Search

In recent years, knowledge graph embedding models have been successfully applied in the transductive setting to tackle various challenging tasks including link prediction, and query answering. Yet, the transductive setting does not allow for reasoning over unseen entities, relations, let alone numerical or non-numerical literals. Although increasing efforts are put into exploring inductive scenarios, inference over unseen entities, relations, and literals has yet to come. This limitation prohibits the existing methods from handling real-world dynamic knowledge graphs involving heterogeneous information about the world. Here, we propose a remedy to this limitation. We propose the attentive byte-pair encoding layer (BytE) to construct a triple embedding from a sequence of byte-pair encoded subword units of entities and relations. Compared to the conventional setting, BytE leads to massive feature reuse via weight tying, since it forces a knowledge graph embedding model to learn embeddings for subword units instead of entities and relations directly. Consequently, the size of the embedding matrices are not anymore bound to the unique number of entities and relations of a knowledge graph. Experimental results show that BytE improves the link prediction performance of 4 knowledge graph embedding models on datasets where the syntactic representations of triples are semantically meaningful. However, benefits of training a knowledge graph embedding model with BytE dissipate on knowledge graphs where entities and relations are represented with plain numbers or URIs. We provide an open source implementation of BytE to foster reproducible research., Comment: 8 pages, 4 figures, ECAI 2024 Workshops (CompAI)

We introduce a novel embedding method diverging from conventional approaches by operating within function spaces of finite dimension rather than finite vector space, thus departing significantly from standard knowledge graph embedding techniques. Initially employing polynomial functions to compute embeddings, we progress to more intricate representations using neural networks with varying layer complexities. We argue that employing functions for embedding computation enhances expressiveness and allows for more degrees of freedom, enabling operations such as composition, derivatives and primitive of entities representation. Additionally, we meticulously outline the step-by-step construction of our approach and provide code for reproducibility, thereby facilitating further exploration and application in the field.

Knowledge Graph Embedding (KGE) transforms a discrete Knowledge Graph (KG) into a continuous vector space facilitating its use in various AI-driven applications like Semantic Search, Question Answering, or Recommenders. While KGE approaches are effective in these applications, most existing approaches assume that all information in the given KG is correct. This enables attackers to influence the output of these approaches, e.g., by perturbing the input. Consequently, the robustness of such KGE approaches has to be addressed. Recent work focused on adversarial attacks. However, non-adversarial attacks on all attack surfaces of these approaches have not been thoroughly examined. We close this gap by evaluating the impact of non-adversarial attacks on the performance of 5 state-of-the-art KGE algorithms on 5 datasets with respect to attacks on 3 attack surfaces-graph, parameter, and label perturbation. Our evaluation results suggest that label perturbation has a strong effect on the KGE performance, followed by parameter perturbation with a moderate and graph with a low effect.

Ensemble models often improve generalization performances in challenging tasks. Yet, traditional techniques based on prediction averaging incur three well-known disadvantages: the computational overhead of training multiple models, increased latency, and memory requirements at test time. To address these issues, the Stochastic Weight Averaging (SWA) technique maintains a running average of model parameters from a specific epoch onward. Despite its potential benefits, maintaining a running average of parameters can hinder generalization, as an underlying running model begins to overfit. Conversely, an inadequately chosen starting point can render SWA more susceptible to underfitting compared to an underlying running model. In this work, we propose Adaptive Stochastic Weight Averaging (ASWA) technique that updates a running average of model parameters, only when generalization performance is improved on the validation dataset. Hence, ASWA can be seen as a combination of SWA with the early stopping technique, where the former accepts all updates on a parameter ensemble model and the latter rejects any update on an underlying running model. We conducted extensive experiments ranging from image classification to multi-hop reasoning over knowledge graphs. Our experiments over 11 benchmark datasets with 7 baseline models suggest that ASWA leads to a statistically better generalization across models and datasets

Clifford algebras are a natural generalization of the real numbers, the complex numbers, and the quaternions. So far, solely Clifford algebras of the form $Cl_{p,q}$ (i.e., algebras without nilpotent base vectors) have been studied in the context of knowledge graph embeddings. We propose to consider nilpotent base vectors with a nilpotency index of two. In these spaces, denoted $Cl_{p,q,r}$, allows generalizing over approaches based on dual numbers (which cannot be modelled using $Cl_{p,q}$) and capturing patterns that emanate from the absence of higher-order interactions between real and complex parts of entity embeddings. We design two new models for the discovery of the parameters $p$, $q$, and $r$. The first model uses a greedy search to optimize $p$, $q$, and $r$. The second predicts $(p, q,r)$ based on an embedding of the input knowledge graph computed using neural networks. The results of our evaluation on seven benchmark datasets suggest that nilpotent vectors can help capture embeddings better. Our comparison against the state of the art suggests that our approach generalizes better than other approaches on all datasets w.r.t. the MRR it achieves on validation data. We also show that a greedy search suffices to discover values of $p$, $q$ and $r$ that are close to optimal.

