许多现有的景点（ROA）分析工具难以解决具有大规模神经网络（NN）政策和/或高维感测模式的反馈系统，如相机。在本文中，我们定制了在对冲学习界中开发的预计梯度下降（PGD）攻击方法作为大型非线性系统的通用ROA分析工具和基于端到端的感知的控制。我们表明ROA分析可以近似为约束的最大化问题，其目标是找到最坏情况的最坏情况初始条件最多。然后我们提出了两个基于PGD的迭代方法，可用于解决所得的受限最大化问题。我们的分析不是基于Lyapunov理论，因此需要问题结构的最低信息。在基于模型的设置中，我们示出了可以使用反向传播有效地执行PGD更新。在无模型设置（与基于感知的控制的ROA分析更相关）中，我们提出了一个有限差异的PGD估计，这是一般的，只需要一个黑盒模拟器来产生闭环系统的轨迹给予任何初始状态。我们在具有大规模NN政策和高维图像观测的几个数字示例下展示了我们分析工具的可扩展性和一般性。我们认为，我们所提出的分析是进一步了解大规模非线性系统的闭环稳定性和基于感知的控制的有意义的初步步骤。

基于政策的强化学习（RL）最近的经验成功，有一项研究趋势，研究了基于政策的RL方法对标准控制基准问题的研究。在本文中，我们研究了基于政策的RL方法的有效性在重要的强大控制问题上，即$ \ mu $综合。我们在强大的对策RL和$ \ mu $综合之间建立连接，并开发出众所周知的$ DK $ antication的无模型版本，用于解决静态$ d $-scaling的状态反馈$ \ mu $ synthesis。在所提出的算法中，$ k $步骤通过将最近开发的双循环对冲RL方法作为子程序来模仿经典的中央路径算法，$ D $步骤基于无模型有限差分近似。还提出了广泛的数值研究以展示我们提出的无模型算法的效用。我们的研究揭示了对抗对抗和鲁棒控制之间的联系。

We introduce a new tool for stochastic convex optimization (SCO): a Reweighted Stochastic Query (ReSQue) estimator for the gradient of a function convolved with a (Gaussian) probability density. Combining ReSQue with recent advances in ball oracle acceleration [CJJJLST20, ACJJS21], we develop algorithms achieving state-of-the-art complexities for SCO in parallel and private settings. For a SCO objective constrained to the unit ball in $\mathbb{R}^d$, we obtain the following results (up to polylogarithmic factors). We give a parallel algorithm obtaining optimization error $\epsilon_{\text{opt}}$ with $d^{1/3}\epsilon_{\text{opt}}^{-2/3}$ gradient oracle query depth and $d^{1/3}\epsilon_{\text{opt}}^{-2/3} + \epsilon_{\text{opt}}^{-2}$ gradient queries in total, assuming access to a bounded-variance stochastic gradient estimator. For $\epsilon_{\text{opt}} \in [d^{-1}, d^{-1/4}]$, our algorithm matches the state-of-the-art oracle depth of [BJLLS19] while maintaining the optimal total work of stochastic gradient descent. We give an $(\epsilon_{\text{dp}}, \delta)$-differentially private algorithm which, given $n$ samples of Lipschitz loss functions, obtains near-optimal optimization error and makes $\min(n, n^2\epsilon_{\text{dp}}^2 d^{-1}) + \min(n^{4/3}\epsilon_{\text{dp}}^{1/3}, (nd)^{2/3}\epsilon_{\text{dp}}^{-1})$ queries to the gradients of these functions. In the regime $d \le n \epsilon_{\text{dp}}^{2}$, where privacy comes at no cost in terms of the optimal loss up to constants, our algorithm uses $n + (nd)^{2/3}\epsilon_{\text{dp}}^{-1}$ queries and improves recent advancements of [KLL21, AFKT21]. In the moderately low-dimensional setting $d \le \sqrt n \epsilon_{\text{dp}}^{3/2}$, our query complexity is near-linear.

Despite the recent progress in language generation models, their outputs may not always meet user expectations. In this work, we study whether informational feedback in natural language can be leveraged to improve generation quality and user preference alignment. To this end, we consider factual consistency in summarization, the quality that the summary should only contain information supported by the input documents, for user preference alignment. We collect a high-quality dataset, DeFacto, containing human demonstrations and informational feedback in natural language consisting of corrective instructions, edited summaries, and explanations with respect to the factual consistency of the summary. Using our dataset, we study two natural language generation tasks: 1) editing a summary using the human feedback, and 2) generating human feedback from the original summary. Using the two tasks, we further evaluate if models can automatically correct factual inconsistencies in generated summaries. We show that the human-edited summaries we collected are more factually consistent, and pre-trained language models can leverage our dataset to improve the factual consistency of original system-generated summaries in our proposed generation tasks. We make the DeFacto dataset publicly available at https://github.com/microsoft/DeFacto.

