Probabilistic Reasoning Research Articles

Learning hierarchical probabilistic logic programs

Probabilistic logic programming (PLP) combines logic programs and probabilities. Due to its expressiveness and simplicity, it has been considered as a powerful tool for learning and reasoning in relational domains characterized by uncertainty. Still, learning the parameter and the structure of general PLP is computationally expensive due to the inference cost. We have recently proposed a restriction of the general PLP language called hierarchical PLP (HPLP) in which clauses and predicates are hierarchically organized. HPLPs can be converted into arithmetic circuits or deep neural networks and inference is much cheaper than for general PLP. In this paper we present algorithms for learning both the parameters and the structure of HPLPs from data. We first present an algorithm, called parameter learning for hierarchical probabilistic logic programs (PHIL) which performs parameter estimation of HPLPs using gradient descent and expectation maximization. We also propose structure learning of hierarchical probabilistic logic programming (SLEAHP), that learns both the structure and the parameters of HPLPs from data. Experiments were performed comparing PHIL and SLEAHP with PLP and Markov Logic Networks state-of-the art systems for parameter and structure learning respectively. PHIL was compared with EMBLEM, ProbLog2 and Tuffy and SLEAHP with SLIPCOVER, PROBFOIL+, MLB-BC, MLN-BT and RDN-B. The experiments on five well known datasets show that our algorithms achieve similar and often better accuracies but in a shorter time.

A New Marine Disaster Assessment Model Combining Bayesian Network with Information Diffusion

There are two challenges in the comprehensive marine hazard assessment. The influencing mechanism of marine disaster is uncertain and disaster data are sparse. Aiming at the uncertain knowledge and small sample in assessment modeling, we combine the information diffusion algorithm and Bayesian network to propose a novel assessment model. The information diffusion algorithm is adopted to expand associated samples between disaster losses and environmental conditions. Then the expanded data sets are used to build the BN-based assessment model through structural learning, parameter learning and probabilistic reasoning. The proposed model is applied to the hazard assessment of marine disasters in Shanghai. Experimental comparison results show that it is capable of dealing with uncertainty effectively and achieving more accuracy risk assessment under the small sample condition.

I did not expect to be dreaming: Explaining realization in lucid dreams with a Bayesian framework

The commonsense view is that a lucid dream starts when the dreamer realizes that they are currently dreaming. The notion of realization, however, has been accepted at face value, with little consideration of whether the dreamer realizes that they are dreaming in the sense of actual reasoning, or if it is a mere epiphenomenon of lucid dream initiation. This article offers a solution to this problem by, first, arguing that the transition to lucidity can occur as a result of successful reasoning, and second, building a model of this reasoning in terms of probabilistic reasoning. The established Bayesian model explains realization in lucid dreams taking under consideration two factors: the beliefs that the dreamer holds on what is generally probable and improbable, and the dreamer’s admissibility of being in a dream. Defended against important objections, the model offers an explanation of lucid dream initiation, relevant for future research on dreaming.

Cognitive heterogeneity and complex belief elicitation.

The Stochastic Becker-DeGroot-Marschak (SBDM) mechanism is a theoretically elegant way of eliciting incentive-compatible beliefs under a variety of risk preferences. However, the mechanism is complex and there is concern that some participants may misunderstand its incentive properties. We use a two-part design to evaluate the relationship between participants’ probabilistic reasoning skills, task complexity, and belief elicitation. We first identify participants whose decision-making is consistent and inconsistent with probabilistic reasoning using a task in which non-Bayesian modes of decision-making lead to violations of stochastic dominance. We then elicit participants’ beliefs in both easy and hard decision problems. Relative to Introspection, there is less variation in belief errors between easy and hard problems in the SBDM mechanism. However, there is a greater difference in belief errors between consistent and inconsistent participants. These results suggest that while the SBDM mechanism encourages individuals to think more carefully about beliefs, it is more sensitive to heterogeneity in probabilistic reasoning. In a follow-up experiment, we also identify participants with high and low fluid intelligence with a Raven task, and high and low proclivities for cognitive effort using an extended Cognitive Reflection Test. Although performance on these tasks strongly predict errors in both the SBDM mechanism and Introspection, there is no significant interaction effect between the elicitation mechanism and either ability or effort. Our results suggest that mechanism complexity is an important consideration when using elicitation mechanisms, and that participants’ probabilistic reasoning is an important consideration when interpreting elicited beliefs.Electronic supplementary materialThe online version of this article (10.1007/s10683-021-09722-x) contains supplementary material, which is available to authorized users.