A variety of knowledge graph embedding approaches have been developed. Most of them obtain embeddings by learning the structure of the knowledge graph within a link prediction setting. As a result, the embeddings reflect only the structure of a single knowledge graph, and embeddings for different knowledge graphs are not aligned, e.g., they cannot be used to find similar entities across knowledge graphs via nearest neighbor search. However, knowledge graph embedding applications such as entity disambiguation require a more global representation, i.e., a representation that is valid across multiple sources. We propose to learn universal knowledge graph embeddings from large-scale interlinked knowledge sources. To this end, we fuse large knowledge graphs based on the owl:sameAs relation such that every entity is represented by a unique identity. We instantiate our idea by computing universal embeddings based on DBpedia and Wikidata yielding embeddings for about 180 million entities, 15 thousand relations, and 1.2 billion triples. We believe our computed embeddings will support the emerging field of graph foundation models. Moreover, we develop a convenient API to provide embeddings as a service. Experiments on link prediction suggest that universal knowledge graph embeddings encode better semantics compared to embeddings computed on a single knowledge graph. For reproducibility purposes, we provide our source code and datasets open access., Comment: 5 pages, 3 tables

Most real-world knowledge graphs, including Wikidata, DBpedia, and Yago are incomplete. Answering queries on such incomplete graphs is an important, but challenging problem. Recently, a number of approaches, including complex query decomposition (CQD), have been proposed to answer complex, multi-hop queries with conjunctions and disjunctions on such graphs. However, all state-of-the-art approaches only consider graphs consisting of entities and relations, neglecting literal values. In this paper, we propose LitCQD -- an approach to answer complex, multi-hop queries where both the query and the knowledge graph can contain numeric literal values: LitCQD can answer queries having numerical answers or having entity answers satisfying numerical constraints. For example, it allows to query (1)~persons living in New York having a certain age, and (2)~the average age of persons living in New York. We evaluate LitCQD on query types with and without literal values. To evaluate LitCQD, we generate complex, multi-hop queries and their expected answers on a version of the FB15k-237 dataset that was extended by literal values.

Concept learning deals with learning description logic concepts from a background knowledge and input examples. The goal is to learn a concept that covers all positive examples, while not covering any negative examples. This non-trivial task is often formulated as a search problem within an infinite quasi-ordered concept space. Although state-of-the-art models have been successfully applied to tackle this problem, their large-scale applications have been severely hindered due to their excessive exploration incurring impractical runtimes. Here, we propose a remedy for this limitation. We reformulate the learning problem as a multi-label classification problem and propose a neural embedding model (NERO) that learns permutation-invariant embeddings for sets of examples tailored towards predicting $F_1$ scores of pre-selected description logic concepts. By ranking such concepts in descending order of predicted scores, a possible goal concept can be detected within few retrieval operations, i.e., no excessive exploration. Importantly, top-ranked concepts can be used to start the search procedure of state-of-the-art symbolic models in multiple advantageous regions of a concept space, rather than starting it in the most general concept $\top$. Our experiments on 5 benchmark datasets with 770 learning problems firmly suggest that NERO significantly (p-value <1%) outperforms the state-of-the-art models in terms of $F_1$ score, the number of explored concepts, and the total runtime. We provide an open-source implementation of our approach., Comment: Accepted at IDA 2023

Knowledge graph embedding research has mainly focused on learning continuous representations of knowledge graphs towards the link prediction problem. Recently developed frameworks can be effectively applied in research related applications. Yet, these frameworks do not fulfill many requirements of real-world applications. As the size of the knowledge graph grows, moving computation from a commodity computer to a cluster of computers in these frameworks becomes more challenging. Finding suitable hyperparameter settings w.r.t. time and computational budgets are left to practitioners. In addition, the continual learning aspect in knowledge graph embedding frameworks is often ignored, although continual learning plays an important role in many real-world (deep) learning-driven applications. Arguably, these limitations explain the lack of publicly available knowledge graph embedding models for large knowledge graphs. We developed a framework based on the frameworks DASK, Pytorch Lightning and Hugging Face to compute embeddings for large-scale knowledge graphs in a hardware-agnostic manner, which is able to address real-world challenges pertaining to the scale of real application. We provide an open-source version of our framework along with a hub of pre-trained models having more than 11.4 B parameters., Comment: accepted in Software Impacts journal

