Linköping Electronic Conference Proceedings
Not a member yet
1113 research outputs found
Sort by
Topics in Periodicals from the Swedish Diabetes Association 1949 – 1990: Extending the Topic Modelling Tool Topics2Themes with a Timeline Visualisation
Get PDFExisting methods for visualising temporal topic models typically present the information in an aggregated form, and do not offer any possibility to track the specific texts responsible for the change in topic prevalence over time. We present a new type of topic modelling-based timeline visualisation. It still provides an overview with aggregated topic information suitable for distant reading, while also allowing the user to gradually zoom into the image for more detail. At the most detailed level, the individual texts can be reached, which makes it possible to switch to close reading. The timeline visualisation was implemented as an extension of the topic modelling tool Topics2Themes, but this visualisation technique can be adapted to other topic modelling tools and algorithms. We showcase the timeline visualisation on a corpus of periodicals from the Swedish Diabetes Association, which is one of the patient organisation corpora studied within the interdisciplinary project ActDisease. One timeline visualisation was generated for the entire corpus. Additionally, we generated a timeline focusing on the texts that contain the word “dietitian”. The two timelines, including the functionality to zoom into the graphs and reach the texts, were used to analyse the topics and how they vary. It could be concluded that some of the topics and topic timelines were predictable, while others revealed content that might be less expected. These results indicate validity of the method applied, and they also show that this visualisation technique could help us learn something new
Enhancing a multi-faceted verb-centered resource to help a language learner: the case of breton
Get PDFThis article builds on two recent resources for breton, a verb-centered database and a set of sentences in the universal dependencies (UD) format. Our focus is on breton, an endangered language in the celtic family. We provide an analysis of the verb resource and show how it can be connected and transformed to a multi-faceted system intended to help a learner in a flexible way. We discuss several scenarios
Evaluating Automatic Pronunciation Scoring with Crowd-sourced Speech Corpus Annotations
Get PDFPronunciation is an important, and difficult aspect of learning a language. Providing feedback to learners automatically can help train pronunciation, but training a model to do so requires corpora annotated for mispronunciation. Such corpora are rare. We investigate the potential of using the crowdsourced annotations included in Common Voice to indicate mispronunciation. We evaluate the quality of ASR generated goodness of pronunciation scores through the Common Voice corpus against a simple baseline. These scores allow us to see how the Common Voice annotations behave in a real use scenario. We also take a qualitative approach to analyzing the corpus and show that the crowdsourced annotations are a poor substitute for mispronunciation annotations as they typically reflect issues in audio quality or misreadings instead of mispronunciation
Examining the Role of Hockey Leadership to Foster Inclusive Coaching Practices: Discussions from Atlantic Canada
No full textCoaching has been widely examined in the sport of ice hockey. Technical skill development, player management, and the ability to improve performance have been very notable areas of inquiry. As the critical roles of coaching leadership and communication become clearer, there is limited research available which explores the context of inclusive hockey coaching leadership to support more equitable practices. This paper will focus on specific data extracted from a previous study completed by the authors in which general hockey leadership skills and professional development were explored. This paper will present the outcomes of fostering inclusion and diversity from a coaching lens. Thirteen minor hockey coaches from Atlantic Canada (i.e., who are members of the Atlantic Hockey Group) participated in this qualitative study. Semi structured interviews were conducted online or in-person. A thematic analysis was used to explore data obtained from the interviews. Results revealed that coaches had limited communication training experience when working with diverse abilities, age groups, languages, genders, or cultures. Limited professional development specific to inclusive training was noted by participants. Our results demonstrated that various self-led leadership strategies were utilized to promote inclusive practices such as informal community-peer mentorship opportunities, and small group instructional sessions. Overall, the results give us insights into coaches’ experiences with inclusive leadership and highlight current gaps. During the conclusion, future recommendations for continued study, specifically within leadership training for diversity within ice hockey, are offered
Poisoning Attacks on Federated Learning for Autonomous Driving
Get PDFFederated Learning (FL) is a decentralized learning paradigm, enabling parties to collaboratively train models while keeping their data confidential. Within autonomous driving, it brings the potential of reducing data storage costs, reducing bandwidth requirements, and to accelerate the learning. FL is, however, susceptible to poisoning attacks. In this paper, we introduce two novel poisoning attacks on FL tailored to regression tasks within autonomous driving: FLStealth and Off-Track Attack (OTA). FLStealth, an untargeted attack, aims at providing model updates that deteriorate the global model performance while appearing benign. OTA, on the other hand, is a targeted attack with the objective to change the global model's behavior when exposed to a certain trigger. We demonstrate the effectiveness of our attacks by conducting comprehensive experiments pertaining to the task of vehicle trajectory prediction. In particular, we show that, among five different untargeted attacks, FLStealth is the most successful at bypassing the considered defenses employed by the server.For OTA, we demonstrate the inability of common defense strategies to mitigate the attack, highlighting the critical need for new defensive mechanisms against targeted attacks within FL for autonomous driving
3D Pointcloud Registration In-the-wild
