Knowledge distillation (KD) has gained a lot of attention in the field of model compression for edge devices thanks to its effectiveness in compressing large powerful networks into smaller lower-capacity models. Online distillation, in which both the teacher and the student are learning collaboratively, has also gained much interest due to its ability to improve on the performance of the networks involved. The Kullback-Leibler (KL) divergence ensures the proper knowledge transfer between the teacher and student. However, most online KD techniques present some bottlenecks under the network capacity gap. By cooperatively and simultaneously training, the models the KL distance becomes incapable of properly minimizing the teacher's and student's distributions. Alongside accuracy, critical edge device applications are in need of well-calibrated compact networks. Confidence calibration provides a sensible way of getting trustworthy predictions. We propose BD-KD: Balancing of Divergences for online Knowledge Distillation. We show that adaptively balancing between the reverse and forward divergences shifts the focus of the training strategy to the compact student network without limiting the teacher network's learning process. We demonstrate that, by performing this balancing design at the level of the student distillation loss, we improve upon both performance accuracy and calibration of the compact student network. We conducted extensive experiments using a variety of network architectures and show improvements on multiple datasets including CIFAR-10, CIFAR-100, Tiny-ImageNet, and ImageNet. We illustrate the effectiveness of our approach through comprehensive comparisons and ablations with current state-of-the-art online and offline KD techniques.

知识蒸馏（KD）是压缩边缘设备深层分类模型的有效工具。但是，KD的表现受教师和学生网络之间较大容量差距的影响。最近的方法已诉诸KD的多个教师助手（TA）设置，该设置依次降低了教师模型的大小，以相对弥合这些模型之间的尺寸差距。本文提出了一种称为“知识蒸馏”课程专家选择的新技术，以有效地增强在容量差距问题下对紧凑型学生的学习。该技术建立在以下假设的基础上：学生网络应逐渐使用分层的教学课程来逐步指导，因为它可以从较低（较高的）容量教师网络中更好地学习（硬）数据样本。具体而言，我们的方法是一种基于TA的逐渐的KD技术，它每个输入图像选择单个教师，该课程是基于通过对图像进行分类的难度驱动的课程的。在这项工作中，我们凭经验验证了我们的假设，并对CIFAR-10，CIFAR-100，CINIC-10和Imagenet数据集进行了严格的实验，并在类似VGG的模型，Resnets和WideresNets架构上显示出提高的准确性。

基于BERT的微调模型在内存，计算和时间上是资源密集的。尽管许多先前的工作旨在通过压缩技术（例如修剪）提高推论效率，但这些作品并未明确解决培训对下游任务的计算挑战。我们介绍了学习者模块和启动，新颖的方法，以利用预训练的语言模型的过度参数化，以获得收敛速度和资源利用率的好处。学习者模块通过微调参数的微调来导航1）有效训练的双结合，以及2）通过确保快速收敛和高度度量得分有效训练。我们在Distilbert上的结果表明，学习者在与基础方面的表现或超过基线。学习者训练7倍的参数比胶水上的最新方法少。在可乐方面，学习者快速调整20％，并且资源利用率显着降低。

Participants in political discourse employ rhetorical strategies -- such as hedging, attributions, or denials -- to display varying degrees of belief commitments to claims proposed by themselves or others. Traditionally, political scientists have studied these epistemic phenomena through labor-intensive manual content analysis. We propose to help automate such work through epistemic stance prediction, drawn from research in computational semantics, to distinguish at the clausal level what is asserted, denied, or only ambivalently suggested by the author or other mentioned entities (belief holders). We first develop a simple RoBERTa-based model for multi-source stance predictions that outperforms more complex state-of-the-art modeling. Then we demonstrate its novel application to political science by conducting a large-scale analysis of the Mass Market Manifestos corpus of U.S. political opinion books, where we characterize trends in cited belief holders -- respected allies and opposed bogeymen -- across U.S. political ideologies.

Deep Reinforcement Learning (RL) agents are susceptible to adversarial noise in their observations that can mislead their policies and decrease their performance. However, an adversary may be interested not only in decreasing the reward, but also in modifying specific temporal logic properties of the policy. This paper presents a metric that measures the exact impact of adversarial attacks against such properties. We use this metric to craft optimal adversarial attacks. Furthermore, we introduce a model checking method that allows us to verify the robustness of RL policies against adversarial attacks. Our empirical analysis confirms (1) the quality of our metric to craft adversarial attacks against temporal logic properties, and (2) that we are able to concisely assess a system's robustness against attacks.

