arXiv 研究发现神经网络权重空间保留学习轨迹信息
Learning Trajectories Leave Decodable Traces in Neural Network Weight Space and Internal Representations
这篇论文挺有意思:网络练完后还能从权重里读出它最初的初始化和任务学习顺序,80 个网络上的探针做到 100% 准确。做模型取证、训练过程分析或者关心可复现性的人都该看看。
一项 arXiv 论文测试了训练后的神经网络权重是否保留学习历史信息。实验中,30 个训练后的网络全部能从每数据集十个候选中正确匹配回自己的初始权重,线性探针在 80 个保留网络上以 100% 准确率解码出此前两个任务的学习顺序。
机制上,共享相同初始化的网络即使在不同 batch 调度下训练,最终的隐层表征仍可被识别出来源;训练会重组整个网络集合,初始的成对距离不能预测最终距离。不过历史信息是选择性的:早期的中间权重状态只能部分预测,单个初始权重无法可靠重建。
另外 2,160 对对齐的端点中有 2,157 对在采样插值点上没有正损失屏障,说明不同解之间通过低损失路径相连,而非收敛到同一权重向量。受扰动的网络能恢复分类性能,但不回到原有训练路径。结论是共同任务约束网络走向相关但各异的解,最终权重同时记录了学到了什么以及如何学到的。
Learning Trajectories Leave Decodable Traces in Neural Network Weight Space and Internal Representations
Neural networks trained on the same task develop related weight-space structures, yet remain distinguishable after learning. Do these differences merely reflect alternative solutions, or do they preserve information about individual learning histories? We tested whether final weights reveal initialization, task order, and earlier trajectory states. Training reorganized the ensemble: initial pairwise distances did not consistently predict final distances. Nevertheless, all 30 trained networks were correctly matched to their initial weights among ten candidates per dataset. Networks sharing an initialization also developed recognizable final hidden representations across different batch schedules, although direct comparisons with initial activity did not recover identity. Learning order remained readable as well. A linear probe decoded the order of two earlier tasks with 100% accuracy on 80 held-out networks after ten epochs of a common final task. Earlier intermediate weight states were partly predictable, but individual initial weights could not be reliably reconstructed. Final weights thus retain selective historical information rather than a complete trajectory record. Perturbed networks recovered classification performance without returning to their unperturbed training paths. This separation persisted across datasets, intervention times, and network layers. Moreover, 2,157 of 2,160 aligned endpoint pairs showed no positive loss barrier at sampled interpolation points, indicating low-loss connections between distinct solutions rather than convergence to one weight vector. Together, these findings show that a shared task constrains networks to related successful solutions without erasing their individual histories. Final weights reflect not only what a network has learned, but also where learning began and how it unfolded.
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include neural networks, linear probes.
Source attribution: curated from arXiv cs.AI (@Patrick Krauss, Achim Schilling, Andreas Maier, Thomas Kinfe, Claus Metzner).