Get PDFThis study assesses two state-of-the-art (SOTA) pointcloud registration approaches on industrially challenging datasets, focusing on two specific cases. The first case involves the application of Lidar-based Simultaneous Localization and Mapping (SLAM) in a tunnel environment, while the second case revolves around aligning RGBD scans from intricately symmetrical cast-iron machine parts within the domain of small-scale industrial production. Our evaluation involves testing state-of-the-art pointcloud registration approaches both with and without fine-tuning, and comparing the results to a classical hand crafted feature extractors. Our experimental findings reveal that existing SOTA models exhibit limited generalization capability when confronted with the more challenging pointcloud data. Moreover, robust generalizable methods beyond training are currently unavailable, highlighting a notable gap in addressing challenges associated with industrial datasetsin pointcloud registration
Local Point-wise Explanations of LambdaMART
Get PDFLambdaMART has been shown to outperform neural network models on tabular Learning-to-Rank (LTR) tasks. Similar to the neural network models, LambdaMART is considered a black-box model due to the complexity of the logic behind its predictions. Explanation techniques can help us understand these models. Our study investigates the faithfulness of point-wise explanation techniques when explaining LambdaMART models. Our analysis includes LTR-specific explanation techniques, such as LIRME and EXS, as well as explanation techniques that are not adapted to LTR use cases, such as LIME, KernelSHAP, and LPI. The explanation techniques are evaluated using several measures: Consistency, Fidelity, (In)fidelity, Validity, Completeness, and Feature Frequency (FF) Similarity. Three LTR benchmark datasets are used in the investigation: LETOR 4 (MQ2008), Microsoft Bing Search (MSLR-WEB10K), and Yahoo! LTR challenge dataset. Our empirical results demonstrate the challenges of accurately explaining LambdaMART: no single explanation technique is consistently faithful across all our evaluation measures and datasets. Furthermore, our results show that LTR-based explanation techniques are not consistently better than their non-LTR-based counterparts across the evaluation measures. Specifically, the LTR-based explanation techniques consistently are the most faithful with respect to (In)fidelity, whereas the non-LTR-specific approaches are shown to frequently provide the most faithful explanations with respect to Validity, Completeness, and FF Similarity
Designing Robots to Help Women
Get PDFRobots are being designed to help people in an increasing variety of settings--but seemingly little attention has been given so far to the specific needs of women, who represent roughly half of the world's population but are underrepresented in robotics. Here we used a speculative prototyping approach to explore this expansive design space: First, we identified some challenges that disproportionately affect women in relation to crime, health, and daily activities, as well as opportunities for designers, which were visualized in five sketches. Then, one of the sketched scenarios was further explored by developing a prototype, of a drone equipped with computer vision to detect hidden cameras that could be used to spy on women. While object detection introduced some errors, hidden cameras were identified with a reasonable accuracy of 80% (Intersection over Union (IoU) score: 0.40). Our aim is that these results could help spark discussion and inspire designers, toward realizing a safer, more inclusive future through responsible use of technology
The Bias that Lies Beneath: Qualitative Uncovering of Stereotypes in Large Language Models
Get PDFThe rapid growth of Large Language Models (LLMs), such as ChatGPT and Mistral, has raised concerns about their ability to generate inappropriate, toxic and ethically problematic content. This problem is further amplified by LLMs' tendency to reproduce the prejudices and stereotypes present in their training datasets, which include misinformation, hate speech and other unethical content. Traditional methods of automatic bias detection rely on static datasets that are unable to keep up with society's constantly changing prejudices, and so fail to capture the large diversity of biases, especially implicit associations related to demographic characteristics like gender, ethnicity, nationality, and so on. In addition, these approaches frequently use adversarial techniques that force models to generate harmful language. In response, this study proposes a novel qualitative protocol based on prompting techniques to uncover implicit bias in LLM-generated texts without explicitly asking for prejudicial content. Our protocol focuses on biases associated with gender, sexual orientation, nationality, ethnicity and religion, with the aim of raising awareness of the stereotypes perpetuated by LLMs. We include the Tree of Thoughts technique (ToT) in our protocol, enabling a systematic and strategic examination of internal biases. Through extensive prompting experiments, we demonstrate the effectiveness of the protocol in detecting and assessing various types of stereotypes, thus providing a generic and reproducible methodology. Our results provide important insights for the ethical evaluation of LLMs, which is essential in the current climate of rapid advancement and implementation of generative AI technologies across various industries.
Warning: This paper contains explicit statements of offensiveor upsetting contents
Analysing Unlabeled Data with Randomness and Noise: The Case of Fishery Catch Reports
Get PDFDetecting violations within fishing activity reports is crucial for ensuring the sustainable utilization of fish resources, and employing machine learning methods holds promise for uncovering hidden patterns within this complex dataset. Given that these violations are infrequent occurrences, as fishermen generally adhere to regulations, identifying them becomes akin to an anomaly outlier detection task. Since labeled data distinguishing between normal and anomalous instances is not available for catch reports from Norwegian waters, we have opted for more conventional approaches, such as clustering methods, to identify potential clusters and outliers. Moreover, the catch reports inherently exhibit randomness and noise due to environmental factors and potential errors made by fishermen during report registration which complicates the processes of scaling, clustering, and anomaly detection. Through experimentation with various scaling and clustering techniques, we have observed that many of these methods tend to group the data based on the species caught, exhibiting a high level of agreement in cluster formation, indicating the stability of the clusters. Anomaly detection methods, however, yield varying potential outliers as it is a more challenging task