Deep Neural Networks are known for impressive results in a wide range of applications, being responsible for many advances in technology over the past few years. However, debugging and understanding neural networks models’ inner workings is a complex task, as there are several parameters and variables involved in every decision. Multidimensional projection techniques have been successfully adopted to display neural network hidden layer outputs in an explainable manner, but comparing different outputs often means overlapping projections or observing them side-by-side, presenting hurdles for users in properly conveying data flow. In this paper, we introduce a novel approach for comparing projections obtained from multiple stages in a neural network model and visualizing differences in data perception. Changes among projections are transformed into trajectories that, in turn, generate vector fields used to represent the general flow of information. This representation can then be used to create layouts that highlight new information about abstract structures identified by neural networks.
This article presents a review and analysis of evaluation practices within the visualization and visual analytics (VIS) domain, with a focus on domain application work accepted at the IEEE VIS conference from 2018 to 2022. Through the analysis of 140 pertinent article, we establish a detailed classification principle for evaluation practices, using the Who, When, What, and How indicators. This principle covers facets such as analysis methods, targets, scenarios, participant expertise, and stages of occurrence. By systematically categorizing the application domains presented in these works, we apply our established classification principle to discern and categorize the evaluation practices within them, identifying the prevailing characteristics and trends. The article explores the variety of evaluation methods employed across different application domains and observes the distinctions in their usage. In conclusion, we provide insights and highlight concerns for conducting evaluations in upcoming domain application research. Our findings are intended to inform and guide subsequent studies in a similar context.
2 May 2024 in IEEE Transactions on Visualization and Computer Graphics
Visualizing event timelines for collaborative text writing is an important application for navigating and understanding such data, as time passes and the size and complexity of both text and timeline increase. They are often employed by applications such as code repositories and collaborative text editors. In this paper, we present a visualization tool to explore historical records of writing of legislative texts, which were discussed and voted on by an assembly of representatives. Our visualization focuses on event timelines from text documents that involve multiple people and different topics, allowing for observation of different proposed versions of said text or tracking data provenance of given text sections, while highlighting the connections between all elements involved. We also describe the process of designing such a tool alongside domain experts, with three steps of evaluation being conducted to verify the effectiveness of our design.