In semi-supervised learning, the paradigm of self-training refers to the idea of learning from pseudo-labels suggested by the learner itself. Across various domains, corresponding methods have proven effective and achieve state-of-the-art performance. However, pseudo-labels typically stem from ad-hoc heuristics, relying on the quality of the predictions though without guaranteeing their validity. One such method, so-called credal self-supervised learning, maintains pseudo-supervision in the form of sets of (instead of single) probability distributions over labels, thereby allowing for a flexible yet uncertainty-aware labeling. Again, however, there is no justification beyond empirical effectiveness. To address this deficiency, we make use of conformal prediction, an approach that comes with guarantees on the validity of set-valued predictions. As a result, the construction of credal sets of labels is supported by a rigorous theoretical foundation, leading to better calibrated and less error-prone supervision for unlabeled data. Along with this, we present effective algorithms for learning from credal self-supervision. An empirical study demonstrates excellent calibration properties of the pseudo-supervision, as well as the competitiveness of our method on several benchmark datasets., Comment: 26 pages, 5 figures, 10 tables, to be published at the 12th Symposium on Conformal and Probabilistic Prediction with Applications (COPA 2023)

Knowledge graph embedding research has mainly focused on learning continuous representations of entities and relations tailored towards the link prediction problem. Recent results indicate an ever increasing predictive ability of current approaches on benchmark datasets. However, this effectiveness often comes with the cost of over-parameterization and increased computationally complexity. The former induces extensive hyperparameter optimization to mitigate malicious overfitting. The latter magnifies the importance of winning the hardware lottery. Here, we investigate a remedy for the first problem. We propose a technique based on Kronecker decomposition to reduce the number of parameters in a knowledge graph embedding model, while retaining its expressiveness. Through Kronecker decomposition, large embedding matrices are split into smaller embedding matrices during the training process. Hence, embeddings of knowledge graphs are not plainly retrieved but reconstructed on the fly. The decomposition ensures that elementwise interactions between three embedding vectors are extended with interactions within each embedding vector. This implicitly reduces redundancy in embedding vectors and encourages feature reuse. To quantify the impact of applying Kronecker decomposition on embedding matrices, we conduct a series of experiments on benchmark datasets. Our experiments suggest that applying Kronecker decomposition on embedding matrices leads to an improved parameter efficiency on all benchmark datasets. Moreover, empirical evidence suggests that reconstructed embeddings entail robustness against noise in the input knowledge graph. To foster reproducible research, we provide an open-source implementation of our approach, including training and evaluation scripts as well as pre-trained models in our knowledge graph embedding framework (https://github.com/dice-group/dice-embeddings)., Comment: Accepted at HT 2022

Many applications require explainable node classification in knowledge graphs. Towards this end, a popular ``white-box'' approach is class expression learning: Given sets of positive and negative nodes, class expressions in description logics are learned that separate positive from negative nodes. Most existing approaches are search-based approaches generating many candidate class expressions and selecting the best one. However, they often take a long time to find suitable class expressions. In this paper, we cast class expression learning as a translation problem and propose a new family of class expression learning approaches which we dub neural class expression synthesizers. Training examples are ``translated'' into class expressions in a fashion akin to machine translation. Consequently, our synthesizers are not subject to the runtime limitations of search-based approaches. We study three instances of this novel family of approaches based on LSTMs, GRUs, and set transformers, respectively. An evaluation of our approach on four benchmark datasets suggests that it can effectively synthesize high-quality class expressions with respect to the input examples in approximately one second on average. Moreover, a comparison to state-of-the-art approaches suggests that we achieve better F-measures on large datasets. For reproducibility purposes, we provide our implementation as well as pretrained models in our public GitHub repository at https://github.com/dice-group/NeuralClassExpressionSynthesis, Comment: 18 pages, 2 figures, ESWC 2023--Research Track

