To Örebro University

We introduce CLEVR-Math, a multi-modal math word problems dataset consisting of simple math word problems involving addition/subtraction, represented partly by a textual description and partly by an image illustrating the scenario. The text describes actions performed on the scene that is depicted in the image. Since the question posed may not be about the scene in the image, but about the state of the scene before or after the actions are applied, the solver envision or imagine the state changes due to these actions. Solving these word problems requires a combination of language, visual and mathematical reasoning. We apply state-of-the-art neural and neuro-symbolic models for visual question answering on CLEVR-Math and empirically evaluate their performances. Our results show how neither method generalise to chains of operations. We discuss the limitations of the two in addressing the task of multi-modal word problem solving.

Compositional generalization is the ability to understand novel combinations of known concepts. Although it is considered as an innate skill for humans, recent studies have shown that neural networks lack this characteristic. In this paper, we focus on compositional generalization with respect to the two specific tasks of word problem solving and visual relation recognition and propose a neuro-symbolic solution, using DeepProbLog, that addresses the problem of compositionality in state-of-the-art neural systems for these tasks.

Emotions are highly useful to model human behavior being at the core of what makes us human. Today, people abundantly express and share emotions through social media. Technological advancements in such platforms enable sharing opinions or expressing any specific emotions towards what others have shared, mainly in the form of textual data. This entails an interesting arena for analysis; as to whether there is a disconnect between the writer's intended emotion and the reader's perception of textual content. In this paper, we present experiments for Readers' Emotion Detection through multi-target regression settings by exploring a Bi-LSTM-based Attention model, where our major intention is to analyze the interpretability and effectiveness of the deep learning model for the task. To conduct experiments, we procure two extensive datasets REN-10k and RENh-4k, apart from using a popular benchmark dataset from SemEval-2007. We perform a two-phase experimental evaluation, first being various coarse-grained and fine-grained evaluations of our model performance in comparison with several baselines belonging to different categories of emotion detection, viz., deep learning, lexicon based, and classical machine learning. Secondly, we evaluate model behavior towards readers' emotion detection assessing attention maps generated by the model through devising a novel set of qualitative and quantitative metrics. The first phase of experiments shows that our Bi-LSTM + Attention model significantly outperforms all baselines. The second analysis reveals that emotions may be correlated to specific words as well as named entities.

Solving kinematics word problems is a specialized task which is best addressed through bespoke logical reasoners. Reasoners, however, require structured input in the form of kinematics parameter values, and translating textual word problems to such structured inputs is a key step in enabling end-to-end automated word problem solving. Span detection for a kinematics parameter is the process of identifying the smallest span of text from a kinematics word problem that has the information to estimate the value of that parameter. A key aspect differentiating kinematics span detection from other span detection tasks is the presence of multiple inter-related parameters for which separate spans need to be identified. State-of-the-art span detection methods are not capable of leveraging the existence of a plurality of inter-dependent span identification tasks. We propose a novel neural architecture that is designed to exploit the inter-relatedness between the separate span detection tasks using a single joint model. This allows us to train the same network for span detection over multiple kinematics parameters, implicitly and automatically transferring knowledge across the kinematics parameters. We show that such a joint training delivers an improvement of accuracies over real-world datasets against state-of-the-art methods for span detection.

Solving kinematics word problems is a specialized task which is best addressed through bespoke logical reasoners. Reasoners, however, require structured input in the form of kinematics parameter values, and translating textual word problems to such structured inputs is a key step in enabling end-to-end automated word problem solving. Span detection for a kinematics parameter is the process of identifying the smallest span of text from a kinematics word problem that has the information to estimate the value of that parameter. A key aspect differentiating kinematics span detection from other span detection tasks is the presence of multiple inter-related parameters for which separate spans need to be identified. State-of-the-art span detection methods are not capable of leveraging the existence of a plurality of inter-dependent span identification tasks. We propose a novel neural architecture that is designed to exploit the inter-relatedness between the separate span detection tasks using a single joint model. This allows us to train the same network for span detection over multiple kinematics parameters, implicitly and automatically transferring knowledge across the kinematics parameters. We show that such a joint training delivers an improvement of accuracies over real-world datasets against state-of-the-art methods for span detection.

An outlier detection method may be considered fair over specified sensitive attributes if the results of outlier detection are not skewed toward particular groups defined on such sensitive attributes. In this paper, we consider the task of fair outlier detection. Our focus is on the task of fair outlier detection over multiple multi-valued sensitive attributes (e.g., gender, race, religion, nationality and marital status, among others), one that has broad applications across modern data scenarios. We propose a fair outlier detection method, FairLOF, that is inspired by the popular LOF formulation for neighborhood-based outlier detection. We outline ways in which unfairness could be induced within LOF and develop three heuristic principles to enhance fairness, which form the basis of the FairLOF method. Being a novel task, we develop an evaluation framework for fair outlier detection, and use that to benchmark FairLOF on quality and fairness of results. Through an extensive empirical evaluation over real-world datasets, we illustrate that FairLOF is able to achieve significant improvements in fairness at sometimes marginal degradations on result quality as measured against the fairness-agnostic LOF method. We also show that a generalization of our method, named FairLOF-Flex, is able to open possibilities of further deepening fairness in outlier detection beyond what is offered by FairLOF.