Reinforcement Learning (RL) can enable agents to learn complex tasks. However, it is difficult to interpret the knowledge and reuse it across tasks. Inductive biases can address such issues by explicitly providing generic yet useful decomposition that is otherwise difficult or expensive to learn implicitly. For example, object-centered approaches decompose a high dimensional observation into individual objects. Expanding on this, we utilize an inductive bias for explicit object-centered knowledge separation that provides further decomposition into semantic representations and dynamics knowledge. For this, we introduce a semantic module that predicts an objects' semantic state based on its context. The resulting affordance-like object state can then be used to enrich perceptual object representations. With a minimal setup and an environment that enables puzzle-like tasks, we demonstrate the feasibility and benefits of this approach. Specifically, we compare three different methods of integrating semantic representations into a model-based RL architecture. Our experiments show that the degree of explicitness in knowledge separation correlates with faster learning, better accuracy, better generalization, and better interpretability.

The Age-of-Information (AoI) metric has been widely studied in the theoretical communication networks and queuing systems literature. However, experimental evaluation of its applicability to complex real-world time-sensitive systems is largely lacking. In this work, we develop, implement, and evaluate an AoI-based application layer middleware that enables the customization of WiFi networks to the needs of time-sensitive applications. By controlling the storage and flow of information in the underlying WiFi network, our middleware can: (i) prevent packet collisions; (ii) discard stale packets that are no longer useful; and (iii) dynamically prioritize the transmission of the most relevant information. To demonstrate the benefits of our middleware, we implement a mobility tracking application using a swarm of UAVs communicating with a central controller via WiFi. Our experimental results show that, when compared to WiFi-UDP/WiFi-TCP, the middleware can improve information freshness by a factor of 109x/48x and tracking accuracy by a factor of 4x/6x, respectively. Most importantly, our results also show that the performance gains of our approach increase as the system scales and/or the traffic load increases.

A fundamental procedure in the analysis of massive datasets is the construction of similarity graphs. Such graphs play a key role for many downstream tasks, including clustering, classification, graph learning, and nearest neighbor search. For these tasks, it is critical to build graphs which are sparse yet still representative of the underlying data. The benefits of sparsity are twofold: firstly, constructing dense graphs is infeasible in practice for large datasets, and secondly, the runtime of downstream tasks is directly influenced by the sparsity of the similarity graph. In this work, we present $\textit{Stars}$: a highly scalable method for building extremely sparse graphs via two-hop spanners, which are graphs where similar points are connected by a path of length at most two. Stars can construct two-hop spanners with significantly fewer similarity comparisons, which are a major bottleneck for learning based models where comparisons are expensive to evaluate. Theoretically, we demonstrate that Stars builds a graph in nearly-linear time, where approximate nearest neighbors are contained within two-hop neighborhoods. In practice, we have deployed Stars for multiple data sets allowing for graph building at the $\textit{Tera-Scale}$, i.e., for graphs with tens of trillions of edges. We evaluate the performance of Stars for clustering and graph learning, and demonstrate 10~1000-fold improvements in pairwise similarity comparisons compared to different baselines, and 2~10-fold improvement in running time without quality loss.

Linear-quadratic regulators (LQR) are a well known and widely used tool in control theory for both linear and nonlinear dynamics. For nonlinear problems, an LQR-based controller is usually only locally viable, thus, raising the problem of estimating the region of attraction (ROA). The need for good ROA estimations becomes especially pressing for underactuated systems, as a failure of controls might lead to unsafe and unrecoverable system states. Known approaches based on optimization or sampling, while working well, might be too slow in time critical applications and are hard to verify formally. In this work, we propose a novel approach to estimate the ROA based on the analytic solutions to linear ODEs for the torque limited simple pendulum. In simulation and physical experiments, we compared our approach to a Lyapunov-sampling baseline approach and found that our approach was faster to compute, while yielding ROA estimations of similar phase space area.

This paper is a technical overview of DeepMind and Google's recent work on reinforcement learning for controlling commercial cooling systems. Building on expertise that began with cooling Google's data centers more efficiently, we recently conducted live experiments on two real-world facilities in partnership with Trane Technologies, a building management system provider. These live experiments had a variety of challenges in areas such as evaluation, learning from offline data, and constraint satisfaction. Our paper describes these challenges in the hope that awareness of them will benefit future applied RL work. We also describe the way we adapted our RL system to deal with these challenges, resulting in energy savings of approximately 9% and 13% respectively at the two live experiment sites.