A Review on the Development of Fuzzy Classifiers with Improved Interpretability and Accuracy Parameters

This review paper of fuzzy classifiers with improved interpretability and accuracy parameter discussed the most fundamental aspect of very effective and powerful tools in form of probabilistic reasoning, The fuzzy logic concept allows the effective realization of ap-proximate, vague, uncertain, dynamic, and more realistic conditions, which is closer to the actual physical world and human thinking. The fuzzy theory has the competency to catch the lack of preciseness of linguistic terms in a speech of natural language. The fuzzy theory provides a more significant competency to model humans like com-mon-sense reasoning and conclusion making to fuzzy set and rules as good membership functions. Also, in this paper reviews discussed the evaluation of the fuzzy set, type-1, type-2, and interval type-2 fuzzy system from traditional Boolean crisp set logic along with interpretability and accuracy issues in the fuzzy system.

Indirect illusory inferences from disjunction: a new bridge between deductive inference and representativeness

We provide a new link between deductive and probabilistic reasoning fallacies. Illusory inferences from disjunction are a broad class of deductive fallacies traditionally explained by recourse to a matching procedure that looks for content overlap between premises. In two behavioral experiments, we show that this phenomenon is instead sensitive to real-world causal dependencies and not to exact content overlap. A group of participants rated the strength of the causal dependence between pairs of sentences. This measure is a near perfect predictor of fallacious reasoning by an independent group of participants in illusory inference tasks with the same materials. In light of these results, we argue that all extant accounts of these deductive fallacies require non-trivial adjustments. Crucially, these novel indirect illusory inferences from disjunction bear a structural similarity to seemingly unrelated probabilistic reasoning problems, in particular the conjunction fallacy from the heuristics and biases literature. This structural connection was entirely obscure in previous work on these deductive problems, due to the theoretical and empirical focus on content overlap. We argue that this structural parallelism provides arguments against the need for rich descriptions and individuating information in the conjunction fallacy, and we outline a unified theory of deductive illusory inferences from disjunction and the conjunction fallacy, in terms of Bayesian confirmation theory.

Individual differences in epistemically suspect beliefs: the role of analytic thinking and susceptibility to cognitive biases

The endorsement of epistemically suspect (i.e., paranormal, conspiracy, and pseudoscientific) beliefs is widespread and has negative consequences. Therefore, it is important to understand the reasoning processes – such as lower analytic thinking and susceptibility to cognitive biases – that might lead to the adoption of such beliefs. In two studies, I constructed and tested a novel questionnaire on epistemically suspect beliefs (Study 1, N = 263), and used it to examine probabilistic reasoning biases and belief bias in syllogistic reasoning as predictors of the endorsement of those beliefs, while accounting for analytic thinking and worldview variables (Study 2, N = 397). Probabilistic reasoning biases, analytic thinking, religious faith, and political liberalism consistently predicted various epistemically suspect beliefs, whereas the effect of syllogistic belief bias was largely restricted to pseudoscientific beliefs. Further research will be needed to examine the role the biased evaluation of evidence plays in the endorsement of epistemically suspect beliefs.

