VizNet

Abstract

and developing benchmark models and algorithms for automating visual analysis. To demonstrate VizNet's utility as a platform for conducting online crowdsourced experiments at scale, we replicate a prior study assessing the influence of user task and data distribution on visual encoding effectiveness, and extend it by considering an additional task: outlier detection. To contend with running such studies at scale, we demonstrate how a metric of perceptual effectiveness can be learned from experimental results, and show its predictive power across test datasets. CCS CONCEPTS• Human-centered computing → Visualization design and evaluation methods; Visualization theory, concepts and paradigms; • Computing methodologies → Machine learning.

To Cut or Not To Cut? A Systematic Exploration of Y-axis Truncation

“…The first group examines how data distributions and user tasks influence the effectiveness of charts [32,33] and visual encodings [26]. Hu et al [22] proposed to view every realized visualization as a tuple of (data 𝐷, visual form 𝑉 , task 𝑇 ).…”

“…We synthesize the ideas from Demiralp et al, Kindlmann and Scheidegger, and Hu et al [16,22,27] and form the following model for reasoning about visualizations: Data 𝐷 ↔ Visual forms 𝑉 ↔ Tasks 𝑇 Given a particular data set 𝑑 in the space of all possible data sets 𝐷, 𝑑 could be mapped to a set of different visual configurations 𝑉 𝑑 ⊆ 𝑉 , some more structure-preserving (or having better "data-visual" correspondence) than others. Each feasible visual configuration 𝑣 ∈ 𝑉 𝑑 can be used to answer tasks, be it low-level [2] or high-level [38].…”

Database Benchmarking for Supporting Real-Time Interactive Querying of Large Data

“…Note that crossfilter-applications generally involve only a small number of attributes [17,41]. Typically, visualization datasets have less than 10 attributes, as reported in [27,Figure 2].…”

Unveiling the power of transfer learning towards efficient artificial intelligence

“…However, most existing works have been focusing on pre-training itself, which only rely on the general distribution information of corpora without considering the difference between semantical and irregular characters. Unlike the natural language, most of the tabular data are organized by non-semantical items, including numbers, strings, or symbols, which approximately remain 70% of the tabular pre-training datasets Table 3.5: Quantitative GED results of student model trained with different loss weight α. like Wikitables and Common Crawl Tables [229,4]. Therefore, the semantical table entities, including both headers and cells, take the remaining 30% but play an important role in high-level table understanding such as column type prediction or table classification based on the semantical attributes.…”

“…"type 0" Table Type Classification Outlier Detection … Figure 3.9: Overview of our model. The proposed model is pretrained by the tabular corpora [4] and external well strctured data such as knowledge graph [5]. Then the model will be applied to multiple downstream tasks by either zero-shot learning or direct finetuning.…”