Attributions are a common local explanation technique for deep learning models on single samples as they are easily extractable and demonstrate the relevance of input values. In many cases, heatmaps visualize such att...
详细信息
ISBN:
(纸本)9781665418171
Attributions are a common local explanation technique for deep learning models on single samples as they are easily extractable and demonstrate the relevance of input values. In many cases, heatmaps visualize such attributions for samples, for instance, on images. However, heatmaps are not always the ideal visualization to explain certain model decisions for other data types. In this review, we focus on attribution visualizations for time series. We collect attribution heatmap visualizations and some alternatives, discuss the advantages as well as disadvantages and give a short position towards future opportunities for attributions and explanations for time series.
Decision-making systems have been widely used in the Financial Services domain. AI is bringing both many innovations and opportunities as well as new risks linked to ethical considerations. Customer trust is at the fo...
详细信息
ISBN:
(纸本)9783030856076;9783030856069
Decision-making systems have been widely used in the Financial Services domain. AI is bringing both many innovations and opportunities as well as new risks linked to ethical considerations. Customer trust is at the forefront of customer retention. To build trust, there is the need to make the decision process Interpretable, Understandable, and Trustworthy for the end-user. Since products offered within the banking sector are usually of an intangible nature, customer trust perception is crucial to maintain a long-standing relationship and to ensure customer loyalty. To this end, in this paper we propose more insightful and user-friendly explanations for decisions made by AI systems in the financial domain.
A well-known method for rendering unstructured volumetric data is tetrahedral marching (tet marching), where rays are marched through a series of tetrahedral elements. However, existing tet marching techniques do not ...
详细信息
ISBN:
(纸本)9781665433358
A well-known method for rendering unstructured volumetric data is tetrahedral marching (tet marching), where rays are marched through a series of tetrahedral elements. However, existing tet marching techniques do not easily generalize to rays with arbitrary origin and direction required for advanced shading effects or non-convex meshes. Additionally, the memory footprint of these methods may exceed GPU memory limits. Interactive performance and high image quality are opposing goals. Our approach significantly lowers the burden to render unstructured datasets with high image fidelity while maintaining real-time and interactive performance even for large datasets. To this end, we leverage hardware-accelerated ray tracing to find entry and exit faces for a given ray into a volume and utilize a compact mesh representation to enable the efficient marching of arbitrary rays, thus allowing for advanced shading effects that ultimately yields more convincing and grounded images. Our approach is also robust, supporting both convex and non-convex unstructured meshes. We show that our method achieves interactive rates even with moderately-sized datasets while secondary effects are applied.
We present a novel magnifying tool for virtual environments, where users are given a view of the world through a handheld window controlled by their real-time eye gaze data. The system builds on the optics of real mag...
详细信息
ISBN:
(纸本)9781665440578
We present a novel magnifying tool for virtual environments, where users are given a view of the world through a handheld window controlled by their real-time eye gaze data. The system builds on the optics of real magnifying glasses and prior work in gaze-directed interfaces. A pilot study is run to evaluate these techniques against a baseline, that reveals no significant improvement in performance, though users appear to prefer the new technique.
We present a VR system that supports the restoration of broken cultural artifacts. As a case study, we demonstrate this approach for the restoration of a funerary monument. Among the challenges of this monument are a ...
详细信息
ISBN:
(纸本)9781665418386
We present a VR system that supports the restoration of broken cultural artifacts. As a case study, we demonstrate this approach for the restoration of a funerary monument. Among the challenges of this monument are a large number of 415 fragments, an unknown amount belongs to another artifact, missing pieces prevent a full reconstruction and the preserved fragments vary strongly in size. Our VR system supports the workflow of digital restoration by offering a configurable self-arranging fragment wall. It supports the user to organize all fragments in an overview representation and to identify relevant fragments quickly. For assembly, we implemented a jigsaw approach comprising two sets of manipulation techniques that allow the user to roughly align fragments first in sub-puzzles and precisely assemble them in a second step. The iterative development and assembling process was accompanied by a professional restorer. We report about the insights we gained from this process and how we optimized the VR system according to her requirements and feedback. Within 14 sessions that took 21 hours, the virtual reconstruction was finalized.
Rule sets are often used in Machine Learning (ML) as a way to communicate the model logic in settings where transparency and intelligibility are necessary. Rule sets are typically presented as a text-based list of log...