The Results Presentation of The Test Plants As A Natural Indicator of The Test Compounds of The Acid-Base Balance In the Teaching Materials of The Atlas to Train Scientific Reasoning Students

The type of research applied in this research is Research and Development. The development model used is Dick and Carrey. The purpose of this research and development is to design an Atlas and apply it to train the students' scientific reasoning abilities. Atlas is a teaching materials developed and applied to high school / vocational / MA level students. The sample chosen for the implementation of the Atlas of teaching materials was grade X of VOCATIONAL Health Yannas Husada Bangkalan students. The application of plants as natural indicators of acid-base tests is the material presented in the developed Atlas. Plants selected as indicators are tested first to be presented in the Atlas. Atlas is a teaching material that can convey information related to the use of natural indicators of the acid-base test to train the students' scientific reasoning skills. The four indicators of scientific reasoning that are trained are Theoretical Reasoning, Proportional Reasoning, Probabilistic Reasoning, and Correlational Reasoning. The data shows the students' scientific reasoning ability on the post-test results, as many as 20% of students are on the high criteria, 60% are moderate criteria, and 20% are still on the low criteria from the post-test results. The post-test result data is quite increased when compared to the pretet results, which 100% of students are in the low criteria.

(Invited) Binary Stochastic Neurons and Compound Synapses from Nano-Magnets

In this presentation I will introduce the nCORE center CAPSL’s activities towards a probabilistic spin logic for Boolean and Non-Boolean computing. While magnetic materials in storage devices are part of our electronic world since many decades and are explored more recently by the semiconductor industry for memory applications such as Spin-transfer torque magnetic random-access memory (STT-RAM), spin based devices for logic applications are still in their infancies. Mimicking a CMOS type logic with magnets requires a “Write” and a “Read” unit as well as proper input/output isolation. Based on this, a device concept in which i) charge-to-spin conversion occurs through the Giant Spin Hall Effect (GSHE) [1], ii) dipolar coupling mediates the information transfer from a magnet in the Write unit to another one in the Read unit, and iii) a magnetic tunnel junction (MTJ) is used as the Read unit was proposed in 2012 under the name charge spin logic (CSL) [2]. Next, I will explain in how far the use of low-barrier magnets cannot only help to reduce power consumption of CSL devices but introduces a conceptually new way of computing through the introduction of “tunable randomness”. This idea, that is also known as probabilistic spin logic (PSL), was introduced in 2017 [3] and uses a counterintuitive aspect of magnets, namely their capability of creating a truly random magnetic signal if low-barrier circular nano-magnets are used. In particular, I will connect PSL with activities in the realm of quantum computing as well as neural networks.[1] L. Liu, C.-F. Pai, Y. Li, H. W. Tseng, D. C. Ralph, and R. a. Buhrman, Science 336, 555 (2012).[2] S. Datta, S. Salahuddin, and B. Behin-Aein, Applied Physics Letters 101, 252411 (2012).[3] K. Y. Camsari, R. Faria, B. Sutton and S. Datta. Physical Review X 7, 031014 (2017).

Decision Support for Integrated Management of Local-Level Adaptation to Climate Changes: The Case of Serbia

Projected climate changes will additionally increase the already significant risk of natural hazard-related disasters in Serbia and the west Balkan region as a whole. Serbia is about to introduce the strategy for climate change adaptation and mitigation. However, a national decision-support system for implementation of the climate change law and strategy is yet to be developed. This study contributes to the implementation of adaptation policies at subnational levels by development of a decision-support model for local-level management of the climate change adaptation process. The study explores the potential for synergetic application of multicriteria decision making analysis and probabilistic reasoning methods by focusing on Bayesian networks, analytical hierarchy processes, and geographic information systems for selection of priority adaptation measures. The study was based on the formation of causal chains, which enable linking management decisions and socioeconomic or biophysical consequences into articulated sequences of conditional relationships. A model was tested in the forestry sector, and it clearly pointed out development of an early warning system and planning of water intake basins as priority adaptation measures. Since the results are shown as a probability distribution for each alternative solutions, the model can assist decision makers with prompt evaluation of various scenarios.

