Tree-of-Evidence: Efficient "System 2" Search for Faithful Multimodal Grounding

TL;DR

Tree-of-Evidence (ToE) is a search algorithm that improves interpretability of multimodal models by identifying minimal, discrete evidence sets that faithfully reproduce model predictions across various domains.

Contribution

Introduces ToE, a novel inference-time search method that provides faithful, auditable evidence traces for multimodal models, enhancing interpretability without sacrificing predictive performance.

Findings

ToE retains over 0.98 AUROC with as few as five evidence units.

Outperforms other methods in decision agreement and fidelity under sparse evidence budgets.

Adapts its search strategy based on case complexity, using different evidence modalities accordingly.

Abstract

Large Multimodal Models (LMMs) achieve state-of-the-art performance in high-stakes domains like healthcare, yet their reasoning remains opaque. Current interpretability methods, such as attention mechanisms or post-hoc saliency, often fail to faithfully represent the model's decision-making process, particularly when integrating heterogeneous modalities like time-series and text. We introduce Tree-of-Evidence (ToE), an inference-time search algorithm that frames interpretability as a discrete optimization problem. Rather than relying on soft attention weights, ToE employs lightweight Evidence Bottlenecks that score coarse groups or units of data (e.g., vital-sign windows, report sentences) and performs a beam search to identify the compact evidence set required to reproduce the model's prediction. We evaluate ToE across six tasks spanning three datasets and two domains: four clinical prediction tasks on MIMIC-IV, cross-center validation on eICU, and non-clinical fault detection on LEMMA-RCA. ToE produces auditable evidence traces while maintaining predictive performance, retaining over 0.98 of full-model AUROC with as few as five evidence units across all settings. Under sparse evidence budgets, ToE achieves higher decision agreement and lower probability fidelity error than other approaches. Qualitative analyses show that ToE adapts its search strategy: it often resolves straightforward cases using only vitals, while selectively incorporating text when physiological signals are ambiguous. ToE therefore provides a practical mechanism for auditing multimodal models by revealing which discrete evidence units support each prediction.