9607 research outputs found
Sort by
From safe teams to superior software: How to establish psychological safety and why it matters
Get PDFThe Calculated Typer (Functional Pearl)
Get PDFWe present a calculational approach to the design of type checkers, showing how they can be derived from behavioural specifications using equational reasoning. We focus on languages whose semantics can be expressed as a fold, and show how the calculations can be simplified using fold fusion. This approach enables the compositional derivation of correct-by-construction type checkers based on solving and composing fusion preconditions. We introduce our approach using a simple expression language, to which we then add support for exception handling and checked exceptions
Impact of the Perceived System Bias and Type of AI Explanations on Decision-Making Effectiveness in Explainable AI Systems: Cognitive and Emotional Mechanisms
No full textArtificial intelligence (AI)-based decision-making systems have been shown to outperform humans. However, in critical decision-making domains like healthcare, human decision-makers often mistrust and are reluctant to follow the recommendations of black-box AI systems because they perceive the system to be biased. This study aims to advance AI for social good by illuminating the mechanisms by which perceived bias in AI systems affects users’ decision-making effectiveness and how different explanation types mitigate these effects. Drawing from Dual-Process Theory and Theory of Effective Use, we propose two mechanisms that mediate the effects of perceived system bias in explainable AI systems: a cognitive mechanism of learning and an emotional mechanism of anticipated regret. Our study found that the cognitive mechanism of learning primarily mediates the relationship between perceived system bias and decision-making effectiveness, and feature importance explanations mitigate the negative effects of perceived system bias more effectively than counterfactual explanations
The Churns and Turns of HCI: Which CHI Papers Make the Most Impact in an Ever-growing Sea of HCI Publications
No full textThe ACM Conference on Human Factors in Computing Systems (CHI) is the premier venue for research in Human-Computer Interaction (HCI). 11,290 full papers have been published and collectively cited almost one million times. Highly cited papers undoubtedly represent influential work, affecting the creation of review standards and conference submission and acceptance practices within and beyond CHI. However, the factors contributing to high citation counts and what constitutes a highly cited CHI paper remain largely unclear. In this panel discussion, we will engage the CHI community in exploring the relationship between paper characteristics, citation numbers, and effective impact on HCI as a discipline, and on HCI as an influential endeavour in technology design and development. To ground this discussion, we present findings from a literature review of the 100 most cited CHI full papers, looking at past and present fields and subfields of influence. We will also share insights from HCI experts. Our goals are to shed light on the meaning of impactful work at CHI and in HCI more broadly, to reflect on key trends in HCI over the years, and to discuss themes that have driven pivotal shifts in HCI research. We will lead the conversation toward a deeper understanding of citation practices, the role of citations in focusing and driving HCI research, and the implications of citation when it comes to shaping what is considered impactful HCI
Bridging Digital Sourcing, Platforms, and Ecosystems:16th International Workshop, DSPE 2025, Obergurgl, Austria, March 4–7, 2025, Revised Selected Papers
No full textA Dynamic Piecewise-Linear Geometric Index with Worst-Case Guarantees.
