique to avoid repeated subcomputations and to terminate loops. We investigate how tabling can be used in ProbLog. The challenge is that we have to reconcile tabling with the advanced ProbLog techniques. While standard tabling collects only the answers for the calls, we do need the SLD proof tree. Finally we discuss how to deal with loops in our probabilistic framework. By avoiding repeated subcomputations, our tabling approach not only improves the execution time of ProbLog programs, but also decreases accordingly the memory consumption. We obtain promising results for ProbLog programs using exact probability inference.

Analyzing the efficiency of context-based grouping on collaboration in VANETs with large-scale simulation

Vehicle-to-vehicle and vehicle-to-infrastructure communication systems enable vehicles to share information captured by their local sensors with other interested vehicles. To ensure that this information is delivered at the right time and location, context-aware routing is vital for intelligent inter-vehicular communication. Traditional network addressing and routing schemes do not scale well for large vehicular networks. The conventional network multicasting and broadcasting cause significant overhead due to a large amount of irrelevant and redundant transmissions. To address these challenges, we first take into account contextual properties such as location, direction, and information interest to reduce the network traffic overhead. Second, to improve the relevancy of the received information we leverage the mobility patterns of vehicles and the road layouts to further optimize the peer-to-peer routing of the information. Third, to ensure our approach is scalable, we propose a context-based grouping mechanism in which relevant information is shared in an intelligent way within and between the groups. We evaluate our approach based on groups with common spatio-temporal characteristics. Our simulation experiments show that our context-based routing scheme and grouping mechanism significantly reduces the propagation of irrelevant and redundant information.

ProbLog Technology for Inference in a Probabilistic First Order Logic

We introduce First Order ProbLog, an extension of first order logic with soft constraints where formulas are guarded by probabilistic facts. The paper defines a semantics for FOProbLog, develops a translation into ProbLog, a system that allows a user to compute the probability of a query in a similar setting restricted to Horn clauses, and reports on initial experience with inference.

Preprocessing boolean formulae for BDDs in a probabilistic context

Inference in many probabilistic logic systems is based on representing the proofs of a query as a DNF Boolean formula. Assessing the probability of such a formula is known as a #P-hard task. In practice, a large DNF is given to a BDD software package to construct the corresponding BDD. The DNF has to be transformed into the input format of the package. This is the preprocessing step. In this paper we investigate and compare different preprocessing methods, including our new trie based approach. Our experiments within the ProbLog system show that the behaviour of the methods changes according to the amount of sharing in the original DNF. The decomposition method is preferred when there is not much sharing in the DNF, whereas DNFs with sharing benefit from our trie based method. While our methods are motivated and applied in the ProbLog context, our results are interesting for other applications that manipulate DNF Boolean formulae.

Variable compression in ProbLog

In order to compute the probability of a query, ProbLog represents the proofs of the query as disjunctions of conjunctions, for which a Reduced Ordered Binary Decision Diagram (ROBDD) is computed. The paper identifies patterns of Boolean variables that occur in Boolean formulae, namely AND-clusters and OR-clusters. Our method compresses the variables in these clusters and thus reduces the size of ROBDDs without affecting the probability. We give a polynomial algorithm that detects AND-clusters in disjunctive normal form (DNF) Boolean formulae, or OR-clusters in conjunctive normal form (CNF) Boolean formulae. We do an experimental evaluation of the effects of AND-cluster compression for a real application of ProbLog. With our prototype implementation we have a significant improvement in performance (up to 87%) for the generation of ROBDDs. Moreover, compressing AND-clusters of Boolean variables in the DNFs makes it feasible to deal with ProbLog queries that give rise to larger DNFs.

