Deep learning models deployed on edge devices frequently encounter resource variability, which arises from fluctuating energy levels, timing constraints, or prioritization of other critical tasks within the system. State-of-the-art machine learning pipelines generate resource-agnostic models that are not capable to adapt at runtime. In this work, we introduce Resource-Efficient Deep Subnetworks (REDS) to tackle model adaptation to variable resources. In contrast to the state-of-the-art, REDS leverages structured sparsity constructively by exploiting permutation invariance of neurons, which allows for hardware-specific optimizations. Specifically, REDS achieves computational efficiency by (1) skipping sequential computational blocks identified by a novel iterative knapsack optimizer, and (2) taking advantage of data cache by re-arranging the order of operations in REDS computational graph. REDS supports conventional deep networks frequently deployed on the edge and provides computational benefits even for small and simple networks. We evaluate REDS on eight benchmark architectures trained on the Visual Wake Words, Google Speech Commands, Fashion-MNIST, CIFAR-10 and ImageNet-1K datasets, and test on four off-the-shelf mobile and embedded hardware platforms. We provide a theoretical result and empirical evidence demonstrating REDS' outstanding performance in terms of submodels' test set accuracy, and demonstrate an adaptation time in response to dynamic resource constraints of under 40$\mu$s, utilizing a fully-connected network on Arduino Nano 33 BLE.

Threat models contribute to strengthen the security of an enterprise system by listing the cybersecurity threats that might affect it as well as possible mitigations to such threats. Given the importance of these models, different tools have been devised to support creating or updating threat models. However, these tools focus mainly on the needs of cybersecurity experts, often overlooking non-expert users.To address this gap, in the first part of this paper we present a chat-based LLM assistant to answer both expert and non-expert queries on threat models. The assistant is based on a microservice architecture and uses Retrieval-Augmented Generation (RAG) to extract relevant information from an industrial threat model.Guaranteeing the reliability of an LLM’s answers in sensitive fields such as cybersecurity is paramount. For this reason, in the second part of this paper, we evaluate the LLM assistant using (i) Bleu and Rouge score metrics, (ii) human evaluation, and (iii) an automatic LLM-as-a-judge approach.The evaluation, conducted on 275 test cases, confirms the accuracy of the answers provided by our LLM-based threat model assistant. Additionally, the results of the human and LLM-as-a-judge evaluations are consistent. This corroborates the effectiveness of automatic LLM-as-a-judge methods in assessing the performance of LLM-based industrial solutions.

Our increasingly digital and connected world has led to the generation of unprecedented amounts of data. This data must be efficiently managed, transmitted, and stored to preserve resources and allow scalability. Data compression has therein been a key technology for a long time, resulting in a vast landscape of available techniques. This largest-to-date study analyzes and compares various lossless data compression methods for time series data. We present a unified framework encompassing two stages: data transformation and entropy encoding. We evaluate compression algorithms across both synthetic and real-world datasets with varying characteristics. Through ablation studies at each compression stage, we isolate the impact of individual components on overall compression performance–revealing the strengths and weaknesses of different algorithms when facing diverse time series properties. Our study underscores the importance of well-configured and complete compression pipelines beyond individual components or algorithms; it offers a comprehensive guide for selecting and composing the most appropriate compression algorithms tailored to specific datasets.

Timing prediction for real-time systems, especially those based on multi-core processor technology, presents enor-mous challenges in the design of modern industrial control sys-tems. Classical timing analysis techniques that have been effective in the past cannot yet keep up with the increased complexity of industrial control systems. Therefore, using hardware-in-the-loop testing to understand the timing behavior of real-time tasks is a prevalent practice across many industries. However, applying this state-of-the-practice to Continuous Integration (CI) software development workflows is expensive, and frequently leads to delayed developer feedback on task timing for code commits. To address this challenge, we propose the Chronos framework, which focuses on improving development efficiency of industrial control system software. Chronos utilizes CI data from both simulation and hardware-in-the-loop testing to build machine learning based, cross-platform timing prediction models, which correlate the simulated performance of the tasks with their actual timing observed on the target embedded hardware. For any new code that is committed, the timing prediction is triggered by the CI server with the trained machine learning models, enabling fast feedback on timing behavior of the committed code. We demonstrate the effectiveness of Chronos with preliminary results on real industrial control system setups.

Great public speakers are made, not born. Practicing a presentation in front of colleagues is common practice and results in a set of subjective judgements what could be improved. In this paper we describe the design and implementation of a mobile app which estimates the quality of speaker's delivery in real time in a fair, repeatable and privacy-preserving way. Quantle estimates the speaker's pace in terms of the number of syllables, words and clauses, computes pitch and duration of pauses. The basic parameters are then used to estimate the talk complexity based on readability scores from the literature to help the speaker adjust his delivery to the target audience. In contrast to speech-to-text-based methods used to implement a digital presentation coach, Quantle does processing locally in real time and works in the flight mode. This design has three implications: (1) Quantle does not interfere with the surrounding hardware, (2) it is power-aware, since 95.2% of the energy used by the app on iPhone 6 is spent to operate the built-in microphone and the screen, and (3) audio data and processing results are not shared with a third party therewith preserving speaker's privacy.

Air pollution is a major concern for public health and urban environments. Conventional air pollution monitoring systems install a few highly accurate, expensive stations at representative locations. Their sparse coverage and low spatial resolution are insufficient to quantify urban air pollution and its impacts on human health and environment. Advances in low-cost portable air pollution sensors have enabled air pollution monitoring deployments at scale to measure air pollution at high spatiotemporal resolution. However, it is challenging to ensure the accuracy of these low-cost sensor deployments because the sensors are more error-prone than high-end sensing infrastructures and they are often deployed in harsh environments. Sensor calibration has proven to be effective to improve the data quality of low-cost sensors and maintain the reliability of long-term, distributed sensor deployments. In this paper, we review the state-of-the-art low-cost air pollution sensors, identify their major error sources, and comprehensively survey calibration models as well as network recalibration strategies suited for different sensor deployments. We also discuss limitations of exiting methods and conclude with open issues for future sensor calibration research.

Sensing tongue movements enables various applications in hands-free interaction and alternative communication. We propose BARTON, a BARometer based low-power and robust TONgue movement sensing system. Using a low sampling rate of below 50 Hz, and only extracting simple temporal features from in-ear pressure signals, we demonstrate that it is plausible to distinguish important tongue gestures (left, right, forward) at low power consumption. We prototype BARTON with commodity earpieces integrated with COTS barometers for in-ear pressure sensing and an ARM micro-controller for signal processing. Evaluations show that BARTON yields 94% classification accuracy and 8.4 mW power consumption, which achieves comparable accuracy, but consumes 44 times lower energy than the state-of-the-art microphone-based solutions. BARTON is also robust to head movements and operates with music played directly from earphones.