Search

Many applications of cyber-physical systems require real-time communication: manufacturing, automotive, etc. Recent Ethernet standards for Time Sensitive Networking (TSN) offer time-triggered scheduling in order to guarantee low latency and jitter bounds. This requires precise frame transmission planning, which becomes especially hard when dealing with many streams, large networks, and dynamically changing communications. A very promising approach uses conflict graphs, modeling conflicting transmission configurations. Since the creation of conflict graphs is the bottleneck in these approaches, we provide an improvement to the conflict graph creation. We present a randomized selection process that reduces the overall size of the graph in half and three heuristics to improve the scheduling success. In our evaluations we show substantial improvements in the graph creation speed and the scheduling success compared to existing work, updating existing schedules in fractions of a second. Additionally, offline planning of 9000 streams was performed successfully within minutes.

While originally developed for novel view synthesis, Neural Radiance Fields (NeRFs) have recently emerged as an alternative to multi-view stereo (MVS). Triggered by a manifold of research activities, promising results have been gained especially for texture-less, transparent, and reflecting surfaces, while such scenarios remain challenging for traditional MVS-based approaches. However, most of these investigations focus on close-range scenarios, with studies for airborne scenarios still missing. For this task, NeRFs face potential difficulties at areas of low image redundancy and weak data evidence, as often found in street canyons, facades or building shadows. Furthermore, training such networks is computationally expensive. Thus, the aim of our work is twofold: First, we investigate the applicability of NeRFs for aerial image blocks representing different characteristics like nadir-only, oblique and high-resolution imagery. Second, during these investigations we demonstrate the benefit of integrating depth priors from tie-point measures, which are provided during presupposed Bundle Block Adjustment. Our work is based on the state-of-the-art framework VolSDF, which models 3D scenes by signed distance functions (SDFs), since this is more applicable for surface reconstruction compared to the standard volumetric representation in vanilla NeRFs. For evaluation, the NeRF-based reconstructions are compared to results of a publicly available benchmark dataset for airborne images.

Goal misalignment, reward sparsity and difficult credit assignment are only a few of the many issues that make it difficult for deep reinforcement learning (RL) agents to learn optimal policies. Unfortunately, the black-box nature of deep neural networks impedes the inclusion of domain experts for inspecting the model and revising suboptimal policies. To this end, we introduce *Successive Concept Bottleneck Agents* (SCoBots), that integrate consecutive concept bottleneck (CB) layers. In contrast to current CB models, SCoBots do not just represent concepts as properties of individual objects, but also as relations between objects which is crucial for many RL tasks. Our experimental results provide evidence of SCoBots' competitive performances, but also of their potential for domain experts to understand and regularize their behavior. Among other things, SCoBots enabled us to identify a previously unknown misalignment problem in the iconic video game, Pong, and resolve it. Overall, SCoBots thus result in more human-aligned RL agents. Our code is available at https://github.com/k4ntz/SCoBots ., Comment: 10 of main text, lots of appendices

In distributed Complex Event Processing (CEP) applications with high load but limited resources, bottleneck operators in the operator graph can significantly slow down processing of event streams, thus compelling the need to shed load. A high-quality load shedding strategy that resolves the bottleneck with high output quality evaluates each event's importance with regards to the application's final output and drops less important events from the event stream for the benefit of important ones. So far, no solution has been proposed that is able to permit good load shedding in distributed, multi-operator CEP applications. On one hand, shedding strategies have been proposed for single-operator CEP applications that can measure an event's importance immediately at the bottleneck operator, only, and thereby ignore the effect of other streams in the application on an event's importance. On the other hand, shedding strategies have been proposed for applications with multiple operators from the area of stream processing that provide a fixed selectivity which is not given in the conditional CEP operators. We, therefore, propose a load-shedding solution for distributed CEP applications that maximizes the application's final output and ensures timely processing of important events by using a set of CEP-tailored selectivity functions and a linear program, which is an abstraction of the CEP application. Moreover, our solution ensures a quality optimal shedder configuration even in the presence of dynamically changing conditions. With the help of extensive evaluations on both synthetic and real data, we show that our solution successfully resolves overload at bottleneck operators and at the same time maximizes the quality of the application's output.