No full textIndexing data is a fundamental problem in computer science. The input is a set S of n distinct integers from a universe U. Indexing queries take a value q ∈ U and return the membership, predecessor or rank of q in S. A range query takes two values q, r ∈ U and returns the set S ∩[q, r]. Recently, various papers study a special case where the the input data behaves in an approximately piece-wise linear way. Given the sorted (rank,value) pairs, and given some constant ε, one wants to maintain a small number of axis-disjoint line-segments such that, for each rank, the value is within ±ε of the corresponding line-segment. Ferragina and Vinciguerra (VLDB 2020) observe that this geometric problem is useful for solving indexing problems, particularly when the number of line-segments is small compared to the size of the dataset. We study the dynamic version of this geometric problem. In the dynamic setting, inserting or deleting just one data point may cause up to three line-segments to be merged, or one line-segment to be split at most three-way. To determine and compute this, we use techniques from dynamic maintenance of convex hulls, and provide new algorithms with worst-case guarantees, including an O(log n) algorithm to compute a separating line between two non-intersecting convex hulls – an operation previously missing from the literature. We then use our fully-dynamic geometry-based subroutine in an indexing data structure, combining it with a natural hashing technique. The resulting indexing data structure has theoretically efficient worst-case guarantees in expectation. We compare its practical performance to the solution of Ferragina and Vinciguerra, which was shown to perform better in certain structured settings [Sun, Zhou, Li VLDB 2023]. Our empirical analysis shows that our solution supports more efficient range queries in the special case where the update sequence contains many deletions
False Promises in Medical Imaging AI?:Assessing Validity of Outperformance Claims
Get PDFPerformance comparisons are fundamental in medical imaging Artificial Intelligence (AI) research, often driving claims of superiority based on relative improvements in common performance metrics. However, such claims frequently rely solely on empirical mean performance. In this paper, we investigate whether newly proposed methods genuinely outperform the state of the art by analyzing a representative cohort of medical imaging papers. We quantify the probability of false claims based on a Bayesian approach that leverages reported results alongside empirically estimated model congruence to estimate whether the relative ranking of methods is likely to have occurred by chance. According to our results, the majority (>80%) of papers claims outperformance when introducing a new method. Our analysis further revealed a high probability (>5%) of false outperformance claims in 86% of classification papers and 53% of segmentation papers. These findings highlight a critical flaw in current benchmarking practices: claims of outperformance in medical imaging AI are frequently unsubstantiated, posing a risk of misdirecting future research efforts
The value of information in multi-scale feedback systems
No full textComplex adaptive systems (CAS) can be described as system of information flows that dynamically interact across scales to adapt and survive. CAS often consist of many components that work toward a shared goal and interact across different informational scales through feedback loops, leading to their adaptation. In this context, understanding how information is transmitted among system components and across scales becomes crucial for understanding the behavior of CAS. Shannon entropy, a measure of syntactic information, is often used to quantify the size and rarity of messages transmitted between objects and observers, but it does not measure the value that information has for each observer. For this, semantic and pragmatic information have been conceptualized as describing the influence on an observer’s knowledge and actions. Building on this distinction, we describe the architecture of multi-scale information flows in CAS through the concept of multi-scale feedback systems and propose a series of syntactic, semantic, and pragmatic information measures to quantify the value of information flows for adaptation. While the measurement of values is necessarily context-dependent, we provide general guidelines on how to calculate semantic and pragmatic measures and concrete examples of their calculation through four case studies: a robotic collective model, a collective decision-making model, a task distribution model, and a hierarchical oscillator model. Our results contribute to an informational theory of complexity that aims to better understand the role played by information in the behavior of multi-scale feedback systems
Network analysis of the Danish bicycle infrastructure: Bikeability across urban-rural divides
No full textResearch on cycling conditions focuses on cities, because cycling is commonly considered an urban phenomenon. People outside of cities should, however, also have access to the benefits of active mobility. To bridge the gap between urban and rural cycling research, we analyze the bicycle network of Denmark, covering around 43,000 km and nearly 6 million inhabitants. We divide the network into four levels of traffic stress and quantify the spatial patterns of bikeability based on network density, fragmentation, and reach. We find that the country has a high share of low-stress infrastructure, but with a very uneven distribution. The widespread fragmentation of low-stress infrastructure results in low mobility for cyclists who do not tolerate high traffic stress. Finally, we partition the network into bikeability clusters and conclude that both high and low bikeability are strongly spatially clustered. Our research confirms that in Denmark, bikeability tends to be high in urban areas. The latent potential for cycling in rural areas is mostly unmet, although some rural areas benefit from previous infrastructure investments. To mitigate the lack of low-stress cycling infrastructure outside urban centers, we suggest prioritizing investments in urban–rural cycling connections and encourage further research in improving rural cycling conditions