TL;DR
BACE is a novel framework that learns phase-specific, directed brain region connectivity from intracranial LFP data, enabling interpretable analysis of neural dynamics during behavior.
Contribution
It introduces an end-to-end method for estimating dynamic, phase-specific brain connectivity directly from neural recordings, improving interpretability and predictive accuracy.
Findings
Accurately recovers ground-truth directed interactions on synthetic data.
Reveals behavior-aligned reconfiguration of inter-regional influence in human subcortical LFP.
Provides interpretable, phase-specific connectivity matrices for neural dynamics analysis.
Abstract
Understanding how distributed brain regions coordinate to produce behavior requires models that are both predictive and interpretable. We introduce Behavior-Adaptive Connectivity Estimation (BACE), an end-to-end framework that learns phase-specific, directed inter-regional connectivity directly from multi-region intracranial local field potentials (LFP). BACE aggregates many micro-contacts within each anatomical region via per-region temporal encoders, applies a learnable adjacency specific to each behavioral phase, and is trained on a forecasting objective. On synthetic multivariate time series with known graphs, BACE accurately recovers ground-truth directed interactions while achieving forecasting performance comparable to state-of-the-art baselines. Applied to human subcortical LFP recorded simultaneously from eight regions during a cued reaching task, BACE yields an explicit…
Peer Reviews
Decision·Submitted to ICLR 2026
- The method is validated on real-world human data, which is extremely difficult to acquire. - The work addresses an important topic in the field of neuroimaging: the discovery of effective connectivity with interpretability.
- The choice of the modules within the architecture is rather simple. The paper would benefit from further explanations and comparative experiments on the architectural choice. - More recent baseline methods can be included in the comparative experiments. - The study remains proof-of-concept, with a single-subject data.
The paper introduces a novel and conceptually well-motivated framework, BACE, that jointly performs neural time-series forecasting and interpretable, behavior-adaptive connectivity estimation. Its main strength lies in combining predictive modeling with explicit, directed graph structures that reveal how neural interactions change across behavioral phases. The approach offers clear interpretability, allowing neuroscientific insights into dynamic brain networks, and demonstrates competitive forec
The main weaknesses of the paper lie in its limited empirical validation and lack of reproducibility. All real-data experiments are conducted on a single-subject, private intracranial LFP dataset, which cannot be shared due to clinical restrictions, making it impossible for others to reproduce or verify the results. The method also depends on manually defined behavioral phase labels, restricting its applicability to structured experimental settings rather than continuous or naturalistic data. Mo
The methods' results on the simulation set are encouraging and look quite good. Moreover, I believe the authors are working on an interesting problem, and that although this paper has a lot of potential, it needs more work before I can recommend it for publication. I commend the authors on their work, and encourage them to continue improving it.
Major weaknesses: 1) Baselines and missing related work. Given the interest in this field, I believe the authors are missing a large amount of relevant previous work and baselines. First, the authors do not compare any baselines on their simulation experiments, so it is unclear whether their proposed method uniquely can perform well on the simulation. Second, although the authors refer to their connectivity as 'effective', which I agree with, they do not compare to any commonly used effective co
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