A Hierarchical Approach to Active Semantic Mapping Using Probabilistic Logic and Information Reward POMDPs

Maintaining a semantic map of a complex and dynamic environment, where the uncertainty originates in both noisy perception and unexpected changes, is a challenging problem. In particular, we focus on the problem of maintaining a semantic map of an environment by a mobile agent. In this paper we address this problem in an hierarchical fashion. Firstly, we employ a probabilistic logic model representing the semantic map, as well as the associated uncertainty. Secondly, we model the interaction of the robot with the environment with a set of information-reward POMDP models, one for each partition of the environment (e.g., a room). The partition is performed in order to address the scalability limitations of POMDP models over very large state spaces. We then use probabilistic inference to determine which POMDP and policy to execute next. Experimental results show the efficiency of this architecture in real domestic service robotic scenarios.

An Explorative Application of Random Forest Algorithm for Archaeological Predictive Modeling. A Swiss Case Study

The present work proposes an innovative approach to surveys and demonstrates the effectiveness of bringing together traditional archaeological questions, such as the exploration and the analysis of settlement patterns, with the most innovative technologies related to Machine Learning. Namely, we applied Random Forest, an ensemble learning method based on decision trees, to perform archaeological predictive modeling (APM) for the Canton of Zurich, in Switzerland. This was done based on a dataset of known archaeological sites dating back to the Roman Age. The APM represents an automated decision-making and probabilistic reasoning tool that is relevant for archaeological risk assessment and cultural heritage management. Machine learning-based approaches can learn from data and make predictions, starting from the acquired knowledge, through the modeling of the hidden relationships between a set of observations, representing the dependent variable (i.e. the archeological sites), and the independent variables (i.e. the geo-environmental features prone to influence the site locations). The main objective of the present study is to assess the spatial probability of presence for Roman settlements within the study area. As results, we produced: 1) a probability map, expressing the likelihood of finding a Roman site at different locations; 2) the importance ranking of the geo-environmental features influencing the presence of the archeological sites. These outputs in our results are of paramount importance, not only in verifying the reliability of the data, but also in stimulating experts in different ways. Also, these results help evaluate the benefits and constraints of using such innovative techniques and, ultimately, help explore the performance of machine learning-based models in processing archaeological information.

Self-Supervised Self-Supervision by Combining Deep Learning and Probabilistic Logic

Labeling training examples at scale is a perennial challenge in machine learning. Self-supervision methods compensate for the lack of direct supervision by leveraging prior knowledge to automatically generate noisy labeled examples. Deep probabilistic logic (DPL) is a unifying framework for self-supervised learning that represents unknown labels as latent variables and incorporates diverse self-supervision using probabilistic logic to train a deep neural network end-to-end using variational EM. While DPL is successful at combining pre-specified self-supervision, manually crafting self-supervision to attain high accuracy may still be tedious and challenging. In this paper, we propose Self-Supervised Self-Supervision (S4), which adds to DPL the capability to learn new self-supervision automatically. Starting from an initial "seed," S4 iteratively uses the deep neural network to propose new self-supervision. These are either added directly (a form of structured self-training) or verified by a human expert (as in feature-based active learning). Experiments show that S4 is able to automatically propose accurate self-supervision and can often nearly match the accuracy of supervised methods with a tiny fraction of the human effort.

Juice: A Julia Package for Logic and Probabilistic Circuits

Juice is an open-source Julia package providing tools for logic and probabilistic reasoning and learning based on logic circuits (LCs) and probabilistic circuits (PCs). It provides a range of efficient algorithms for probabilistic inference queries, such as computing marginal probabilities (MAR), as well as many more advanced queries. Certain structural circuit properties are needed to achieve this tractability, which Juice helps validate. Additionally, it supports several parameter and structure learning algorithms proposed in the recent literature. By leveraging parallelism (on both CPU and GPU), Juice provides a fast implementation of circuit-based algorithms, which makes it suitable for tackling large-scale datasets and models.

