Joint Manifold Diffusion for Combining Predictions on Decoupled Observations
This addresses a practical limitation in multi-model prediction systems for machine learning practitioners, though it appears to be an incremental improvement over existing predictor combination frameworks.
The paper tackles the problem of combining predictions from different models when they operate on disjoint feature sets, developing a non-parametric algorithm that aligns heterogeneous predictors without requiring joint evaluation data. The method outperforms existing approaches in relative attributes ranking tasks and extends the applicability of predictor combination techniques.
We present a new predictor combination algorithm that improves a given task predictor based on potentially relevant reference predictors. Existing approaches are limited in that, to discover the underlying task dependence, they either require known parametric forms of all predictors or access to a single fixed dataset on which all predictors are jointly evaluated. To overcome these limitations, we design a new non-parametric task dependence estimation procedure that automatically aligns evaluations of heterogeneous predictors across disjoint feature sets. Our algorithm is instantiated as a robust manifold diffusion process that jointly refines the estimated predictor alignments and the corresponding task dependence. We apply this algorithm to the relative attributes ranking problem and demonstrate that it not only broadens the application range of predictor combination approaches but also outperforms existing methods even when applied to classical predictor combination settings.