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

arXiv: https://arxiv.org/abs/2604.07692 · HTML: https://mlsanity.com/paper/2604.07692/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Medical vision-language models, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 16 of its 39 references that are in the map.
- First posted on arXiv 9 Apr 2026.

## Placed in

- Multimodal LLMs / vision-language models / [Medical vision-language models](https://mlsanity.com/mllm/medical-vision-language-models/)

## Builds on (most cited first, 16 of 16)

- [A Unified Approach to Interpreting Model Predictions](https://mlsanity.com/paper/1705.07874/) (2017)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- ["Why Should I Trust You?": Explaining the Predictions of Any Classifier](https://mlsanity.com/paper/1602.04938/) (2016)
- [Counterfactual Explanations without Opening the Black Box: Automated Decisions and the GDPR](https://mlsanity.com/paper/1711.00399/) (2017)
- [Concept Bottleneck Models](https://mlsanity.com/paper/2007.04612/) (2020)
- [Publicly Available Clinical BERT Embeddings](https://mlsanity.com/paper/1904.03323/) (2019)
- [Attention is not Explanation](https://mlsanity.com/paper/1902.10186/) (2019)
- [Attention is not not Explanation](https://mlsanity.com/paper/1908.04626/) (2019)
- [Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?](https://mlsanity.com/paper/2004.03685/) (2020)
- [ERASER: A Benchmark to Evaluate Rationalized NLP Models](https://mlsanity.com/paper/1911.03429/) (2019)
- [Rationalizing Neural Predictions](https://mlsanity.com/paper/1606.04155/) (2016)
- [An Information Bottleneck Approach for Controlling Conciseness in Rationale Extraction](https://mlsanity.com/paper/2005.00652/) (2020)
- [A Comparative Study of Faithfulness Metrics for Model Interpretability Methods](https://mlsanity.com/paper/2204.05514/) (2022)
- [One Explanation is Not Enough: Structured Attention Graphs for Image Classification](https://mlsanity.com/paper/2011.06733/) (2020)
- [The Solvability of Interpretability Evaluation Metrics](https://mlsanity.com/paper/2205.08696/) (2022)
- [Normalized AOPC: Fixing Misleading Faithfulness Metrics for Feature Attribution Explainability](https://mlsanity.com/paper/2408.08137/) (2024)

## Abstract (arXiv)

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.