Progression Heuristics for Planning with Probabilistic LTL Constraints

Probabilistic planning subject to multi-objective probabilistic temporal logic (PLTL) constraints models the problem of computing safe and robust behaviours for agents in stochastic environments. We present novel admissible heuristics to guide the search for cost-optimal policies for these problems. These heuristics project and decompose LTL formulae obtained by progression to estimate the probability that an extension of a partial policy satisfies the constraints. Their computation with linear programming is integrated with the recent PLTL-dual heuristic search algorithm, enabling more aggressive pruning of regions violating the constraints. Our experiments show that they further widen the scalability gap between heuristic search and verification approaches to these planning problems.

Diagnosis of Acute Poisoning using explainable artificial intelligence

IntroductionMedical toxicology is the clinical specialty that treats the toxic effects of substances, for example, an overdose, a medication error, or a scorpion sting. The volume of toxicological knowledge and research has, as with other medical specialties, outstripped the ability of the individual clinician to entirely master and stay current with it. The application of machine learning/artificial intelligence (ML/AI) techniques to medical toxicology is challenging because initial treatment decisions are often based on a few pieces of textual data and rely heavily on experience and prior knowledge. ML/AI techniques, moreover, often do not represent knowledge in a way that is transparent for the physician, raising barriers to usability. Logic-based systems are more transparent approaches, but often generalize poorly and require expert curation to implement and maintain. MethodsWe constructed a probabilistic logic network to model how a toxicologist recognizes a toxidrome, using only physical exam findings. Our approach transparently mimics the knowledge representation and decision-making of practicing clinicians. We created a library of 300 synthetic cases of varying clinical complexity. Each case contained 5 physical exam findings drawn from a mixture of 1 or 2 toxidromes. We used this library to evaluate the performance of our probabilistic logic network, dubbed Tak, against 2 medical toxicologists, a decision tree model, as well as its ability to recover the actual diagnosis. ResultsThe inter-rater reliability between Tak and the consensus of human raters was κ = 0.8432 for straightforward cases, 0.4396 for moderately complex cases, and 0.3331 for challenging cases. The inter-rater reliability between the decision tree classifier and the consensus of human raters was, κ = 0.2522 for straightforward cases, 0.1963 for moderately complex cases and 0.0331 for challenging cases. ConclusionsThe software, dubbed Tak, performs comparably to humans on straightforward cases and intermediate difficulty cases, but is outperformed by humans on challenging clinical cases. Tak outperforms a decision tree classifier at all levels of difficulty. Our results are a proof-of-concept that, in a restricted domain, probabilistic logic networks can perform medical reasoning comparably to humans.

Causal inference via string diagram surgery

Abstract Extracting causal relationships from observed correlations is a growing area in probabilistic reasoning, originating with the seminal work of Pearl and others from the early 1990s. This paper develops a new, categorically oriented view based on a clear distinction between syntax (string diagrams) and semantics (stochastic matrices), connected via interpretations as structure-preserving functors. A key notion in the identification of causal effects is that of an intervention, whereby a variable is forcefully set to a particular value independent of any prior propensities. We represent the effect of such an intervention as an endo-functor which performs ‘string diagram surgery’ within the syntactic category of string diagrams. This diagram surgery in turn yields a new, interventional distribution via the interpretation functor. While in general there is no way to compute interventional distributions purely from observed data, we show that this is possible in certain special cases using a calculational tool called comb disintegration. We demonstrate the use of this technique on two well-known toy examples: one where we predict the causal effect of smoking on cancer in the presence of a confounding common cause and where we show that this technique provides simple sufficient conditions for computing interventions which apply to a wide variety of situations considered in the causal inference literature; the other one is an illustration of counterfactual reasoning where the same interventional techniques are used, but now in a ‘twinned’ set-up, with two version of the world – one factual and one counterfactual – joined together via exogenous variables that capture the uncertainties at hand.

