跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+479新增论文
+361其中 AI 相关
+63已有论文修订
+93新增 AI 导读

arXiv 最近完成扫描:2026-11-16 01:34 · 全站最近入库:2026-11-16 13:44各信源进度 →

统计口径与数据新鲜度

统计 2026-11-15 17:46 — 2026-11-16 17:46(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-16 01:34最近发现新增版本:2026-11-16 13:44库内论文最新更新时间:2026-11-14 02:59

每分钟刷新已保存的数据。空检查不代表采集异常;新收录可能包含历史论文。AI 导读按新增版本计数。

23,827 已收录论文19,941 AI 相关6,921 有修订112 HF 社区发现

浏览论文库

时间范围跟随排序:版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常,查看状态 →

找到 1 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-16 01:33(北京时间)
新论文 · v1arXiv:2611.13543

Sanity Checks for Attention Surgery: Null Models and Matched Interventions for Representation-Motivated Design

Basavaraj R M, Aryan Bharat Kumar

原文摘要

Transformer architectural modifications frequently measure a statistic on internal representations, observe that it is large, remove it architecturally, and report an improvement. Interpreting that sequence requires evaluating whether the statistic exceeds its noise floor, whether it survives a geometric null, and whether the training gain requires the specific direction removed. We formulate these requirements as three reusable checks and apply them to self-attention removal, which deletes the component of an attention head's output aligned with its own value vector. Across nine multi-head pretrained models (5,408 heads, 124M to 6.9B parameters), an earlier-position causal null accounts for 48% to 63% of the head-value cosine alignment, with no monotone scale trend. Under grouped-query attention, this alignment is specific to each head's own key-value group and reverses across groups (+0.24 to +0.27 within-group, -0.11 to -0.19 across-group). Split-half reliability demonstrates that the per-head intervention effect is resolvable in GPT-2 ($r_Δ= 0.80$) but attenuated in Pythia-160M (0.42) and Pythia-410M (0.53), whereas structural self-cosine reliability is near-perfect in all three ($r_{\text{stat}} > 0.99$); which statistic predicts its own intervention is model-dependent. In a pre-registered 3-arm by 8-seed paired factorial at $4.0 \times 10^8$ tokens per run (399,900,672 tokens), a fixed arbitrary rank-one direction does not differ from baseline (+0.0011 nats, $p = 0.197$), while self-directed removal lowers validation loss (-0.0029 nats, Holm-corrected $p = 0.0023$) and outperforms the fixed arbitrary control directly (-0.0040 nats, $p = 0.0018$). At 50.9M scale the training benefit requires the own-value direction rather than a fixed arbitrary one, but a large cosine alignment alone does not establish redundancy in frozen models.

cs.LG · cs.CL