We consider the exact and probably approximately correct (PAC) learning frameworks from computational learning theory and discuss opportunities and challenges for applying notions developed within these frameworks to extract information from black-box machine learning models, in particular, from language models. We discuss recent works that consider algorithms designed for the exact and PAC frameworks to extract information in the format of automata, Horn theories, and ontologies from machine learning models and possible applications of these approaches for understanding the models, studying biases, and knowledge acquisition.
Knowledge graphs (KGs) have in recent years gained a large momentum both in academic research and in business applications. They have become a bridge between databases, artificial intelligence (AI), data science, the (semantic) web, linked data, and many other areas. In particular, in declarative AI, they have become a bridge between logic-based reasoning, and machine learning-based reasoning. Languages for KGs on the one hand, and systems for KGs – i.e., Knowledge Graph Managament System (KGMS) – on the other hand, have garnered increasing attention. Of particular importance are language and system extensions – such as probabilistic reasoning, numeric reasoning, etc. – supporting various real-world applications, and the business applications that can be built using such extensions. In this work, we give an overview of the Vadalog language and system, a KGMS. We focus on three areas: (1) a basic overview, including an introduction to dependencies, the Datalog and Vadalog languages, (2) the extensions of the system, including arithmetic and aggregation, real-world data interfaces, temporal reasoning, and machine learning, and (3) the business applications, including: corporate governance, media intelligence, supply chains, collateral eligibility, hostile takeovers, smart anonymization, and anti-money laundering.
In recent years, multimodal knowledge has become a popular research topic in many fields, such as knowledge graphs and natural language processing. Multimodal knowledge involves multimodal knowledge graphs, multimodal pre-trained language models, multimodal knowledge inference, etc.; from online shopping to medical care, whether it is theoretical research or engineering application, the knowledge representation, discovery, and inference of multimodal knowledge have become the core technologies of the academic and industrial concern. This tutorial focuses on the state of the art of cross-modal knowledge discovery and inference and presents future research opportunities and challenges.
Causal inference is a fundamental concept that goes beyond simple correlation and model-based prediction analysis, and is highly relevant in domains such as health, medicine, and the social sciences. Causal inference enables the estimation of the impact of an intervention or treatment on the world, making it critical for sound and robust policy making. However, randomized controlled experiments, which are typically considered as the gold standard for inferring causal conclusions, are often not feasible due to ethical, cost, or other constraints. Fortunately, there is a rich literature in Artificial Intelligence (AI), Machine Learning (ML), and Statistics on observational studies, which are methods for causal inference on observed or collected data under certain assumptions. In this paper, we provide an overview of popular formal and rigorous techniques for causal inference on observed data from the AI and Statistics literature. Furthermore, we discuss how concepts from causal inference can be used to infer fairness and enable explainability in machine learning models, which are critical in responsible data science when ML is used in making high-stake decisions in various contexts. Our discussion highlights the importance of using causal inference in ML models and provides insights on how to develop more transparent and responsible AI systems.
This tutorial presents a statistical relational extension of the answer set programming language called $$\textrm{LP}^\mathrm{{MLN}}$$ , which incorporates the concept of weighted rules into the stable model semantics following the log-linear models of Markov Logic. An $$\textrm{LP}^\mathrm{{MLN}}$$ program defines a probability distribution over "soft" stable models, which may not satisfy all rules, but the more rules with larger weights they satisfy, the higher their probabilities, thus allowing for an intuitive and elaboration tolerant representation of problems that require both logical and probabilistic reasoning. The extension provides a natural way to overcome the deterministic nature of the stable model semantics, such as resolving inconsistencies in answer set programs, associating probability to stable models, and applying statistical inference and learning with probabilistic stable models. We also present formal relations between $$\textrm{LP}^\mathrm{{MLN}}$$ and other related formalisms, which produce ways of performing inference and learning in $$\textrm{LP}^\mathrm{{MLN}}$$ .
In this expository article we highlight the relevance of explanations for artificial intelligence, in general, and for the newer developments in {\em explainable AI}, referring to origins and connections of and among different approaches. We describe in simple terms, explanations in data management and machine learning that are based on attribution-scores, and counterfactuals as found in the area of causality. We elaborate on the importance of logical reasoning when dealing with counterfactuals, and their use for score computation.
The last decade witnessed an ever-increasing stream of successes in Machine Learning (ML). These successes offer clear evidence that ML is bound to become pervasive in a wide range of practical uses, including many that directly affect humans. Unfortunately, the operation of the most successful ML models is incomprehensible for human decision makers. As a result, the use of ML models, especially in high-risk and safety-critical settings is not without concern. In recent years, there have been efforts on devising approaches for explaining ML models. Most of these efforts have focused on so-called model-agnostic approaches. However, all model-agnostic and related approaches offer no guarantees of rigor, hence being referred to as non-formal. For example, such non-formal explanations can be consistent with different predictions, which renders them useless in practice. This paper overviews the ongoing research efforts on computing rigorous model-based explanations of ML models; these being referred to as formal explanations. These efforts encompass a variety of topics, that include the actual definitions of explanations, the characterization of the complexity of computing explanations, the currently best logical encodings for reasoning about different ML models, and also how to make explanations interpretable for human decision makers, among others.