Complex and contact-rich robotic manipulation tasks, particularly those that involve multi-fingered hands and underactuated object manipulation, present a significant challenge to any control method. Methods based on reinforcement learning offer an appealing choice for such settings, as they can enable robots to learn to delicately balance contact forces and dexterously reposition objects without strong modeling assumptions. However, running reinforcement learning on real-world dexterous manipulation systems often requires significant manual engineering. This negates the benefits of autonomous data collection and ease of use that reinforcement learning should in principle provide. In this paper, we describe a system for vision-based dexterous manipulation that provides a "programming-free" approach for users to define new tasks and enable robots with complex multi-fingered hands to learn to perform them through interaction. The core principle underlying our system is that, in a vision-based setting, users should be able to provide high-level intermediate supervision that circumvents challenges in teleoperation or kinesthetic teaching which allow a robot to not only learn a task efficiently but also to autonomously practice. Our system includes a framework for users to define a final task and intermediate sub-tasks with image examples, a reinforcement learning procedure that learns the task autonomously without interventions, and experimental results with a four-finger robotic hand learning multi-stage object manipulation tasks directly in the real world, without simulation, manual modeling, or reward engineering.

Free-text rationales (FTRs) follow how humans communicate by explaining reasoning processes via natural language. A number of recent works have studied how to improve language model (LM) generalization by using FTRs to teach LMs the correct reasoning processes behind correct task outputs. These prior works aim to learn from FTRs by appending them to the LM input or target output, but this may introduce an input distribution shift or conflict with the task objective, respectively. We propose KNIFE, which distills FTR knowledge from an FTR-augmented teacher LM (takes both task input and FTR) to a student LM (takes only task input), which is used for inference. Crucially, the teacher LM's forward computation has a bottleneck stage in which all of its FTR states are masked out, which pushes knowledge from the FTR states into the task input/output states. Then, FTR knowledge is distilled to the student LM by training its task input/output states to align with the teacher LM's. On two question answering datasets, we show that KNIFE significantly outperforms existing FTR learning methods, in both fully-supervised and low-resource settings.

As information extraction (IE) systems have grown more capable at whole-document extraction, the classic task of \emph{template filling} has seen renewed interest as a benchmark for evaluating them. In this position paper, we call into question the suitability of template filling for this purpose. We argue that the task demands definitive answers to thorny questions of \emph{event individuation} -- the problem of distinguishing distinct events -- about which even human experts disagree. We show through annotation studies and error analysis that this raises concerns about the usefulness of template filling evaluation metrics, the quality of datasets for the task, and the ability of models to learn it. Finally, we consider possible solutions.

Targeted syntactic evaluations of language models ask whether models show stable preferences for syntactically acceptable content over minimal-pair unacceptable inputs. Most targeted syntactic evaluation datasets ask models to make these judgements with just a single context-free sentence as input. This does not match language models' training regime, in which input sentences are always highly contextualized by the surrounding corpus. This mismatch raises an important question: how robust are models' syntactic judgements in different contexts? In this paper, we investigate the stability of language models' performance on targeted syntactic evaluations as we vary properties of the input context: the length of the context, the types of syntactic phenomena it contains, and whether or not there are violations of grammaticality. We find that model judgements are generally robust when placed in randomly sampled linguistic contexts. However, they are substantially unstable for contexts containing syntactic structures matching those in the critical test content. Among all tested models (GPT-2 and five variants of OPT), we significantly improve models' judgements by providing contexts with matching syntactic structures, and conversely significantly worsen them using unacceptable contexts with matching but violated syntactic structures. This effect is amplified by the length of the context, except for unrelated inputs. We show that these changes in model performance are not explainable by simple features matching the context and the test inputs, such as lexical overlap and dependency overlap. This sensitivity to highly specific syntactic features of the context can only be explained by the models' implicit in-context learning abilities.

Autonomous vehicles are suited for continuous area patrolling problems. However, finding an optimal patrolling strategy can be challenging for many reasons. Firstly, patrolling environments are often complex and can include unknown and evolving environmental factors. Secondly, autonomous vehicles can have failures or hardware constraints such as limited battery lives. Importantly, patrolling large areas often requires multiple agents that need to collectively coordinate their actions. In this work, we consider these limitations and propose an approach based on a distributed, model-free deep reinforcement learning based multi-agent patrolling strategy. In this approach, agents make decisions locally based on their own environmental observations and on shared information. In addition, agents are trained to automatically recharge themselves when required to support continuous collective patrolling. A homogeneous multi-agent architecture is proposed, where all patrolling agents have an identical policy. This architecture provides a robust patrolling system that can tolerate agent failures and allow supplementary agents to be added to replace failed agents or to increase the overall patrol performance. This performance is validated through experiments from multiple perspectives, including the overall patrol performance, the efficiency of the battery recharging strategy, the overall robustness of the system, and the agents' ability to adapt to environment dynamics.

Large Neighborhood Search (LNS) is a popular heuristic algorithm for solving combinatorial optimization problems (COP). It starts with an initial solution to the problem and iteratively improves it by searching a large neighborhood around the current best solution. LNS relies on heuristics to select neighborhoods to search in. In this paper, we focus on designing effective and efficient heuristics in LNS for integer linear programs (ILP) since a wide range of COPs can be represented as ILPs. Local Branching (LB) is a heuristic that selects the neighborhood that leads to the largest improvement over the current solution in each iteration of LNS. LB is often slow since it needs to solve an ILP of the same size as input. Our proposed heuristics, LB-RELAX and its variants, use the linear programming relaxation of LB to select neighborhoods. Empirically, LB-RELAX and its variants compute as effective neighborhoods as LB but run faster. They achieve state-of-the-art anytime performance on several ILP benchmarks.