Classifying nodes in knowledge graphs is an important task, e.g., for predicting missing types of entities, predicting which molecules cause cancer, or predicting which drugs are promising treatment candidates. While black-box models often achieve high predictive performance, they are only post-hoc and locally explainable and do not allow the learned model to be easily enriched with domain knowledge. Towards this end, learning description logic concepts from positive and negative examples has been proposed. However, learning such concepts often takes a long time and state-of-the-art approaches provide limited support for literal data values, although they are crucial for many applications. In this paper, we propose EvoLearner - an evolutionary approach to learn concepts in ALCQ(D), which is the attributive language with complement (ALC) paired with qualified cardinality restrictions (Q) and data properties (D). We contribute a novel initialization method for the initial population: starting from positive examples, we perform biased random walks and translate them to description logic concepts. Moreover, we improve support for data properties by maximizing information gain when deciding where to split the data. We show that our approach significantly outperforms the state of the art on the benchmarking framework SML-Bench for structured machine learning. Our ablation study confirms that this is due to our novel initialization method and support for data properties., Comment: Accepted at WWW 2022

Concept learning approaches based on refinement operators explore partially ordered solution spaces to compute concepts, which are used as binary classification models for individuals. However, the number of concepts explored by these approaches can grow to the millions for complex learning problems. This often leads to impractical runtimes. We propose to alleviate this problem by predicting the length of target concepts before the exploration of the solution space. By these means, we can prune the search space during concept learning. To achieve this goal, we compare four neural architectures and evaluate them on four benchmarks. Our evaluation results suggest that recurrent neural network architectures perform best at concept length prediction with a macro F-measure ranging from 38% to 92%. We then extend the CELOE algorithm, which learns ALC concepts, with our concept length predictor. Our extension yields the algorithm CLIP. In our experiments, CLIP is at least 7.5 times faster than other state-of-the-art concept learning algorithms for ALC -- including CELOE -- and achieves significant improvements in the F-measure of the concepts learned on 3 out of 4 datasets. For reproducibility, we provide our implementation in the public GitHub repository at https://github.com/dice-group/LearnALCLengths., Comment: 15 pages, 2 figures, 7 tables

Approaches based on refinement operators have been successfully applied to class expression learning on RDF knowledge graphs. These approaches often need to explore a large number of concepts to find adequate hypotheses. This need arguably stems from current approaches relying on myopic heuristic functions to guide their search through an infinite concept space. In turn, deep reinforcement learning provides effective means to address myopia by estimating how much discounted cumulated future reward states promise. In this work, we leverage deep reinforcement learning to accelerate the learning of concepts in $\mathcal{ALC}$ by proposing DRILL -- a novel class expression learning approach that uses a convolutional deep Q-learning model to steer its search. By virtue of its architecture, DRILL is able to compute the expected discounted cumulated future reward of more than $10^3$ class expressions in a second on standard hardware. We evaluate DRILL on four benchmark datasets against state-of-the-art approaches. Our results suggest that DRILL converges to goal states at least 2.7$\times$ faster than state-of-the-art models on all benchmark datasets. We provide an open-source implementation of our approach, including training and evaluation scripts as well as pre-trained models.

Knowledge graph embedding research has mainly focused on the two smallest normed division algebras, $\mathbb{R}$ and $\mathbb{C}$. Recent results suggest that trilinear products of quaternion-valued embeddings can be a more effective means to tackle link prediction. In addition, models based on convolutions on real-valued embeddings often yield state-of-the-art results for link prediction. In this paper, we investigate a composition of convolution operations with hypercomplex multiplications. We propose the four approaches QMult, OMult, ConvQ and ConvO to tackle the link prediction problem. QMult and OMult can be considered as quaternion and octonion extensions of previous state-of-the-art approaches, including DistMult and ComplEx. ConvQ and ConvO build upon QMult and OMult by including convolution operations in a way inspired by the residual learning framework. We evaluated our approaches on seven link prediction datasets including WN18RR, FB15K-237 and YAGO3-10. Experimental results suggest that the benefits of learning hypercomplex-valued vector representations become more apparent as the size and complexity of the knowledge graph grows. ConvO outperforms state-of-the-art approaches on FB15K-237 in MRR, Hit@1 and Hit@3, while QMult, OMult, ConvQ and ConvO outperform state-of-the-approaches on YAGO3-10 in all metrics. Results also suggest that link prediction performances can be further improved via prediction averaging. To foster reproducible research, we provide an open-source implementation of approaches, including training and evaluation scripts as well as pretrained models.

