RACE (Residual Alignment for Consistency Estimation) is a forward‑pass statistical framework that assesses the domain‑wide functional consistency of Transformer neurons. It outperforms gradient‑based point estimates in terms of domain specificity and shows a clear association between selected neurons and target domains through token‑distribution experiments. Importantly, RACE’s computational overhead is two orders of magnitude lower than that of gradient‑based methods, enabling scalable domain‑wide analysis.
Discovering stable neuron behavior across entire domains remains a challenge in mechanistic interpretability. Existing methods often rely on instance-level point estimates or computationally expensive...
SYNAPSE is a training‑free framework that probes Transformer models by extracting per‑layer [CLS] representations, ranking neurons with a lightweight linear probe, and applying forward‑hook interventions during inference. It enables controlled, repeatable experiments on internal representations without altering the original model, revealing that task‑relevant information is encoded in broad, overlapping neuron subsets and that small weight or logit manipulations can redirect predictions. The framework highlights functional stability, class‑specific specialization, and complementary vulnerability profiles across domains.
arXiv:2608.24482v1 Announce Type: cross Abstract: Mechanistic Localization bridges mechanistic interpretability and post-training optimization by isolating critical parameters via interpretative appr...
The paper introduces a neural-operator framework that learns a shared operator across multiple neurons of the same cell type, conditioned on biologically interpretable electrophysiological features. This approach scales biophysically realistic neuron models to large populations while preserving rich cellular dynamics and within-cell-type variability. Evaluation shows accurate reproduction of subthreshold membrane dynamics, spike waveforms, firing-rate responses, and spike timing across major cortical inhibitory neuron types, with predicted electrophysiological feature distributions closely matching detailed simulations and human recordings.
Mechanistic Localization bridges mechanistic interpretability and post-training optimization by isolating critical parameters via interpretative approaches and then guiding parameter-efficient Supervi...
arXiv:2607. 24031v1 Announce Type: new Abstract: Brain-Machine Interfaces (BMIs), which link the brain to external devices, hold great potential in rehabilitation, human performance augmentation, and human-centered robotics.
BrainWideBench is a benchmark that evaluates across‑animal transfer on multi‑region neural recordings from 139 mice, covering 276 brain regions. It comprises three task suites—behavior decoding, neural activity prediction, and anatomical organization recovery—to test whether learned representations support diverse downstream objectives. The benchmark shows that while pretraining improves performance over single‑session baselines, current methods vary in transfer ability and none perform uniformly well across all suites, highlighting the challenge of developing general‑purpose neural representations.
arXiv:2606. 11236v1 Announce Type: cross Abstract: Training deep spiking neural networks (SNNs) remains challenging due to sharp loss landscapes and temporal inconsistency caused by surrogate gradients.
arXiv:2512. 01362v2 Announce Type: replace Abstract: Neural prediction offers a promising approach to forecasting the individual variability of neurocognitive functions and disorders and providing prognostic indicators for personalized invention.
arXiv:2606. 00073v1 Announce Type: cross Abstract: We investigate how internal representations emerge across hierarchical processing systems by introducing a neuroscience-inspired framework for analyzing deep spiking neural networks (SNN) through the lens of functional connectivity.
arXiv:2608. 13702v1 Announce Type: cross Abstract: Spiking neural networks (SNNs) offer an energy-efficient alternative to conventional deep neural networks by exploiting sparse event-driven computation, but their training remains challenging because the non-differentiable spike function requires surrogate gradients whose fixed shape may be suboptimal across layers and training stages.
The study introduces a dual‑rule contrastive learning framework that aligns neural dynamics from mice and humans by learning representations organized by biological state rather than species. It demonstrates that this cross‑species representation can recover conserved sensory‑response structures, differentiate disease phenotypes, and track drug efficacy in epilepsy models. The framework also identifies shared disease‑associated neural dynamics between genetically distinct mouse models and human carriers of 16p11.2 copy‑number variants.