The paper introduces MATCHA, a method that learns explicit many-to-many layer maps between independently trained language models. MATCHA decomposes the alignment into a layer map—an explicit target‑by‑source matrix—and a shared feature map, allowing joint learning of layer correspondence and feature transformation from prompts. Experiments on 42 model pairs across three families show that MATCHA reconstructs target activations more faithfully and improves retrieval metrics, while enabling transfer of activation‑space interventions between models.
arXiv:2603.18908v5 Announce Type: replace Abstract: Independently trained language models often learn compatible late-stage representations, despite differences in training objectives, architectures,...
arXiv:2604.03867v2 Announce Type: replace Abstract: Steering vectors have emerged as a lightweight and effective approach for aligning large language models (LLMs) at inference time, enabling modulat...
The paper extends mechanistic interpretability of large language models by modeling concepts as low‑dimensional non‑linear manifolds rather than linear subspaces. It introduces a concept‑based alignment (CBA) score to compare these manifolds across layers and models, revealing block structures in intermediate layers, a shift from syntax‑dominated to mixed syntactic‑semantic concepts, and training‑dependent multilingual sharing. The study also shows that alignment patterns differ across model families and training stages, with adjacent stages aligning more closely than distant ones.
arXiv:2506. 14126v2 Announce Type: replace-cross Abstract: Modern deep learning is increasingly characterized by the use of open-weight foundation models that can be fine-tuned on specialized datasets.
arXiv:2602. 12952v3 Announce Type: replace Abstract: Adapting large pre-trained models to downstream tasks often produces task-specific parameter updates that are expensive to relearn for every model variant.
arXiv:2605. 11911v2 Announce Type: replace Abstract: Predictive Coding (PC) is an influential account of cortical learning.
arXiv:2607. 17467v1 Announce Type: cross Abstract: Few-shot Test-Time Domain Adaptation (FSTT-DA) seeks to adapt models to novel domains using only a handful of unlabeled target samples.
arXiv:2606. 14900v1 Announce Type: new Abstract: Multi-source transfer learning faces a fundamental scalability bottleneck: existing approaches require either loading all K source models into memory simultaneously during parameter fusion, requiring O(K) memory, or deploying all models at inference time, making production deployment infeasible.
arXiv:2608. 13596v1 Announce Type: cross Abstract: Heterogeneous model fusion seeks to combine models that differ in tasks, initializations, architectures, or scales.
arXiv:2606. 11123v1 Announce Type: new Abstract: Backpropagation (BP) is widely viewed as biologically implausible, in part because it requires feedback weights to be the transpose of forward weights for error propagation.
arXiv:2604. 26170v2 Announce Type: replace Abstract: Adapting large language models (LLMs) to a targeted task efficiently and effectively remains a fundamental challenge.
The paper introduces a method for selecting the best vision‑language model for a downstream task by analyzing the internal dynamics of the visual encoder. It represents each task with layer‑wise conductance and uses an entropy‑regularized alignment to derive a target‑conditioned block importance distribution. The proposed Directional Conductance Divergence (DCD) metric captures asymmetric transferability, enabling accurate prediction of model rankings without direct inference, and achieves a 14.7% NDCG@5 improvement over SWAB on 48 VLMs across 21 datasets.