Real-world Devices Research Articles

Emergent updates and future directions on white light emitting diodes based on luminous lead-free halide perovskites

Lead halide perovskites possess exceptional optical and electrical properties, yet their inherent toxicity and instability present significant challenges for practical application in white light-emitting diodes (WLEDs) for displays and lighting. Alhough, techniques such as reduced dimensionality, transition metal doping, and protective encapsulation have shown some success in enhancing stability. These advancements remain limited and do not fully meet the requirements for real-world WLED devices. This featured letter delves into the emerging advancements in lead-free halide perovskite WLEDs includes color-conversion WLEDs, and single-emissive-layer WLEDs. Despite significant advancements, issues with stability, efficiency, and scalable fabrication still exist. Future approaches centre on methods to enhance these features, opening the door for lead-free perovskite WLEDs to become commercially viable and bringing in a new age of environmentally friendly lighting options. Additionally, this featured letter provides a comprehensive review of the current advancements and future possibilities in WLEDs based on lead-free halide perovskites materials. This letter serves as an invaluable resource for researchers, scientists, and engineers dedicated to advancing this field.

Few-shot Algorithms for Consistent Neural Decoding (FALCON) Benchmark.

Intracortical brain-computer interfaces (iBCIs) can restore movement and communication abilities to individuals with paralysis by decoding their intended behavior from neural activity recorded with an implanted device. While this activity yields high-performance decoding over short timescales, neural data are often nonstationary, which can lead to decoder failure if not accounted for. To maintain performance, users must frequently recalibrate decoders, which requires the arduous collection of new neural and behavioral data. Aiming to reduce this burden, several approaches have been developed that either limit recalibration data requirements (few-shot approaches) or eliminate explicit recalibration entirely (zero-shot approaches). However, progress is limited by a lack of standardized datasets and comparison metrics, causing methods to be compared in an ad hoc manner. Here we introduce the FALCON benchmark suite (Few-shot Algorithms for COnsistent Neural decoding) to standardize evaluation of iBCI robustness. FALCON curates five datasets of neural and behavioral data that span movement and communication tasks to focus on behaviors of interest to modern-day iBCIs. Each dataset includes calibration data, optional few-shot recalibration data, and private evaluation data. We implement a flexible evaluation platform which only requires user-submitted code to return behavioral predictions on unseen data. We also seed the benchmark by applying baseline methods spanning several classes of possible approaches. FALCON aims to provide rigorous selection criteria for robust iBCI decoders, easing their translation to real-world devices.

Logic Diagnosis Based on Logic Built-In Self-Test Signatures Collected In-Field from Failing System-on-Chips

Modern embedded nanoelectronic devices, particularly in safety-critical sectors, require high dependability throughout their lifecycle. To address this, designers have started integrating extra circuitry for on-device self-testing, such as the Logic Built-In Self-Test (LBIST). However, while automatic test equipment (ATE) ensures exhaustive testing during manufacturing, in-field testing capabilities are limited. This study introduces a novel methodology for in-field data collection of failure information from LBIST engines and a subsequent logic diagnosis strategy to facilitate failure analysis of field returns. The information is collected from key-on and key-off self-tests, executed by central processing units (CPUs) with a fixed seed and different frequency configurations, primarily addressing transition delay (TRN) faults. The proposed approach capitalizes on the constrained in-field configurability of LBIST and does not require a custom architecture, making it highly practical and readily applicable to real-world devices. The logic diagnosis strategy significantly reduces the number of candidate faults by exploiting the first failing pattern index found during the in-field testing and data collection. Reducing fault candidates could enhance accuracy during failure analysis, especially when field return devices exhibit a “No Trouble Found” (NTF) behavior. The experimental results are reported for ITC’99 benchmarks and an industrial automotive system-on-chip (SoC) produced by STMicroelectronics, with about 20 million gates.

Enhancing Cybersecurity: Ransomware Detection - A Proof of Concept Study

Presently, our digital landscape faces a pervasive onslaught of diverse cyber threats, encompassing distributed denial of service (DDoS), phishing, ransomware, and smishing, all orchestrated with malicious intent. Counteracting these malicious incursions poses a formidable challenge, particularly in devising efficacious detection solutions. Ransomware incidents are on the ascent, particularly within critical sectors such as healthcare, finance, and telecommunications. Consequently, this paper introduces a proof of concept (POC) aimed at detecting ransomware activities targeting Internet of Medical Things (IoMT) devices. The primary objective of this paper revolves around the identification and evaluation of factors correlating with ransomware attacks on IoMT and then developing a ransomware detector. The experimental framework involves the utilization of a simulated environment mirroring real-world IoMT devices and networks. The methodology integrates diverse approaches, encompassing data collection from IoMT devices, analysis of ransomware behaviour through the study of encryption patterns, and anomaly detection. The POC assesses the efficacy of these methodologies in detecting and responding to ransomware threats, with the experimentation conducted through hybrid analysis within a controlled laboratory setting. The dataset consists of 13 families with a total malware of 9251 taken from GitHub. For POC, a total of 3459 ransomware dataset samples have been selected. As a result of the experiment, thirteen (13) distinct features have been identified as trigger factors for ransomware attacks. These features are obtained by capturing of the processes initiated by the ransomware samples using a process monitor (procmon). A temporal pattern of the processes which include the file system, API, and access based on their frequency of occurrence were used to develop a ransomware detection model. The proposed approach achieved an accuracy of 99.2% and an error rate of 0.8 %, when evaluated on temporal pattern dataset and by using an enhanced artificial neural network (EANN).