Overload situations, in the presence of resource limitations, in complex event processing (CEP) systems are typically handled using load shedding to maintain a given latency bound. However, load shedding might negatively impact the quality of results (QoR). To minimize the shedding impact on QoR, CEP researchers propose shedding approaches that drop events/internal state with the lowest importances/utilities. In both black-box and white-box shedding approaches, different features are used to predict these utilities. In this work, we propose a novel black-box shedding approach that uses a new set of features to drop events from the input event stream to maintain a given latency bound. Our approach uses a probabilistic model to predict these event utilities. Moreover, our approach uses Zobrist hashing and well-known machine learning models, e.g., decision trees and random forests, to handle the predicted event utilities. Through extensive evaluations on several synthetic and two real-world datasets and a representative set of CEP queries, we show that, in the majority of cases, our load shedding approach outperforms state-of-the-art black-box load shedding approaches, w.r.t. QoR.

Cyber-Physical Systems (CPSs) play a central role in the behavior of a wide range of autonomous physical systems such as medical devices, autonomous vehicles, and smart homes, many of which are safety-critical. CPSs are often specified iteratively as a sequence of models at different levels that can be tested via simulation systems at early stages of their development cycle. One such model is a hybrid automaton; these are used frequently for CPS applications and have the advantage of encapsulating both continuous and discrete CPS behaviors. When testing CPSs, engineers can take advantage of these models to generate test cases that target both types of these behaviors. Moreover, since these models are constructed early in the development process for CPSs, they allow test cases to be generated early in that process for those CPSs, even before simulation models of the CPSs have been designed. One challenge when testing CPSs is that these systems may operate differently even under an identically applied test scenario. In such cases, we cannot employ test oracles that use predetermined deterministic behaviors; instead, test oracles should consider sets of desired behaviors in order to determine whether the CPS has behaved appropriately. In this paper we present a test case generation technique, HYTEST, that generates test cases based on hybrid models, accompanied by appropriate test oracles, for use in testing CPSs early in their development cycle. To evaluate the effectiveness and efficiency of HYTEST, we conducted an empirical study in which we applied the technique to several CPSs and measured its ability to detect faults in those CPSs and the amount of time required to perform the testing process. The results of the study show that HYTEST was able to detect faults more effectively and efficiently than the baseline techniques we compare it to., Comment: 15 pages, Submitted to IEEE Transaction on Software Engineering on 9/14/2023

We present forecasts on cosmological parameters for a CMB-HD survey. For a $\Lambda$CDM + $N_{eff}$ + $\sum m_\nu$ model, we find $\sigma(n_s) = 0.0013$ and $\sigma(N_{eff}) = 0.014$ using CMB and CMB lensing multipoles in the range of $\ell \in [30, 20000]$, after adding anticipated residual foregrounds, delensing the acoustic peaks, and adding DESI BAO data. This is about a factor of two improvement in ability to probe inflation via $n_s$ compared to precursor CMB surveys. The $N_{eff}$ constraint can rule out light thermal particles back to the end of inflation with 95% CL; for example, it can rule out the QCD axion in a model-independent way assuming the Universe's reheating temperature was high enough that the axion thermalized. We find that delensing the acoustic peaks and adding DESI BAO tightens parameter constraints. We also find that baryonic effects can bias parameters if not marginalized over, and that uncertainties in baryonic effects can increase parameter error bars; however, the latter can be mitigated by including information about baryonic effects from kinetic and thermal Sunyaev-Zel'dovich measurements by CMB-HD. The CMB-HD likelihood and Fisher estimation codes used here are publicly available; the likelihood is integrated with Cobaya to facilitate parameter forecasting., Comment: 29 pages, 18 figures, 10 tables. The mock CMB-HD likelihood and Fisher estimation codes are public at https://github.com/CMB-HD/hdlike and https://github.com/CMB-HD/hdfisher ; v3: a few clarifying sentences added; version matches that accepted by PRD

Real-time video analytics typically require video frames to be processed by a query to identify objects or activities of interest while adhering to an end-to-end frame processing latency constraint. Such applications impose a continuous and heavy load on backend compute and network infrastructure because of the need to stream and process all video frames. Video data has inherent redundancy and does not always contain an object of interest for a given query. We leverage this property of video streams to propose a lightweight Load Shedder that can be deployed on edge servers or on inexpensive edge devices co-located with cameras and drop uninteresting video frames. The proposed Load Shedder uses pixel-level color-based features to calculate a utility score for each ingress video frame, which represents the frame's utility toward the query at hand. The Load Shedder uses a minimum utility threshold to select interesting frames to send for query processing. Dropping unnecessary frames enables the video analytics query in the backend to meet the end-to-end latency constraint with fewer compute and network resources. To guarantee a bounded end-to-end latency at runtime, we introduce a control loop that monitors the backend load for the given query and dynamically adjusts the utility threshold. Performance evaluations show that the proposed Load Shedder selects a large portion of frames containing each object of interest while meeting the end-to-end frame processing latency constraint. Furthermore, the Load Shedder does not impose a significant latency overhead when running on edge devices with modest compute resources., Comment: This work was supported by the German Research Foundation (DFG) under the research grant "PRECEPT II" (RO 1086/19-2 and BH 154/1-2)

