基础模型正在成为主要的深度学习技术。由于模型参数和训练数据集的大规模，预处理基础模型始终耗时。除了计算密集型外，培训过程还非常密集和沟通密集。这些功能使得需要应用3D并行性，该平行性整合数据并行性，管道模型并行性和张量模型并行性，以实现高训练效率。为了实现这一目标，开发了一些自定义软件框架，例如Megatron-LM和DeepSpeed。但是，当前的3D平行框架仍然符合两个问题：i）它们对模型开发人员不透明，这些开发人员需要手动修改模型以并行化培训。 ii）它们对计算，GPU存储器和网络带宽的利用不足。我们提出了Merak，这是一个自动化的3D并行性深度学习培训框架，并具有高度资源利用。 Merak会自动使用自动模型分区仪部署，该分区仪在模型的代理表示上使用图形sharding算法。 Merak还提出了非侵入性的API，用于通过最小的代码修改来扩展基础模型培训。此外，我们在Merak设计了高性能的3D平行运行时引擎。它使用多种技术来利用可用的培训资源，包括移动的关键路径管道时间表，该计划带来了更高的计算利用率，阶段感知的重新计算，可利用空闲工作者的记忆以及子额定张量的模型并行性，这些模型并联与通信和计算重叠。 64 GPU的实验显示，Merak可以加快在最新的3D平行性框架上，具有1.5、2.5、8.3和20亿的模型框架，最高可达1.42x，1.39x，1.43x和1.61 x分别。

有限的GPU记忆资源阻碍了深度神经网络的进一步发展。因此，高度要求GPU内存资源的优化。通常应用交换和重新计算，以更好地利用GPU记忆。但是，作为一个新兴领域，仍然存在一些挑战：1）静态和动态方法的重新计算效率受到限制。 2）交换需要手动卸载参数，这会产生巨大的时间成本。 3）没有这种动态和细粒的方法，涉及张量与当今的张量重新组件一起交换。为了纠正上述问题，我们提出了一个名为Delta（动态张量卸载和重新组件）的新型调度程序经理。据我们所知，我们是第一个在没有用户监督的情况下进行张量交换和张量重新组合的合理的动态运行时间调度程序。在Delta中，我们提出了一种过滤器算法，以选择要从GPU内存中释放出来的最佳张量，并提出导演算法，以选择每个张量的适当动作。此外，故意考虑预取和重叠以克服交换和重新计算张量引起的时间成本。实验结果表明，DELTA不仅节省了40％-70％的GPU记忆，从而超过了最新方法，而且还获得了可比的收敛结果，并获得了可接受的时间延迟。此外，与基准相比，当训练Resnet-101训练Resnet-101时，Delta在训练Resnet-50和2.25 $ \ times $时获得2.04 $ \ times $最大批量。此外，我们实验中的交换成本和重新计算成本之间的比较表明，在张量交换和张量重新计算上制定合理的动态调度程序的重要性，这在某些相关工作中反驳了交换应该是第一个也是最好的选择。

分布式数据并行训练已被广泛用于深神经网络（DNN）模型。尽管当前的深度学习（DL）框架对于图像分类模型（例如图像分类模型）的密集模型很好地扩展了，但我们发现这些DL框架对于具有高度稀疏嵌入表的稀疏模型（NLP）模型（NLP）模型（NLP）模型具有相对较低的可扩展性。大多数现有作品忽略了模型参数的稀疏性，因此遭受了重要但不必要的沟通开销。在本文中，我们提出了Ablace，这是一个有效的沟通框架，以加快稀疏模型分布式培训的通信。 Embrace引入了稀疏感知的混合通信，将AlltoAll和模型并行置于数据并行训练中，以减少高度稀疏参数的交流开销。为了有效地重叠稀疏的通信与后向前和前向计算，采用进一步设计的2D通信调度方法，该方法优化了模型计算过程，放松嵌入式的依赖性，并计划以优先级的排队来安排每个嵌入行的稀疏通信。我们已经基于Pytorch和Horovod实施了Embrace的原型，并通过四个代表性的NLP模型进行了全面的评估。实验结果表明，与最先进的分布式训练基线相比，Embrace的速度高达2.41倍。

分布式随机梯度下降（SGD）方法已广泛应用于大型深度学习，梯度集体方法至关重要，以确保分布式深度学习系统的培训可扩展性。已广泛采用分布式SGD过程广泛采用诸如解释的集体通信，以减少通信时间。但是，allreduce会引发大带宽资源，而在许多情况下大多数梯度稀疏，因为许多梯度值是零，并且应该有效地压缩以用于节省带宽。为了减少稀疏梯度通信开销，我们提出了一种稀疏的剪影减速器（S2减速器），这是一种具有收敛保证的新型草图的稀疏梯度聚合方法。 S2减速机仅通过Count-Sketch和Bitmap压缩非零梯度来降低通信成本，并实现有效的已有SGD培训的有效恢复运算符。我们在五种培训模型中对四种最先进的方法进行广泛的评估。我们的结果表明，S2减速机收敛到相同的准确性，降低了81 \％稀疏通信开销，与最先进的方法相比，实现了1.8 $ \ times $ Speedup。

With the fast development of big data, it has been easier than before to learn the optimal decision rule by updating the decision rule recursively and making online decisions. We study the online statistical inference of model parameters in a contextual bandit framework of sequential decision-making. We propose a general framework for online and adaptive data collection environment that can update decision rules via weighted stochastic gradient descent. We allow different weighting schemes of the stochastic gradient and establish the asymptotic normality of the parameter estimator. Our proposed estimator significantly improves the asymptotic efficiency over the previous averaged SGD approach via inverse probability weights. We also conduct an optimality analysis on the weights in a linear regression setting. We provide a Bahadur representation of the proposed estimator and show that the remainder term in the Bahadur representation entails a slower convergence rate compared to classical SGD due to the adaptive data collection.

