Deep neural networks have strong capabilities of memorizing the underlying training data, which can be a serious privacy concern. An effective solution to this problem is to train models with differential privacy, which provides rigorous privacy guarantees by injecting random noise to the gradients. This paper focuses on the scenario where sensitive data are distributed among multiple participants, who jointly train a model through federated learning (FL), using both secure multiparty computation (MPC) to ensure the confidentiality of each gradient update, and differential privacy to avoid data leakage in the resulting model. A major challenge in this setting is that common mechanisms for enforcing DP in deep learning, which inject real-valued noise, are fundamentally incompatible with MPC, which exchanges finite-field integers among the participants. Consequently, most existing DP mechanisms require rather high noise levels, leading to poor model utility. Motivated by this, we propose Skellam mixture mechanism (SMM), an approach to enforce DP on models built via FL. Compared to existing methods, SMM eliminates the assumption that the input gradients must be integer-valued, and, thus, reduces the amount of noise injected to preserve DP. Further, SMM allows tight privacy accounting due to the nice composition and sub-sampling properties of the Skellam distribution, which are key to accurate deep learning with DP. The theoretical analysis of SMM is highly non-trivial, especially considering (i) the complicated math of differentially private deep learning in general and (ii) the fact that the mixture of two Skellam distributions is rather complex, and to our knowledge, has not been studied in the DP literature. Extensive experiments on various practical settings demonstrate that SMM consistently and significantly outperforms existing solutions in terms of the utility of the resulting model.
translated by 谷歌翻译
Neural-symbolic computing aims at integrating robust neural learning and sound symbolic reasoning into a single framework, so as to leverage the complementary strengths of both of these, seemingly unrelated (maybe even contradictory) AI paradigms. The central challenge in neural-symbolic computing is to unify the formulation of neural learning and symbolic reasoning into a single framework with common semantics, that is, to seek a joint representation between a neural model and a logical theory that can support the basic grounding learned by the neural model and also stick to the semantics of the logical theory. In this paper, we propose differentiable fuzzy $\mathcal{ALC}$ (DF-$\mathcal{ALC}$) for this role, as a neural-symbolic representation language with the desired semantics. DF-$\mathcal{ALC}$ unifies the description logic $\mathcal{ALC}$ and neural models for symbol grounding; in particular, it infuses an $\mathcal{ALC}$ knowledge base into neural models through differentiable concept and role embeddings. We define a hierarchical loss to the constraint that the grounding learned by neural models must be semantically consistent with $\mathcal{ALC}$ knowledge bases. And we find that capturing the semantics in grounding solely by maximizing satisfiability cannot revise grounding rationally. We further define a rule-based loss for DF adapting to symbol grounding problems. The experiment results show that DF-$\mathcal{ALC}$ with rule-based loss can improve the performance of image object detectors in an unsupervised learning way, even in low-resource situations.
translated by 谷歌翻译