Due to the growing complexity of modern data centers, failures are not uncommon any more. Therefore, fault tolerance mechanisms play a vital role in fulfilling the availability requirements. Multiple availability models have been proposed to assess compute systems, among which Bayesian network models have gained popularity in industry and research due to its powerful modeling formalism. In particular, this work focuses on assessing the availability of redundant and replicated cloud computing services with Bayesian networks. So far, research on availability has only focused on modeling either infrastructure or communication failures in Bayesian networks, but have not considered both simultaneously. This work addresses practical modeling challenges of assessing the availability of large-scale redundant and replicated services with Bayesian networks, including cascading and common-cause failures from the surrounding infrastructure and communication network. In order to ease the modeling task, this paper introduces a high-level modeling formalism to build such a Bayesian network automatically. Performance evaluations demonstrate the feasibility of the presented Bayesian network approach to assess the availability of large-scale redundant and replicated services. This model is not only applicable in the domain of cloud computing it can also be applied for general cases of local and geo-distributed systems., Comment: 16 pages, 12 figures, journal

Deterministic real-time communication with bounded delay is an essential requirement for many safety-critical cyber-physical systems, and has received much attention from major standardization bodies such as IEEE and IETF. In particular, Ethernet technology has been extended by time-triggered scheduling mechanisms in standards like TTEthernet and Time-Sensitive Networking. Although the scheduling mechanisms have become part of standards, the traffic planning algorithms to create time-triggered schedules are still an open and challenging research question due to the problem's high complexity. In particular, so-called plug-and-produce scenarios require the ability to extend schedules on the fly within seconds. The need for scalable scheduling and routing algorithms is further supported by large-scale distributed real-time systems like smart energy grids with tight communication requirements. In this paper, we tackle this challenge by proposing two novel algorithms called Hierarchical Heuristic Scheduling (H2S) and Cost-Efficient Lazy Forwarding Scheduling (CELF) to calculate time-triggered schedules for TTEthernet. H2S and CELF are highly efficient and scalable, calculating schedules for more than 45,000 streams on random networks with 1,000 bridges as well as a realistic energy grid network within sub-seconds to seconds.

In this article, a general model for 1D thermal wave interference is derived for multi-layered coating systems on a thermally thick substrate using the same principles as for the well established one-layered and two-layered coating cases. Using the lock-in thermography principle, an illumination source modulates the surface of those systems periodically by a planar, sinusoidal wave form with a fixed frequency. The coating systems absorb the optical energy on its surface and convert it into thermal energy, resulting in the propagation of a spatially and temporally periodic thermal wave with the same frequency. These thermal waves, originating at the surface, are reflected and transmitted at each interface leading to infinitely many wave trains that need to be tracked in order to formulate the final surface temperature as a superposition of all these waves. The heat transfer inside the object depends on the layer thickness of each coating, but also on the thermal properties of each layer material. The goal is to have a mathematical and physical model which describes the phase angle data measured by an infrared camera. Having these data, the main objective of this paper is to determine the thickness of each coating layer. In practice, the thermal properties of the layers usually are unknown, which makes the process even more difficult. For that reason, this article presents a concept to determine the thermal properties in advance., Comment: 16 pages, 8 figures

Recent breakthroughs in the development of agents to solve challenging sequential decision making problems such as Go, StarCraft, or DOTA, have relied on both simulated environments and large-scale datasets. However, progress on this research has been hindered by the scarcity of open-sourced datasets and the prohibitive computational cost to work with them. Here we present the NetHack Learning Dataset (NLD), a large and highly-scalable dataset of trajectories from the popular game of NetHack, which is both extremely challenging for current methods and very fast to run. NLD consists of three parts: 10 billion state transitions from 1.5 million human trajectories collected on the NAO public NetHack server from 2009 to 2020; 3 billion state-action-score transitions from 100,000 trajectories collected from the symbolic bot winner of the NetHack Challenge 2021; and, accompanying code for users to record, load and stream any collection of such trajectories in a highly compressed form. We evaluate a wide range of existing algorithms including online and offline RL, as well as learning from demonstrations, showing that significant research advances are needed to fully leverage large-scale datasets for challenging sequential decision making tasks., Comment: 9 pages, published in the Proceedings of the 36th Conference on Neural Information Processing Systems (NeurIPS 2022) Track on Datasets and Benchmarks. New links to hosting location. Revised results, same conclusions

