arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2606.29331cs.LGstat.ML
收藏

科学发现的样本复杂度:组合函数树的PAC可学习性

Sample Complexity of Scientific Discovery: PAC Learnability of Compositional Function Trees

Şuayp Talha Kocabay, Talha Rüzgar Akkuş, Kerem Yalçın

AI总结:

研究组合函数树的PAC可学习性,证明其Rademacher复杂度由深度和算子Lipschitz常数控制,风险界为O(L^d/√n),并通过实验验证理论。

摘要:

通过符号回归的科学发现通常被视为统计和计算上难以处理的,因为表达式的假设空间随深度组合增长。本文通过PAC学习的视角重新审视统计方面,聚焦于由有限平滑算子(如$\{+,\times,\sin,\exp\}$和仿射映射)词汇表构建的组合函数树。我们证明了相关的泛化量——Rademacher复杂度,进而过剩风险,不一定随不同符号结构的数量指数爆炸,而是由(i)深度$d$和(ii)沿组合计算图的基础算子的Lipschitz常数控制。具体地,在算子的温和Lipschitz条件和有界仿射叶节点下,词汇表大小$K=|\mathcal{H}_{\mathrm{base}}|$的有限并集界与Maurer型向量收缩结合,得到$\mathfrak{R}_n(\mathcal{H}_{\mathrm{comp}}^{d}) \leq (Kb\sqrt{2}L)^{d-1}\mathfrak{R}_n(\mathcal{H}_{\mathrm{comp}}^{1})$,其中arity界为$b$;对应的高概率风险界为$\mathcal{O}(L^{d}/\sqrt{n})$,当$K,b=O(1)$且$\mathfrak{R}_n(\mathcal{H}_{\mathrm{comp}}^{1})=O(n^{-1/2})$时。我们用一个模块化代码库补充理论,该库在合成“类物理”目标(具有受控深度)上训练可微算子树(非MLP),并显示经验泛化差距与预测的复杂度项$(\widehat{L}^{d})/\sqrt{n}$正相关。

英文摘要:

Scientific discovery via symbolic regression is often viewed as statistically and computationally intractable because the hypothesis space of expressions grows combinatorially with depth. This paper revisits the statistical side through the lens of PAC learning, focusing on compositional function trees built from a finite vocabulary of smooth operators (e.g., $\{+,\times,\sin,\exp\}$ and affine maps). We prove that the relevant generalization quantity, Rademacher complexity, hence the excess risk, does not necessarily blow up exponentially with the number of distinct symbolic structures, but is controlled by (i) the depth $d$ and (ii) the Lipschitz constants of the base operators along the composed computation graph. Concretely, under mild Lipschitz conditions on operators and bounded affine leaves, a finite-union bound over a vocabulary of size $K=|\mathcal{H}_{\mathrm{base}}|$ together with Maurer-type vector contraction yields $\mathfrak{R}_n(\mathcal{H}_{\mathrm{comp}}^{d}) \leq (Kb\sqrt{2}L)^{d-1}\mathfrak{R}_n(\mathcal{H}_{\mathrm{comp}}^{1})$ with arity bound $b$; corresponding high-probability risk bounds scale as $\mathcal{O}(L^{d}/\sqrt{n})$ when $K,b=O(1)$ and $\mathfrak{R}_n(\mathcal{H}_{\mathrm{comp}}^{1})=O(n^{-1/2})$. We complement the theory with a modular codebase that trains differentiable operator trees (not MLPs) on synthetic "physics-like" targets of controlled depth and shows that the empirical generalization gap correlates positively with the predicted complexity term $(\widehat{L}^{d})/\sqrt{n}$.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