Problem
Research questions and friction points this paper is trying to address.
visual encoding
bias
vision models
time-series classification
chart-based representation
VEIL: How Visual Encoding Hijacking Induces Bias In Vision Models
🤖 AI Summary
This study investigates whether visual models trained on chart images for time series classification learn genuine temporal patterns or are misled by spurious visual cues introduced by chart encodings. Through similarity analysis, cross-encoding transferability tests, and attribution methods, the authors systematically evaluate the sensitivity of convolutional neural network representations to chart design choices, extending graphical perception research from human readers to machine vision models for the first time. Their findings reveal that model representations are substantially influenced by chart encodings; attention-guided training proves effective only when sensitivity across different encodings is consistent, and can otherwise be ineffective or even detrimental. The work demonstrates that chart design decisions actively shape learned representations and argues that chart-based time series classification should be framed as both a representation and measurement problem.
📝 Abstract
Rendering time series as chart images for CNN-based classification has become increasingly common in time-series classification (TSC). However, it remains unclear whether models learn underlying temporal patterns or rely on encoding-specific visual cues introduced by chart design. We present VEIL: a systematic study examining how chart encodings influence learned representations through complementary analyses of similarity, transferability, and attribution. Attention-guided training appears to mitigate this effect when encoding sensitivity is consistently identified across diagnostics, but provides limited or negative benefit when such signals are absent. These findings position VEIL within the broader question of how machines perceive visualizations -- extending graphical perception from human readers to vision models -- and show that visualization design choices shape learned representations in ways that warrant treating chart-based TSC as a representation and measurement problem rather than a simple modeling decision.
Innovation
Methods, ideas, or system contributions that make the work stand out.
visual encoding
time-series classification
representation bias
chart-based TSC
attention-guided training
🔎 Similar Papers
No similar papers found.
