arXiv:2608. 28150v1 Announce Type: new Abstract: Which geometry controls the rank complexity of normalized softmax attention?
arXiv:2608.28150v2 Announce Type: replace Abstract: How much matrix rank is required to preserve every bounded value output of normalized softmax attention? We study the unrestricted maximum-row-\(\e...
arXiv:2608. 26052v1 Announce Type: new Abstract: Choosing the rank of a low-rank adaptation (LoRA) update is usually an empirical task.
arXiv:2607. 23050v1 Announce Type: new Abstract: Neural scaling laws describe how loss decreases as models, data, and compute grow, but they do not answer a prior question: for a fixed task, what is the minimum model capacity required to solve it?
arXiv:2602. 18849v2 Announce Type: replace-cross Abstract: We develop a sensitivity analysis for transformer attention in a geometry aligned with tokenwise computation.
arXiv:2609.21523v1 Announce Type: new Abstract: A system may be compressed before its downstream task is fully known. We ask how much retained state is then necessary and how much can be saved by lim...
arXiv:2609.06327v2 Announce Type: replace-cross Abstract: A query-oblivious coreset for a softmax-attention head is a subset of the key-value pairs whose attention output is within $\varepsilon$ of t...
The paper proposes a new update geometry for the language‑model head by treating the head and softmax as a single module and using Hilbert’s projective distance to measure functional change. It replaces the spectral norm with the Euclidean row diameter, derives a tractable RowNorm update rule, and demonstrates that RowNorm substantially reduces step diameters and Hilbert perturbations while only slightly increasing validation loss.
The paper introduces Outlier-Aware Attention Linearization (OAL), a method that improves linearized softmax attention by explicitly accounting for a few outlier dimensions that disproportionately affect query-key dot products. OAL builds on a second-order Taylor approximation, uses a Hadamard-product formulation to expose dimension-specific contributions, and groups dimensions with similar scales to share coefficients while treating outliers separately. Experiments on Qwen2.5-0.5B show an 8.9% reduction in frozen-attention output MSE, and fine-tuning with LoRA lowers WikiText-103 perplexity from 31.71 to 27.50 while boosting performance on PIQA, HellaSwag, and ARC-E.
arXiv:2609.32002v2 Announce Type: replace Abstract: The rank $r$ in LoRA is widely treated as a capacity control: a smaller rank is assumed to yield a simpler model that generalizes better. We show t...
arXiv:2609.09130v1 Announce Type: new Abstract: An input may activate few hidden units even when different inputs collectively use an entire network. We study the statistical complexity of this input...
arXiv:2604. 14727v2 Announce Type: replace Abstract: To quantify the geometric capacity of transformers, we develop a tropical-geometric framework for analyzing the spatial partitions induced by conditioned self-attention.
An input may activate few hidden units even when different inputs collectively use an entire network. We study the statistical complexity of this input-dependent sparsity in the one-hidden-layer ReLU model of Awasthi et al.