Semantic Allocation in Ordered Bottlenecks: Predictive Residual Inference for Visual Representation Learning
For researchers in visual representation learning, PRIOR provides a more robust method for ordered bottlenecks that overcomes limitations of masking-based ordering pressure, particularly in discrete/quantized settings.
The paper introduces PRIOR, a framework for ordered bottlenecks that uses log2-scaled levels and level-wise predictors to separate explained from unexplained information, replacing activation-rate control. PRIOR achieves well-ordered representations across contrastive learning and image reconstruction tasks, with higher full-budget performance than baselines in most settings and near-continuous performance in discrete/quantized settings.
Ordered bottlenecks aim to provide utility at flexible budgets by assigning coarse information to early tokens and task-relevant detail to later ones. Prior work, including tail dropping (TD), typically enforces ordering by means of a masking-based ordering pressure (MBOP): Late tokens are masked more frequently than early tokens and are therefore encouraged to store less essential fine details. We introduce predictive residual inference for ordered representations (PRIOR), a framework designed to address inherent weaknesses of MBOP. MBOP is prone to weak late-token utility because it lacks an explicit refinement objective and uses gradient exposure as a proxy for importance. Furthermore, representations may become particularly brittle in optimization-sensitive settings, such as when using discrete or quantized token representations. PRIOR replaces activation-rate control with log2-scaled levels and level-wise predictors. These predictors separate already explained from unexplained information, focusing each level on residual error. We compare PRIOR against MBOP-TD and independent tail-biased dropout (MBOP-ITD) in contrastive learning and image reconstruction tasks. Unlike the baselines, PRIOR learns well-ordered representations across experiments: low budgets provide coarse descriptors, while high budgets add refinements. Simultaneously, full-budget performance with PRIOR is higher in all but one experimental setting, where performance remains comparable. MBOP baselines are severely limited in discrete and quantized settings, while PRIOR approaches the performance of continuous counterparts. Taken together, these findings establish PRIOR as an effective framework for ordered representation learning.