New architecture GPUs like A100 are now equipped with multi-instance GPU (MIG) technology, which allows the GPU to be partitioned into multiple small, isolated instances. This technology provides more flexibility for users to support both deep learning training and inference workloads, but efficiently utilizing it can still be challenging. The vision of this paper is to provide a more comprehensive and practical benchmark study for MIG in order to eliminate the need for tedious manual benchmarking and tuning efforts. To achieve this vision, the paper presents MIGPerf, an open-source tool that streamlines the benchmark study for MIG. Using MIGPerf, the authors conduct a series of experiments, including deep learning training and inference characterization on MIG, GPU sharing characterization, and framework compatibility with MIG. The results of these experiments provide new insights and guidance for users to effectively employ MIG, and lay the foundation for further research on the orchestration of hybrid training and inference workloads on MIGs. The code and results are released on https://github.com/MLSysOps/MIGProfiler. This work is still in progress and more results will be published soon.
translated by 谷歌翻译
我们表明,将人类的先验知识与端到端学习相结合可以通过引入基于零件的对象分类模型来改善深神经网络的鲁棒性。我们认为,更丰富的注释形式有助于指导神经网络学习更多可靠的功能,而无需更多的样本或更大的模型。我们的模型将零件分割模型与一个微小的分类器结合在一起,并经过训练的端到端,以同时将对象分割为各个部分,然后对分段对象进行分类。从经验上讲,与所有三个数据集的Resnet-50基线相比,我们的基于部分的模型既具有更高的精度和更高的对抗性鲁棒性。例如,鉴于相同的鲁棒性,我们部分模型的清洁准确性高达15个百分点。我们的实验表明,这些模型还减少了纹理偏见,并对共同的腐败和虚假相关性产生更好的鲁棒性。该代码可在https://github.com/chawins/adv-part-model上公开获得。
translated by 谷歌翻译
时间动作本地化旨在预测未修剪长视频中每个动作实例的边界和类别。基于锚或建议的大多数先前方法忽略了整个视频序列中的全局本地上下文相互作用。此外,他们的多阶段设计无法直接生成动作边界和类别。为了解决上述问题,本文提出了一种新颖的端到端模型,称为自适应感知变压器(简称apperformer)。具体而言,Adaperformer探索了双支球多头的自我发项机制。一个分支会照顾全球感知的关注,该注意力可以模拟整个视频序列并汇总全球相关环境。而其他分支集中于局部卷积转移,以通过我们的双向移动操作来汇总框架内和框架间信息。端到端性质在没有额外步骤的情况下产生视频动作的边界和类别。提供了广泛的实验以及消融研究,以揭示我们设计的有效性。我们的方法在Thumos14数据集上实现了最先进的准确性(根据map@0.5、42.6 \%map@0.7和62.7 \%map@avg),并在活动网络上获得竞争性能, -1.3数据集,平均地图为36.1 \%。代码和型号可在https://github.com/soupero/adaperformer上找到。
translated by 谷歌翻译
当今AI应用程序的成功不仅需要模型培训(以模型为中心),还需要数据工程(以数据为中心)。在以数据为中心的AI中,主动学习(AL)起着至关重要的作用,但是当前的AL工具无法有效执行AL任务。为此,本文介绍了一个有效的MLOPS系统,该系统名为Alaas(主动学习-AS-A-Service)。具体而言,ALAAS采用服务器客户架构来支持AL管道并实现阶段级并行性以提高效率。同时,使用缓存和批处理技术进一步加速了AL过程。除效率外,ALAAS还可以借助于配置的设计理念,以确保可访问性。它还将AL过程抽象到多个组件,并为高级用户提供丰富的API,以将系统扩展到新方案。广泛的实验表明,在潜伏期和吞吐量方面,ALAAS优于所有其他基线。进一步的消融研究证明了我们的设计和Alaas易于使用的有效性。我们的代码可在\ url {https://github.com/mlsysops/alaas}中获得。
translated by 谷歌翻译
我们介绍了一个3D实例表示,称为实例内核,其中实例由一维向量表示,该向量编码3D实例的语义,位置和形状信息。我们显示了实例内核通过简单地在整个场景中扫描内核,避免对标准3D实例分段管道中的建议或启发式聚类算法的严重依赖,从而实现了简单的掩盖推理。实例内核的想法是受到2D/3D实例分割中动态卷积的最新成功的启发。但是,我们发现由于点云数据的无序和非结构化的性质,代表3D实例是非平凡的,例如,糟糕的实例定位可以显着降低实例表示。为了解决这个问题,我们构建了一个编码范式的新颖3D实例。首先,潜在的实例质心定位为候选。然后,设计了一个候选人合并方案,以同时汇总重复的候选人,并收集围绕合并的质心的背景,以形成实例内核。一旦实例内核可用,就可以通过在实例内核调节的动态卷积来重建实例掩码。整个管道是通过动态内核网络(DKNET)实例化的。结果表明,DKNET的表现优于ScannETV2和S3DIS数据集的艺术状态,并具有更好的实例本地化。可用代码:https://github.com/w1zheng/dknet。
translated by 谷歌翻译
我们考虑两个马尔可夫决策过程(MDP)之间的政策转移问题。我们基于现有的加强学习理论结果(RL)引入引理,以衡量两个任意MDP之间的相对性,这是在不同的政策和环境动态上定义的任何两个累积预期收益之间的差异。基于此引理,我们提出了两种称为相对策略优化(RPO)和相对过渡优化(RTO)的新算法,它们可以分别提供快速的策略转移和动态建模。 RPO使用相对策略梯度更新策略,以转移在一个环境中评估的策略以最大化另一个环境的返回,而RTO使用相对过渡梯度更新参数化的动态模型(如果存在),以减少差异两个环境。然后,集成两种算法提供完整的算法相对策略转换优化(RPTO),其中策略同时与两个环境进行交互,从而使两个环境中的数据收集,策略和过渡更新以一个封闭的循环完成,以形成一个封闭式循环政策转移的原则学习框架。我们通过通过变体动态创建策略转移问题来证明RPTO在OpenAI Gym的经典控制任务中的有效性。
translated by 谷歌翻译
合作多代理增强学习(CMARL)具有许多真实的应用程序,但是在部署时,现有CMARL算法培训的政策不够强大。关于RL系统的对抗攻击也存在许多方法,这意味着RL系统可能会遭受对抗攻击,但大多数都集中在单个代理RL上。在本文中,我们在CMARL系统上提出了一个\ textit {稀疏对抗攻击}。我们将(MA)RL与正规化一起训练攻击政策。我们的实验表明,当当前CMARL算法训练的政策可以在团队中只有一名或几个代理(例如,25个中的1个或5个中的1个)在几个时间段攻击时(例如,攻击3的总数3或5)可以获得较差的性能40个时间段)。
translated by 谷歌翻译
数以百万计的流浪动物在街头遭受痛苦或每天在世界各地的庇护所中被安乐死。为了更好地采用流浪动物,对流浪动物的爪子(可爱)进行评分非常重要,但是评估动物的爪子是非常劳动密集型的事情。因此,开发一种分数动物的算法引起了迫切关注的兴趣。但是,Kaggle中的数据集不仅具有图像,还具有描述图像的元数据。大多数方法基本上都集中在近年来最先进的图像回归方法上,但是没有很好的方法来处理图像的元数据。为了应对上述挑战,本文提出了一个称为PETS-SWINF的图像回归模型,该模型考虑了图像的元数据。我们的结果基于Kaggle竞争的数据集“ Petfinder.my”,表明PETS-SWINF比仅基于基于的图像模型具有优势。我们的结果表明,测试数据集上提出的模型的RMSE丢失为17.71876,但没有元数据为17.76449。提出的方法的优点是,Pets-Swinf可以考虑元数据的低阶和高阶特征,并自适应地调整图像模型和元数据模型的权重。表现很有希望,因为我们的Leadboard得分在3545支球队(金牌)中排名第15,目前在2021 Kaggle比赛中参加了挑战“ Petfinder.my”。
translated by 谷歌翻译