arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2501.18537cs.LGstat.ML
收藏

由f-散度生成的损失函数与算子

Loss Functions and Operators Generated by f-Divergences

Google DeepMind(谷歌DeepMind)

Vincent Roulet, Tianlin Liu, Nino Vieillard, Michael E. Sander, Mathieu Blondel

AI总结:

本研究基于$f$-散度构建推广逻辑损失的凸损失函数,提出对应$f$-softargmax算子及可并行二分计算算法,实证发现α=1.5的α-散度损失在语言模型多任务中表现良好。

摘要:

逻辑损失(又称交叉熵损失)是多分类任务中最常用的损失函数之一,也是语言建模中下一词元预测的首选损失函数,它与Kullback–Leibler(KL)散度以及softargmax算子相关联。在本研究中,我们提出基于$f$-散度构建新的凸损失函数。我们的损失函数从两个方向推广了逻辑损失:其一,用$f$-散度替代KL散度;其二,允许使用非均匀参考测度。我们针对多种$f$-散度实例化了该框架,既复现了已有的损失函数,也构建了新的损失函数。与逻辑损失类似,由$f$-散度生成的损失函数对应一个算子,我们将其命名为$f$-softargmax。我们推导了一种新颖的可并行二分算法,用于计算与任意$f$-散度对应的$f$-softargmax。在实证层面,本文的目标之一是探究经典交叉熵之外的损失函数在语言模型场景下的有效性,涵盖预训练、后训练(SFT,即监督微调)和蒸馏阶段。实验表明,当$α=1.5$时,由$α$-散度(在单位参考测度下等价于Tsallis $α$-负熵)生成的损失函数在多项任务上表现优异。

英文摘要:

The logistic loss (a.k.a. cross-entropy loss) is one of the most popular loss functions used for multiclass classification. It is also the loss function of choice for next-token prediction in language modeling. It is associated with the Kullback--Leibler (KL) divergence and the softargmax operator. In this work, we propose to construct new convex loss functions based on $f$-divergences. Our loss functions generalize the logistic loss in two directions: i) by replacing the KL divergence with $f$-divergences and ii) by allowing non-uniform reference measures. We instantiate our framework for numerous $f$-divergences, recovering existing losses and creating new ones. By analogy with the logistic loss, the loss function generated by an $f$-divergence is associated with an operator, that we dub $f$-softargmax. We derive a novel parallelizable bisection algorithm for computing the $f$-softargmax associated with any $f$-divergence. On the empirical side, one of the goals of this paper is to determine the effectiveness of loss functions beyond the classical cross-entropy in a language model setting, including on pre-training, post-training (SFT) and distillation. We show that the loss function generated by the $α$-divergence (which is equivalent to Tsallis $α$-negentropy in the case of unit reference measures) with $α=1.5$ performs well across several tasks.

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。

↑