详细信息
ISBN:
(纸本)9781665433358
Rule sets are often used in Machine Learning (ML) as a way to communicate the model logic in settings where transparency and intelligibility are necessary. Rule sets are typically presented as a text-based list of logical statements (rules). Surprisingly, to date there has been limited work on exploring visual alternatives for presenting rules. In this paper, we explore the idea of designing alternative representations of rules, focusing on a number of visual factors we believe have a positive impact on rule readability and understanding. We then presents a user study exploring their impact. The results show that some design factors have a strong impact on how efficiently readers can process the rules while having minimal impact on accuracy. This work can help practitioners employ more effective solutions when using rules as a communication strategy to understand ML models.
This work shows an alternative way to shortening social distancing without breaking physical distancing, a medical recommendation in some health conditions, as the current COVID-19 pandemic. As an alternative to tradi...
详细信息
ISBN:
(纸本)9781665440578
This work shows an alternative way to shortening social distancing without breaking physical distancing, a medical recommendation in some health conditions, as the current COVID-19 pandemic. As an alternative to traditional video conferencing, this proposal allows persons to interact with their natural space while actively participating in a remote call. Each participant's reality is extended with a visual representation of the other and binaural sound, enriching the conversation by the sense of the other's proximity. This proposal uses quotidian technology, i.e., smartphones, and an immersive homemade device, to provide an affordable system to display and capture audio stereo and video.
This study examined the influence of set-interval and repetition-interval rest time of virtual reality (VR) boxing game in supine-lying posture. Fifty healthy middle-aged adults were randomly assigned into VR and non-...
详细信息
This study examined the influence of set-interval and repetition-interval rest time of virtual reality (VR) boxing game in supine-lying posture. Fifty healthy middle-aged adults were randomly assigned into VR and non-VR groups to perform six different exercise protocols with varying set-interval and repetition-interval rest times (S0R0, S0R1/3, S0R2/3, S40R0, S40R1/3, and S40R2/3). Analysis on the non-VR group showed significant differences between exercise protocols for average heart rate (p < 0.001), maximum ventilation volume (p < 0.001), respiratory quo-tient (p < 0.001), oxygen pulse (p < 0.001), and excess post-exercise oxygen con-sumption (EPOC) (p = 0.003). VR appeared to have no further improvement on physical training effects in middle-aged adults, while the participants reported negative experience that might be associated with the over-exertion. Future study might need to explore game design elements that can accommodate high-exertion exercises.
This paper presents a web-based Virtual Tour (VT) pipeline to allow organisations to develop a VT experience rapidly, which integrates a collaborative virtual environment with avatars and a 3D Environment to provide v...
详细信息
ISBN:
(纸本)9781665440578
This paper presents a web-based Virtual Tour (VT) pipeline to allow organisations to develop a VT experience rapidly, which integrates a collaborative virtual environment with avatars and a 3D Environment to provide visitors an interactive and shareable experience. Users from an organisation with minimal coding experience can develop a low-cost web-based VT using a 360 camera, the Open Source tools Marzipano and Hubs, the developed software framework and documentation directly from the GitHub repository. The contribution of this work is both the software framework for use with the Marzipano, Mozilla Hubs tool and Tin Can API. A University case study along with a user evaluation demonstrate the effectiveness of the approach.
Cinematic virtual reality, or 360 degrees video, provides viewers with an immersive experience, allowing them to enjoy a video while moving their head to watch in any direction. However, there is an inevitable problem...
详细信息
ISBN:
(纸本)9781665401586
Cinematic virtual reality, or 360 degrees video, provides viewers with an immersive experience, allowing them to enjoy a video while moving their head to watch in any direction. However, there is an inevitable problem of feeling fear of missing out (FOMO) when viewing a 360 degrees video, as only a part of the video is visible to the viewer at any given time. To solve this problem, we developed a technique to present a panoramic thumbnail of a full 360 degrees video to users through a head-mounted display. With this technique, the user can grasp the overall view of the video as needed. We conducted an experiment to evaluate the FOMO, presence, and quality of viewing experience while using this technique compared to normal viewing without it. The results of the experiment show that the proposed technique relieved FOMO, the quality of viewing experience was improved, and there was no difference in presence. We also investigated how users interacted with this new interface based on eye tracking and head tracking data during viewing, which suggested that users used the panoramic thumbnail to actively explore outside their field of view.
暂无评论