We introduce CLeVer, a novel visualization system designed to analyze and enhance the performance of continual learning models by detecting task transition failures. Based on the literature review, we discovered and classified three primary causes of task transition failure in classification tasks: class/position inconsistency, biased/noisy samples, and diverse class scopes. These problems are critical in terms of model performance but were not tackled in prior research on continual learning. CLeVer is designed to address these challenges as an integrated system for model experiments and visual analysis. Firstly, users can easily configure the tasks for continual learning through a single JSON file. Then, our system automatically simulates the continual learning process in a relatively short time while generating data for visualization. Finally, the transition visualizer provides an effective visual representation of the task transition process where users can easily detect the task transition failures in continual learning. Our interview with machine learning experts and a case study with three participants demonstrate CLeVer’s utility in detecting and addressing such failures. We also discuss the system’s potential applicability and adaptability for various computer vision tasks while suggesting our future work.

Partitioning a dynamic network into subsets (i.e., snapshots) based on disjoint time intervals is a widely used technique for understanding how structural patterns of the network evolve. However, selecting an appropriate time window (i.e., slicing a dynamic network into snapshots) is challenging and time-consuming, often involving a trial-and-error approach to investigating underlying structural patterns. To address this challenge, we present MoNetExplorer, a novel interactive visual analytics system that leverages temporal network motifs to provide recommendations for window sizes and support users in visually comparing different slicing results. MoNetExplorer provides a comprehensive analysis based on window size, including (1) a temporal overview to identify the structural information, (2) temporal network motif composition, and (3) node-link-diagram-based details to enable users to identify and understand structural patterns at various temporal resolutions. To demonstrate the effectiveness of our system, we conducted a case study with network researchers using two real-world dynamic network datasets. Our case studies show that the system effectively supports users to gain valuable insights into the temporal and structural aspects of dynamic networks.

The quality of parallel corpora used to train a Neural Machine Translation (NMT) model can critically influence the model's performance. Various approaches for refining parallel corpora have been introduced, but there is still much room for improvements, such as enhancing the efficiency and the quality of refinement. We introduce VANT, a novel visual analytics system for refining parallel corpora used in training an NMT model. Our system helps users to readily detect and filter noisy parallel corpora by (1) aiding the quality estimation of individual sentence pairs within the corpora by providing diverse quality metrics (e.g., cosine similarity, BLEU, length ratio) and (2) allowing users to visually examine and manage the corpora based on the pre-computed metrics scores. Our system's effectiveness and usefulness are demonstrated through a qualitative user study with eight participants, including four domain experts with real-world datasets.

A literature review is a critical task in performing research. However, even browsing an academic database and choosing must-read items can be daunting for novice researchers. In this paper, we introduce Papers101, an interactive system that supports novice researchers' discovery of papers relevant to their research topics. Prior to system design, we performed a formative study to investigate what difficul-ties novice researchers often face and how experienced researchers address them. We found that novice researchers have difficulty in identifying appropriate search terms, choosing which papers to read first, and ensuring whether they have examined enough candidates. In this work, we identified key requirements for the system dedicated to novices: prioritizing search results, unifying the contexts of multiple search results, and refining and validating the search queries. Accordingly, Papers101 provides an opinionated perspective on selecting important metadata among papers. It also visualizes how the priority among papers is developed along with the users' knowledge discovery process. Finally, we demonstrate the potential usefulness of our system with the case study on the metadata collection of papers in visualization and HCI community.

Extracting data from pictures of medical records is a common task in the insurance industry as the patients often send their medical invoices taken by smartphone cameras. However, the overall process is still challenging to be fully automated because of low image quality and variation of templates that exist in the status quo. In this paper, we propose a mixed-initiative pipeline for extracting data from pictures of medical invoices, where deep-learning-based automatic prediction models and task-specific heuristics work together under the mediation of a user. In the user study with 12 participants, we confirmed our mixed-initiative approach can supplement the drawbacks of a fully automated approach within an acceptable completion time. We further discuss the findings, limitations, and future works for designing a mixed-initiative system to extract data from pictures of a complicated table.