Model counting is a fundamental problem which has been influential in many applications, from artificial intelligence to formal verification. Due to the intrinsic hardness of model counting, approximate techniques have been developed to solve real-world instances of model counting. This paper designs a new anytime approach called PartialKC for approximate model counting. The idea is a form of partial knowledge compilation to provide an unbiased estimate of the model count which can converge to the exact count. Our empirical analysis demonstrates that PartialKC achieves significant scalability and accuracy over prior state-of-the-art approximate counters, including satss and STS. Interestingly, the empirical results show that PartialKC reaches convergence for many instances and therefore provides exact model counting performance comparable to state-of-the-art exact counters.

Robots are traditionally bounded by a fixed embodiment during their operational lifetime, which limits their ability to adapt to their surroundings. Co-optimizing control and morphology of a robot, however, is often inefficient due to the complex interplay between the controller and morphology. In this paper, we propose a learning-based control method that can inherently take morphology into consideration such that once the control policy is trained in the simulator, it can be easily deployed to robots with different embodiments in the real world. In particular, we present the Embodiment-aware Transformer (EAT), an architecture that casts this control problem as conditional sequence modeling. EAT outputs the optimal actions by leveraging a causally masked Transformer. By conditioning an autoregressive model on the desired robot embodiment, past states, and actions, our EAT model can generate future actions that best fit the current robot embodiment. Experimental results show that EAT can outperform all other alternatives in embodiment-varying tasks, and succeed in an example of real-world evolution tasks: stepping down a stair through updating the morphology alone. We hope that EAT will inspire a new push toward real-world evolution across many domains, where algorithms like EAT can blaze a trail by bridging the field of evolutionary robotics and big data sequence modeling.

Persuasion modeling is a key building block for conversational agents. Existing works in this direction are limited to analyzing textual dialogue corpus. We argue that visual signals also play an important role in understanding human persuasive behaviors. In this paper, we introduce the first multimodal dataset for modeling persuasion behaviors. Our dataset includes 199 dialogue transcriptions and videos captured in a multi-player social deduction game setting, 26,647 utterance level annotations of persuasion strategy, and game level annotations of deduction game outcomes. We provide extensive experiments to show how dialogue context and visual signals benefit persuasion strategy prediction. We also explore the generalization ability of language models for persuasion modeling and the role of persuasion strategies in predicting social deduction game outcomes. Our dataset, code, and models can be found at https://persuasion-deductiongame.socialai-data.org.

Deep reinforcement learning has recently emerged as an appealing alternative for legged locomotion over multiple terrains by training a policy in physical simulation and then transferring it to the real world (i.e., sim-to-real transfer). Despite considerable progress, the capacity and scalability of traditional neural networks are still limited, which may hinder their applications in more complex environments. In contrast, the Transformer architecture has shown its superiority in a wide range of large-scale sequence modeling tasks, including natural language processing and decision-making problems. In this paper, we propose Terrain Transformer (TERT), a high-capacity Transformer model for quadrupedal locomotion control on various terrains. Furthermore, to better leverage Transformer in sim-to-real scenarios, we present a novel two-stage training framework consisting of an offline pretraining stage and an online correction stage, which can naturally integrate Transformer with privileged training. Extensive experiments in simulation demonstrate that TERT outperforms state-of-the-art baselines on different terrains in terms of return, energy consumption and control smoothness. In further real-world validation, TERT successfully traverses nine challenging terrains, including sand pit and stair down, which can not be accomplished by strong baselines.

Graphene quantum dots provide a platform for manipulating electron behaviors in two-dimensional (2D) Dirac materials. Most previous works were of the "forward" type in that the objective was to solve various confinement, transport and scattering problems with given structures that can be generated by, e.g., applying an external electrical field. There are applications such as cloaking or superscattering where the challenging problem of inverse design needs to be solved: finding a quantum-dot structure according to certain desired functional characteristics. A brute-force search of the system configuration based directly on the solutions of the Dirac equation is computational infeasible. We articulate a machine-learning approach to addressing the inverse-design problem where artificial neural networks subject to physical constraints are exploited to replace the rigorous Dirac equation solver. In particular, we focus on the problem of designing a quantum dot structure to generate both cloaking and superscattering in terms of the scattering efficiency as a function of the energy. We construct a physical loss function that enables accurate prediction of the scattering characteristics. We demonstrate that, in the regime of Klein tunneling, the scattering efficiency can be designed to vary over two orders of magnitudes, allowing any scattering curve to be generated from a proper combination of the gate potentials. Our physics-based machine-learning approach can be a powerful design tool for 2D Dirac material-based electronics.