The Interplay Between Postsynaptic Striatal D2/3 Receptor Availability, Adversity Exposure and Odd Beliefs: A [11C]-Raclopride PET Study

Between unaffected mental health and diagnosable psychiatric disorders, there is a vast continuum of functioning. The hypothesized link between striatal dopamine signaling and psychosis has guided a prolific body of research. However, it has been understudied in the context of multiple interacting factors, subclinical phenotypes, and pre-postsynaptic dynamics. This work investigated psychotic-like experiences and D2/3 dopamine postsynaptic receptor availability in the dorsal striatum, quantified by in vivo [11C]-raclopride positron emission tomography, in a sample of 24 healthy male individuals. Additional mediation and moderation effects with childhood trauma and key dopamine-regulating genes were examined. An inverse relationship between nondisplaceable binding potential and subclinical symptoms was identified. D2/3 receptor availability in the left putamen fully mediated the association between traumatic childhood experiences and odd beliefs, that is, inclinations to see meaning in randomness and unfounded interpretations. Moreover, the effect of early adversity was moderated by a DRD2 functional variant (rs1076560). The results link environmental and neurobiological influences in the striatum to the origination of psychosis spectrum symptomology, consistent with the social defeat and diathesis-stress models. Adversity exposure may affect the dopamine system as in association with biases in probabilistic reasoning, attributional style, and salience processing. The inverse relationship between D2/3 availability and symptomology may be explained by endogenous dopamine occupying the receptor, postsynaptic compensatory mechanisms, and/or altered receptor sensitivity. This may also reflect a cognitively stabilizing mechanism in non-help-seeking individuals. Future research should comprehensively characterize molecular parameters of dopamine neurotransmission along the psychosis spectrum and according to subtype profiling.

Indicative conditionals: probabilities and relevance

We propose a new account of indicative conditionals, giving acceptability and logical closure conditions for them. We start from Adams’ Thesis: the claim that the acceptability of a simple indicative equals the corresponding conditional probability. The Thesis is widely endorsed, but arguably false and refuted by empirical research. To fix it, we submit, we need a relevance constraint: we accept a simple conditional varphi rightarrow psi to the extent that (i) the conditional probability mathrm{p}(psi |varphi ) is high, provided that (ii) varphi is relevant for psi. How (i) should work is well-understood. It is (ii) that holds the key to improve our understanding of conditionals. Our account has (i) a probabilistic component, using Popper functions; (ii) a relevance component, given via an algebraic structure of topics or subject matters. We present a probabilistic logic for simple indicatives, and argue that its (in)validities are both theoretically desirable and in line with empirical results on how people reason with conditionals.

Hack distributions of rill networks and nonlinear slope length–soil loss relationships

Abstract. Surface flow on rilled hillslopes tends to produce sediment yields that scale nonlinearly with total hillslope length. The widespread observation lacks a single unifying theory for such a nonlinear relationship. We explore the contribution of rill network geometry to the observed yield–length scaling relationship. Relying on an idealized network geometry, we formally develop probability functions for geometric variables of contributing area and rill length. In doing so, we contribute towards a complete probabilistic foundation for the Hack distribution. Using deterministic and empirical functions, we then extend the probability theory to the hydraulic variables that are related to sediment detachment and transport. A Monte Carlo simulation samples hydraulic variables from hillslopes of different lengths to provide estimates of sediment yield. The results of this analysis demonstrate a nonlinear yield–length relationship as a result of the rill network geometry. Theory is supported by numerical modeling, wherein surface flow is routed over an idealized numerical surface and a natural surface from northern Arizona. Numerical flow routing demonstrates probability functions that resemble the theoretical ones. This work provides a unique application of the Scheidegger network to hillslope settings which, because of their finite lengths, result in unique probability functions. We have addressed sediment yields on rilled slopes and have contributed towards understanding Hack's law from a probabilistic reasoning.