To date, no "information-theoretic" frameworks for reasoning about generalization error have been shown to establish minimax rates for gradient descent in the setting of stochastic convex optimization. In this work, we consider the prospect of establishing such rates via several existing information-theoretic frameworks: input-output mutual information bounds, conditional mutual information bounds and variants, PAC-Bayes bounds, and recent conditional variants thereof. We prove that none of these bounds are able to establish minimax rates. We then consider a common tactic employed in studying gradient methods, whereby the final iterate is corrupted by Gaussian noise, producing a noisy "surrogate" algorithm. We prove that minimax rates cannot be established via the analysis of such surrogates. Our results suggest that new ideas are required to analyze gradient descent using information-theoretic techniques.
translated by 谷歌翻译
Practitioners prune neural networks for efficiency gains and generalization improvements, but few scrutinize the factors determining the prunability of a neural network the maximum fraction of weights that pruning can remove without compromising the model's test accuracy. In this work, we study the properties of input data that may contribute to the prunability of a neural network. For high dimensional input data such as images, text, and audio, the manifold hypothesis suggests that these high dimensional inputs approximately lie on or near a significantly lower dimensional manifold. Prior work demonstrates that the underlying low dimensional structure of the input data may affect the sample efficiency of learning. In this paper, we investigate whether the low dimensional structure of the input data affects the prunability of a neural network.
translated by 谷歌翻译
我们研究了学习算法的输出及其$ n $培训数据之间（某些摘要）之间的共同信息，以$ n+1 $ i.i.d.的超级样本为条件。随机选择训练数据而无需更换的数据。这些算法（Steinke and Zakynthinou，2020）的条件相互信息（CMI）的这些剩余变体也被认为可以控制具有有界损耗函数的学习算法的平均通用误差。为了学习在0-1损失（即插值算法）下实现零经验风险的学习算法，我们提供了剩余的CMI与风险的经典保留误差估计之间的明确联系。使用此连接，我们就（评估）保留的CMI获得了上限和下限。当限制风险恒定或多项式衰减时，边界会收敛到两个恒定因子。作为应用程序，我们分析了单个包含图算法的人口风险，这是一种在可实现的环境中的VC类的通用转导学习算法。使用一对一的CMI，我们匹配在可实现的设置中学习VC课程的最佳界限，回答了Steinke和Zakynthinou（2020）提出的开放挑战。最后，为了理解剩余的CMI在研究概括中的作用，我们将剩余的CMI放在措施层次结构中，并在根本上使用新颖的无条件相互信息。对于0-1的损失和插值学习算法，观察到此相互信息恰恰是风险。
translated by 谷歌翻译
在这项工作中，我们调查了Steinke和Zakynthinou（2020）的“条件互信息”（CMI）框架的表现力，以及使用它来提供统一框架，用于在可实现的环境中证明泛化界限。我们首先证明可以使用该框架来表达任何用于从一类界限VC维度输出假设的任何学习算法的非琐碎（但是次优）界限。我们证明了CMI框架在用于学习半个空间的预期风险上产生最佳限制。该结果是我们的一般结果的应用，显示稳定的压缩方案Bousquet al。 （2020）尺寸$ k $有统一有限的命令$ o（k）$。我们进一步表明，适当学习VC类的固有限制与恒定的CMI存在适当的学习者的存在，并且它意味着对Steinke和Zakynthinou（2020）的开放问题的负面分辨率。我们进一步研究了价值最低限度（ERMS）的CMI的级别$ H $，并表明，如果才能使用有界CMI输出所有一致的分类器（版本空间），只有在$ H $具有有界的星号（Hanneke和杨（2015）））。此外，我们证明了一般性的减少，表明“休假”分析通过CMI框架表示。作为推论，我们研究了Haussler等人提出的一包图算法的CMI。 （1994）。更一般地说，我们表明CMI框架是通用的，因为对于每一项一致的算法和数据分布，当且仅当其评估的CMI具有样品的载位增长时，预期的风险就会消失。
translated by 谷歌翻译
We study whether a neural network optimizes to the same, linearly connected minimum under different samples of SGD noise (e.g., random data order and augmentation). We find that standard vision models become stable to SGD noise in this way early in training. From then on, the outcome of optimization is determined to a linearly connected region. We use this technique to study iterative magnitude pruning (IMP), the procedure used by work on the lottery ticket hypothesis to identify subnetworks that could have trained in isolation to full accuracy. We find that these subnetworks only reach full accuracy when they are stable to SGD noise, which either occurs at initialization for small-scale settings (MNIST) or early in training for large-scale settings (ResNet-50 and Inception-v3 on ImageNet).
