In this paper, we investigate whether we could use pruning as a reliable
method to boost the generalization ability of the model. We found that existing
pruning method like L2 can already offer small improvement on the target domain
performance. We further propose a novel pruning scoring method, called DSS,
designed not to maintain source accuracy as typical pruning work, but to
directly enhance the robustness of the model. We conduct empirical experiments
to validate our method and demonstrate that it can be even combined with
state-of-the-art generalization work like MIRO(Cha et al., 2022) to further
boost the performance. On MNIST to MNIST-M, we could improve the baseline
performance by over 5 points by introducing 60% channel sparsity into the
model. On DomainBed benchmark and state-of-the-art MIRO, we can further boost
its performance by 1 point only by introducing 10% sparsity into the model.
Code can be found at:
this https URL

本文研究：剪枝是否可靠地提升模型的泛化能力。我们发现现有剪枝方法像 L2 已经可以在目标域性能上提供小的改进。我们进一步提出了一种被称为 DSS 的新的剪枝评分方法，旨在直接提高模型的鲁棒性。通过实证实验，我们验证了我们的方法，并表明它可以与最先进的泛化工作（如 MIRO）相结合，进一步提高性能。在 MNIST 到 MNIST-M 上，我们可以通过将 60％通道稀疏引入模型来提高基线性能 5 个点以上。在 DomainBed 基准和最先进的 MIRO 上，我们只需将 10％稀疏引入模型即可进一步提高其性能。

剪枝以实现更好的领域通用性

Pruning for Better Domain Generalizability

State space models (SSM) have recently been shown to be very effective as a
deep learning layer as a promising alternative to sequence models such as RNNs,
CNNs, or Transformers. The first version to show this potential was the S4
model, which is particularly effective on tasks involving long-range
dependencies by using a prescribed state matrix called the HiPPO matrix. While
this has an interpretable mathematical mechanism for modeling long
dependencies, it introduces a custom representation and algorithm that can be
difficult to implement. On the other hand, a recent variant of S4 called DSS
showed that restricting the state matrix to be fully diagonal can still
preserve the performance of the original model when using a specific
initialization based on approximating S4's matrix. This work seeks to
systematically understand how to parameterize and initialize such diagonal
state space models. While it follows from classical results that almost all
SSMs have an equivalent diagonal form, we show that the initialization is
critical for performance. We explain why DSS works mathematically, by showing
that the diagonal restriction of S4's matrix surprisingly recovers the same
kernel in the limit of infinite state dimension. We also systematically
describe various design choices in parameterizing and computing diagonal SSMs,
and perform a controlled empirical study ablating the effects of these choices.
Our final model S4D is a simple diagonal version of S4 whose kernel computation
requires just 2 lines of code and performs comparably to S4 in almost all
settings, with state-of-the-art results for image, audio, and medical
time-series domains, and averaging 85\% on the Long Range Arena benchmark.

本文提出了一种简单的 S4 模型的对角变体 S4D，通过限制状态矩阵为全对角可保持原模型性能而又易于实现。