SkILL - A Stochastic Inductive Logic Learner

Probabilistic Inductive Logic Programming (PILP) is a relatively unexplored area of Statistical Relational Learning which extends classic Inductive Logic Programming (ILP). Within this scope, we introduce SkILL, a Stochastic Inductive Logic Learner, which takes probabilistic annotated data and produces First Order Logic (FOL) theories. Data in several domains such as medicine and bioinformatics have an inherent degree of uncertainty, and because SkILL can handle this type of data, the models produced for these areas are closer to reality. SkILL can then use probabilistic data to extract non-trivial knowledge from databases, and also address efficiency issues by introducing an efficient search strategy for finding hypotheses in PILP environments. SkILLs capabilities are demonstrated using a real world medical dataset in the breast cancer domain.

Compacting Boolean Formulae for Inference in Probabilistic Logic Programming

Knowledge compilation converts Boolean formulae for which some inference tasks are computationally expensive into a representation where the same tasks are tractable. ProbLog is a state-of-the-art Probabilistic Logic Programming system that uses knowledge compilation to reduce the expensive probabilistic inference to an efficient weighted model counting. Motivated to improve ProbLog’s performance we present an approach that optimizes Boolean formulae in order to speed-up knowledge compilation. We identify 7 Boolean subformulae patterns that can be used to re-write Boolean formulae. We implemented an algorithm with polynomial complexity which detects and compacts 6 of these patterns. We employ our method in the inference pipeline of ProbLog and conduct extensive experiments. We show that our compaction method improves knowledge compilation and consecutively the overall inference performance. Furthermore, using compaction reduces the number of time-outs, allowing us to solve previously unsolvable problems.

Analysing a publish/subscribe system for mobile ad hoc networks with ProbLog

Fadip is a Publish/Subscribe system for Mobile Ad hoc Networks which uses probabilistic routing of messages to deal with the volatile nature of the network. It uses controlled propagation of publications and subscriptions, with the fading gossip technique to reduce the number of broadcasts. We present a probabilistic logic program in ProbLog that models Fadip. This allows us to calculate the probabilities that messages are successfully received by subscribers and to analyse the performance of the Fadip system.

On modifying the temporal modeling of HSMMs for pediatric heart sound segmentation

Heart sounds are difficult to interpret because a) they are composed by several different sounds, all contained in very tight time windows; b) they vary from physiognomy even if the show similar characteristics; c) human ears are not naturally trained to recognize heart sounds. Computer assisted decision systems may help but they require robust signal processing algorithms. In this paper, we use a real life dataset in order to compare the performance of a hidden Markov model and several hidden semi Markov models that used the Poisson, Gaussian, Gamma distributions, as well as a non-parametric probability mass function to model the sojourn time. Using a subject dependent approach, a model that uses the Poisson distribution as an approximation for the sojourn time is shown to outperform all other models. This model was able to recreate the “true” state sequence with a positive predictability per state of 96%. Finally, we used a conditional distribution in order to compute the confidence of our classifications. By using the proposed confidence metric, we were able to identify wrong classifications and boost our system (in average) from an ≈ 83% up to ≈90% of positive predictability per sample.

Using Iterative Deepening for Probabilistic Logic Inference

We present a novel approach that uses an iterative deepening algorithm in order to perform probabilistic logic inference for ProbLog, a probabilistic extension of Prolog. The most used inference method for ProbLog is exact inference combined with tabling. Tabled exact inference first collects a set of SLG derivations which contain the probabilistic structure of the ProbLog program including the cycles. At a second step, inference requires handling these cycles in order to create a non-cyclic Boolean representation of the probabilistic information. Finally, the Boolean representation is compiled to a data structure where inference can be performed in linear time. Previous work has illustrated that there are two limiting factors for ProbLog’s exact inference. The first factor is the target compilation language and the second factor is the handling of the cycles. In this paper, we address the second factor by presenting an iterative deepening algorithm which handles cycles and produces solutions to problems that previously ProbLog was not able to solve. Our experimental results show that our iterative deepening approach gets approximate bounded values in almost all cases and in most cases we are able to get the exact result for the same or one lower scaling factor.