More Like this

1. On Removing Algorithmic Priority Inversion from Mission-critical Machine Inference Pipelines

   https://doi.org/10.1109/RTSS49844.2020.00037  

   Liu, Shengzhong; Yao, Shuochao; Fu, Xinzhe; Tabish, Rohan; Yu, Simon; Bansal, Ayoosh; Yun, Heechul; Sha, Lui; Abdelzaher, Tarek (December 2020, IEEE Real-Time Systems Symposium)

   Chen, Jian-Jia (Ed.)

   The paper discusses algorithmic priority inversion in mission-critical machine inference pipelines used in modern neural-network-based cyber-physical applications and develops a scheduling solution to mitigate its effect. In general, priority inversion occurs in real-time systems when computations that are of lower priority are performed together with or ahead of those that are of higher priority. 1 In current machine intelligence software, significant priority inversion occurs on the path from perception to decision-making, where the execution of underlying neural network algorithms does not differentiate between critical and less critical data. We describe a scheduling framework to resolve this problem and demonstrate that it improves the system's ability to react to critical inputs, while at the same time reducing platform cost. 

   more » « less
2. Dynamic Feature Sharing for Cooperative Perception from Point Clouds

   https://doi.org/10.1109/ITSC57777.2023.10422242  

   Bai, Zhengwei; Wu, Guoyuan; Barth, Matthew J; Liu, Yongkang; Sisbot, Emrah Akin; Oguchi, Kentaro (September 2023, IEEE)

   Perceiving the surrounding environment is critical to enable cooperative driving automation, which is regarded as a transformative solution to improving our transportation system. Cooperative perception, by cooperating information from spatially separated nodes, can innately unlock the bottleneck caused by physical occlusions and has become an important research topic. Although cooperative perception aims to resolve practical problems, most of the current research work is designed based on the default assumption that the communication capacities of collaborated perception entities are identical. In this work, we introduce a fundamental approach - Dynamic Feature Sharing (DFS) - for cooperative perception from a more pragmatic context. Specifically, a DFS-based cooperative perception framework is designed to dynamically reduce the feature data required for sharing among the cooperating entities. Convolution-based Priority Filtering (CPF) is proposed to enable DFS under different communication constraints (e.g., bandwidth) by filtering the feature data according to the designed priority values. Zero-shot experiments demonstrate that the proposed CPF method can improve cooperative perception performance by approximately +22% under a dynamic communication-capacity condition and up to +130% when the communication bandwidth is reduced by 90 %. 

   more » « less
3. Bayesian real-time classification of multi-messenger electromagnetic and gravitational-wave observations

   https://doi.org/10.1088/1361-6382/ad3279  

   Berbel, Marina; Miravet-Tenés, Miquel; Sharma_Chaudhary, Sushant; Albanesi, Simone; Cavaglià, Marco; Magaña_Zertuche, Lorena; Tseneklidou, Dimitra; Zheng, Yanyan; Coughlin, Michael W; Toivonen, Andrew (April 2024, Classical and Quantum Gravity)

   Abstract Because of the electromagnetic (EM) radiation produced during the merger, compact binary coalescences with neutron stars may result in multi-messenger observations. In order to follow up on the gravitational-wave (GW) signal with EM telescopes, it is critical to promptly identify the properties of these sources. This identification must rely on the properties of the progenitor source, such as the component masses and spins, as determined by low-latency detection pipelines in real time. The output of these pipelines, however, might be biased, which could decrease the accuracy of parameter recovery. Machine learning algorithms are used to correct this bias. In this work, we revisit this problem and discuss two new implementations of supervised machine learning algorithms,K-nearest neighbors and random forest, which are able to predict the presence of a neutron star and post-merger matter remnant in low-latency compact binary coalescence searches across different search pipelines and data sets. Additionally, we present a novel approach for calculating the Bayesian probabilities for these two metrics. Instead of metric scores derived from binary machine learning classifiers, our scheme is designed to provide the astronomy community well-defined probabilities. This would deliver a more direct and easily interpretable product to assist EM telescopes in deciding whether to follow up on GW events in real time. 

   more » « less
4. Optimal Synthesis of Robust IDK Classifier Cascades

   https://doi.org/10.1145/3609129  

   Baruah, Sanjoy; Burns, Alan; Davis, Robert Ian (September 2023, ACM Transactions on Embedded Computing Systems)

   AnIDK classifieris a computing component that categorizes inputs into one of a number of classes, if it is able to do so with the required level of confidence, otherwise it returns “I Don’t Know” (IDK).IDK classifier cascadeshave been proposed as a way of balancing the needs for fast response and high accuracy in classification-based machine perception. Efficient algorithms for the synthesis of IDK classifier cascades have been derived; however, the responsiveness of these cascades is highly dependent on the accuracy of predictions regarding the run-time behavior of the classifiers from which they are built. Accurate predictions of such run-time behavior is difficult to obtain for many of the classifiers used for perception. By applying thealgorithms using predictionsframework, we propose efficient algorithms for the synthesis of IDK classifier cascades that arerobustto inaccurate predictions in the following sense: the IDK classifier cascades synthesized by our algorithms have short expected execution durations when the predictions are accurate, and these expected durations increase only within specified bounds when the predictions are inaccurate. 

   more » « less
5. Optimizing Confidential Deep Learning for Real-Time Systems

   https://doi.org/10.1145/3724126  

   Babar, Mohammad Fakhruddin; Hasan, Monowar (March 2025, ACM Transactions on Cyber-Physical Systems)

   Deep neural networks (DNNs) are increasingly used in time-critical, learning-enabled cyber-physical applications such as autonomous driving and robotics. Despite the growing use of various deep learning models, protecting DNN inference from adversarial threats while preserving model privacy and confidentiality remains a key concern for resource and timing-constrained autonomous cyber-physical systems. One potential solution, primarily used in general-purpose systems, is the execution of the DNN workloads withintrusted enclavesavailable on current off-the-shelf processors. However, ensuring temporal guarantees when running DNN inference within these enclaves poses significant challenges in real-time applications due to(a)the large computational and memory demands of DNN models and(b)the overhead introduced by frequent context switches between “normal” and “trusted” execution modes. This paper introduces new time-aware schemes for dynamic (EDF) and fixed-priority (RM) schedulers to preserve the confidentiality of DNN tasks by running them inside trusted enclaves. We first propose a technique thatsliceseach DNN layer and runs them sequentially in the enclave. However, due to the extra context switch overheads of individual layer slices, we further introduce a novellayer fusiontechnique. Layer fusion improves real-time guarantees by grouping multiple layers of DNN workload from multiple tasks, thus allowing them to fit and run concurrently within the enclaves while maintaining timing constraints. We implemented and tested our ideas on the Raspberry Pi platform running a DNN-enabled trusted operating system (OP-TEE with DarkNet-TZ) and three DNN architectures (AlexNet-squeezed, Tiny Darknet, YOLOv3-tiny). Compared to the layer-wise partitioning approach, layer fusion can(a)schedule up to 3x more tasksets for EDF and 5x for RM and(b)reduce context switches by up to 11.12x for EDF and by up to 11.06x for RM. 

   more » « less