In this report, we summarize the takeaways from the first NeurIPS 2021 NetHack Challenge. Participants were tasked with developing a program or agent that can win (i.e., 'ascend' in) the popular dungeon-crawler game of NetHack by interacting with the NetHack Learning Environment (NLE), a scalable, procedurally generated, and challenging Gym environment for reinforcement learning (RL). The challenge showcased community-driven progress in AI with many diverse approaches significantly beating the previously best results on NetHack. Furthermore, it served as a direct comparison between neural (e.g., deep RL) and symbolic AI, as well as hybrid systems, demonstrating that on NetHack symbolic bots currently outperform deep RL by a large margin. Lastly, no agent got close to winning the game, illustrating NetHack's suitability as a long-term benchmark for AI research., Comment: Under review at PMLR for the NeuRIPS 2021 Competition Workshop Track, 10 pages + 10 in appendices

Positive behavioral health outcomes are affected by a variety of factors including social, biological, economic and physical environments. These social determinants of mental health (SDOMH) affect not only individual children but the communities and neighborhoods in which children are expected to live, play and thrive. The State must ensure appropriate supports are available from pre-natal to age 21, so that children can grow in environments that have minimized the root causes of common illnesses like anxiety and depression, while equipping children with the ability to recognize and regulate their emotions and maintain healthy relationships. The following agenda provides a roadmap for state leaders on the components needed to establish a successful children's behavioral health system including community investments, individual service supports, community education, and data-sharing. The examples provided are not exhaustive but are key recommendations for a delivery system that develops and supports a well-trained workforce, educates communities, identifies kids in need, and provides appropriate treatment and support. [Additional funding was provided by the Well Being Trust.]

Self-supervised pre-training of large-scale transformer models on text corpora followed by finetuning has achieved state-of-the-art on a number of natural language processing tasks. Recently, Lu et al. (2021, arXiv:2103.05247) claimed that frozen pretrained transformers (FPTs) match or outperform training from scratch as well as unfrozen (fine-tuned) pretrained transformers in a set of transfer tasks to other modalities. In our work, we find that this result is, in fact, an artifact of not tuning the learning rates. After carefully redesigning the empirical setup, we find that when tuning learning rates properly, pretrained transformers do outperform or match training from scratch in all of our tasks, but only as long as the entire model is finetuned. Thus, while transfer from pretrained language models to other modalities does indeed provide gains and hints at exciting possibilities for future work, properly tuning hyperparameters is important for arriving at robust findings., Comment: Accepted to ICML 2021 Workshop: Self-Supervised Learning for Reasoning and Perception

Many networked applications, e.g., in the domain of cyber-physical systems, require strict service guarantees, usually in the form of jitter and latency bounds, for time-triggered traffic flows. It is a notoriously hard problem to compute a network-wide traffic plan that satisfies these requirements, and dynamic changes in the flow set add even more challenges. Existing traffic-planning methods are ill-suited for dynamic scenarios because they either suffer from high computational cost, can result in low network utilization, or provide no explicit guarantees when transitioning to a new traffic plan that incorporates new flows. Therefore, we present a novel approach for dynamic traffic planning of time-triggered flows. Our conflict-graph based modeling of the traffic planning problem allows to reconfigure active flows to increase the network utilization, while also providing per-flow QoS guarantees during the transition to the new traffic plan. Additionally, we introduce a novel heuristic for computing the new traffic plans. Evaluations of our prototypical implementation show that we can efficiently compute new traffic plans in scenarios with hundreds of active flows for a wide range of scenarios.

Mathematical models describing the behavior of viscoelastic materials are often based on evolution equations that measure the change in stress depending on its material parameters such as stiffness, viscosity or relaxation time. In this article, we introduce a Maxwell-based rheological model, define the associated forward operator and the inverse problem in order to determine the number of Maxwell elements and the material parameters of the underlying viscoelastic material. We perform a relaxation experiment by applying a strain to the material and measure the generated stress. Since the measured data varies with the number of Maxwell elements, the forward operator of the underlying inverse problem depends on parts of the solution. By introducing assumptions on the relaxation times, we propose a clustering algorithm to resolve this problem. We provide the calculations that are necessary for the minimization process and conclude with numerical results by investigating unperturbed as well as noisy data. We present different reconstruction approaches based on minimizing a least squares functional. Furthermore, we look at individual stress components to analyze different displacement rates. Finally, we study reconstructions with shortened data sets to obtain assertions on how long experiments have to be performed to identify conclusive material parameters., Comment: 23 pages, 11 figures, 6 tables