translated by 谷歌翻译
Distance Metric Learning (DML) has attracted much attention in image processing in recent years. This paper analyzes its impact on supervised fine-tuning language models for Natural Language Processing (NLP) classification tasks under few-shot learning settings. We investigated several DML loss functions in training RoBERTa language models on known SentEval Transfer Tasks datasets. We also analyzed the possibility of using proxy-based DML losses during model inference. Our systematic experiments have shown that under few-shot learning settings, particularly proxy-based DML losses can positively affect the fine-tuning and inference of a supervised language model. Models tuned with a combination of CCE (categorical cross-entropy loss) and ProxyAnchor Loss have, on average, the best performance and outperform models with only CCE by about 3.27 percentage points -- up to 10.38 percentage points depending on the training dataset.
translated by 谷歌翻译
This paper presents an analysis regarding an influence of the Distance Metric Learning (DML) loss functions on the supervised fine-tuning of the language models for classification tasks. We experimented with known datasets from SentEval Transfer Tasks. Our experiments show that applying the DML loss function can increase performance on downstream classification tasks of RoBERTa-large models in few-shot scenarios. Models fine-tuned with the use of SoftTriple loss can achieve better results than models with a standard categorical cross-entropy loss function by about 2.89 percentage points from 0.04 to 13.48 percentage points depending on the training dataset. Additionally, we accomplished a comprehensive analysis with explainability techniques to assess the models' reliability and explain their results.
translated by 谷歌翻译
本文介绍了有关开发的原型的研究，以服务公共政策设计的定量研究。政治学的这种子学科着重于确定参与者，之间的关系以及在健康，环境，经济和其他政策方面可以使用的工具。我们的系统旨在自动化收集法律文件，用机构语法注释它们的过程，并使用超图来分析关键实体之间的相互关系。我们的系统经过了《联合国教科文组织公约》的保护，以保护2003年的无形文化遗产，这是一份法律文件，该文件规定了确保文化遗产的国际关系的基本方面。
translated by 谷歌翻译
提出了联合学习（FL），以促进分布式环境中模型的培训。它支持（本地）数据隐私的保护，并使用本地资源进行模型培训。到目前为止，大多数研究一直致力于“核心问题”，例如机器学习算法对FL，数据隐私保护或处理客户之间不均匀数据分布的影响。此贡献锚定在实际的用例中，在这种情况下，FL将实际部署在生态系统的互联网中。因此，在文献中发现了一些流行的考虑之外，还需要考虑一些不同的问题。此外，引入了一种构建灵活和适应性的FL解决方案的体系结构。
translated by 谷歌翻译
我们介绍了一个大规模实验，该实验对编码器进行了预处理，其参数计数范围从700m到9.3b不等，随后蒸馏到较小的型号中，范围为17m-170亿参数，其应用到自然语言理解（NLU）组件（NLU）组件（虚拟助手系统。尽管我们使用70％的口语数据训练，但在对书面形式的跨语性自然语言推论（XNLI）语料库进行评估时，我们的教师模型与XLM-R和MT5相当。我们使用系统中的内域数据对教师模型进行了第二阶段的训练，以提高了3.86％的相对分类，而相对7.01％的插槽填充。我们发现，即使是从我们的2阶段教师模型中提取的170亿参数模型，与仅接受公共数据的2.3B参数老师相比，与2.3B参数老师相比，意图分类更好2.88％，并且7.69％的插槽填充错误率更好（第1阶段），强调了。内域数据对训练的重要性。当使用标记的NLU数据进行离线评估时，我们的17m参数阶段2蒸馏模型的表现分别优于XLM-R碱基（85m Params）和Distillbert（42m Params），分别优于4.23％至6.14％。最后，我们介绍了一个完整的虚拟助手实验平台的结果，在该平台中，我们发现使用经过预训练和蒸馏管道训练的模型超过了从8500万参数教师蒸馏的模型，在自动测量全系统用户不满的自动测量中，从8500万参数教师蒸馏出3.74％-4.91％。
translated by 谷歌翻译