Knowledge graph embedding techniques are key to making knowledge graphs amenable to the plethora of machine learning approaches based on vector representations. Link prediction is often used as a proxy to evaluate the quality of these embeddings. Given that the creation of benchmarks for link prediction is a time-consuming endeavor, most work on the subject matter uses only a few benchmarks. As benchmarks are crucial for the fair comparison of algorithms, ensuring their quality is tantamount to providing a solid ground for developing better solutions to link prediction and ipso facto embedding knowledge graphs. First studies of benchmarks pointed to limitations pertaining to information leaking from the development to the test fragments of some benchmark datasets. We spotted a further common limitation of three of the benchmarks commonly used for evaluating link prediction approaches: out-of-vocabulary entities in the test and validation sets. We provide an implementation of an approach for spotting and removing such entities and provide corrected versions of the datasets WN18RR, FB15K-237, and YAGO3-10. Our experiments on the corrected versions of WN18RR, FB15K-237, and YAGO3-10 suggest that the measured performance of state-of-the-art approaches is altered significantly with p-values <1%, <1.4%, and <1%, respectively. Overall, state-of-the-art approaches gain on average absolute $3.29 \pm 0.24\%$ in all metrics on WN18RR. This means that some of the conclusions achieved in previous works might need to be revisited. We provide an open-source implementation of our experiments and corrected datasets at at https://github.com/dice-group/OOV-In-Link-Prediction.

In recent years, we observe an increasing amount of software with machine learning components being deployed. This poses the question of quality assurance for such components: how can we validate whether specified requirements are fulfilled by a machine learned software? Current testing and verification approaches either focus on a single requirement (e.g., fairness) or specialize on a single type of machine learning model (e.g., neural networks). In this paper, we propose property-driven testing of machine learning models. Our approach MLCheck encompasses (1) a language for property specification, and (2) a technique for systematic test case generation. The specification language is comparable to property-based testing languages. Test case generation employs advanced verification technology for a systematic, property-dependent construction of test suites, without additional user-supplied generator functions. We evaluate MLCheck using requirements and data sets from three different application areas (software discrimination, learning on knowledge graphs and security). Our evaluation shows that despite its generality MLCheck can even outperform specialised testing approaches while having a comparable runtime.

Knowledge graph completion refers to predicting missing triples. Most approaches achieve this goal by predicting entities, given an entity and a relation. We predict missing triples via the relation prediction. To this end, we frame the relation prediction problem as a multi-label classification problem and propose a shallow neural model (SHALLOM) that accurately infers missing relations from entities. SHALLOM is analogous to C-BOW as both approaches predict a central token (p) given surrounding tokens ((s,o)). Our experiments indicate that SHALLOM outperforms state-of-the-art approaches on the FB15K-237 and WN18RR with margins of up to $3\%$ and $8\%$ (absolute), respectively, while requiring a maximum training time of 8 minutes on these datasets. We ensure the reproducibility of our results by providing an open-source implementation including training and evaluation scripts at {\url{https://github.com/dice-group/Shallom}.}, Comment: 15th IEEE International Conference on Semantic Computing, ICSC-2021

In this paper, we study the problem of learning continuous vector representations of knowledge graphs for predicting missing links. We present a new approach called ConEx, which infers missing links by leveraging the composition of a 2D convolution with a Hermitian inner product of complex-valued embedding vectors. We evaluate ConEx against state-of-the-art approaches on the WN18RR, FB15K-237, KINSHIP and UMLS benchmark datasets. Our experimental results show that ConEx achieves a performance superior to that of state-of-the-art approaches such as RotatE, QuatE and TuckER on the link prediction task on all datasets while requiring at least 8 times fewer parameters. We ensure the reproducibility of our results by providing an open-source implementation which includes the training, evaluation scripts along with pre-trained models at https://github.com/conex-kge/ConEx.

Knowledge graph embedding methods learn continuous vector representations for entities in knowledge graphs and have been used successfully in a large number of applications. We present a novel and scalable paradigm for the computation of knowledge graph embeddings, which we dub PYKE . Our approach combines a physical model based on Hooke's law and its inverse with ideas from simulated annealing to compute embeddings for knowledge graphs efficiently. We prove that PYKE achieves a linear space complexity. While the time complexity for the initialization of our approach is quadratic, the time complexity of each of its iterations is linear in the size of the input knowledge graph. Hence, PYKE's overall runtime is close to linear. Consequently, our approach easily scales up to knowledge graphs containing millions of triples. We evaluate our approach against six state-of-the-art embedding approaches on the DrugBank and DBpedia datasets in two series of experiments. The first series shows that the cluster purity achieved by PYKE is up to 26% (absolute) better than that of the state of art. In addition, PYKE is more than 22 times faster than existing embedding solutions in the best case. The results of our second series of experiments show that PYKE is up to 23% (absolute) better than the state of art on the task of type prediction while maintaining its superior scalability. Our implementation and results are open-source and are available at http://github.com/dice-group/PYKE., Comment: 9th Joint International Conference, JIST 2019, Hangzhou, China