Fenestrated Endovascular Aortic Repair of Complex Aneurysm Using the Anaconda Fenestrated Device: 1-Year Results From the French Multicenter Registry.

This study aims to assess the safety and efficacy of the fenestrated Anaconda device for the treatment of complex aortic aneurysms over 1 year in daily clinical practice. All patients who received the graft between October 2019 and October 2020 were prospectively enrolled in an observational, multicenter national registry. The primary endpoint was the aneurysm-related 1-year mortality rate. Secondary endpoints included all-cause mortality, technical success, device-related adverse events (AEs) and major complications, secondary reinterventions, endoleaks, target vessel (TV) outcomes, and a center-volume analysis comparing mortality and secondary reintervention rate between expert centers (>15 F/BEVAR per year) and nonexpert centers (≤15 F/BEVAR per year). Ninety-seven patients from 28 centers were treated. Aneurysms types included juxta-renal and short neck (84.6%), pararenal (5.2%), paravisceral (3.1%), and type IV thoracoabdominal (2.5%). Configurations with 1, 2, 3, and 4 fenestrations were used in 2, 12, 18, and 65 cases, respectively, totaling 350 TVs. Technical success was 98.0%, with 1 type Ic and 1 type IIIc endoleak. Mean follow-up was 468 days. Ten patients died, with 8 deaths (8.2%) due to non-aortic causes and 2 deaths (2.1%) from unknown causes. The estimated 1-year survival rate was 92.7% (95% CI: 87.8-98.1%). The reintervention rate at 30 days was 11.3% and the estimated reintervention rate at 1 year was 21.6% (95% CI: 12.7-29.6%). No type Ia endoleak occurred during follow-up. Three type III endoleaks occurred and required reintervention. Three TV occlusion occurred, yielding a 1-year TVI-free rate of 92.4% (95% CI: 89.4-99.3). No device-related major complication was recorded, and 3 device-related AEs occurred: 1 limb migration, 1 limb kinking and 1 limb thrombosis. There was no statistically significant difference in mortality and survival without reintervention between expert and nonexpert centers (Log rank p=0.52 and p=0.16, respectively). The fenestrated Anaconda device demonstrated acceptable early and at 1-year safety and efficacy for treating primarily juxta-renal and short-neck aortic aneurysms, in centers with varying levels of experience. The Anaconda Fenestrated, as the second commercially available custom-made device for complex endovascular aortic surgery, might adress some anatomical challenges. Following market authorization, this registry presents the 1-year outcomes of real-world device usage in a large number of centers, including small clinics. The results indicate that the device provided adequate efficacy and safety for treating mostly juxta-renal and short neck aneurysms, with notable improvement in technical success rate compared to older reports.

Real-world data to support post-market safety and performance of embolization coils: evidence generation from a medical device manufacturer and data institute partnership

BackgroundRecognizing the limitations of pre-market clinical data, regulatory authorities have embraced total product lifecycle management with post-market surveillance (PMS) data to assess medical device safety and performance. One method of proactive PMS involves the analysis of real-world data (RWD) through retrospective review of electronic health records (EHR). Because EHRs are patient-centered and focused on providing tools that clinicians use to determine care rather than collecting information on individual medical products, the process of transforming RWD into real-world evidence (RWE) can be laborious, particularly for medical devices with broad clinical use and extended clinical follow-up. This study describes a method to extract RWD from EHR to generate RWE on the safety and performance of embolization coils.MethodsThrough a partnership between a non-profit data institute and a medical device manufacturer, information on implantable embolization coils’ use was extracted, linked, and analyzed from clinical data housed in an electronic data warehouse from the state of Indiana’s largest health system. To evaluate the performance and safety of the embolization coils, technical success and safety were defined as per the Society of Interventional Radiology guidelines. A multi-prong strategy including electronic and manual review of unstructured (clinical chart notes) and structured data (International Classification of Disease codes), was developed to identify patients with relevant devices and extract data related to the endpoints.ResultsA total of 323 patients were identified as treated using Cook Medical Tornado, Nester, or MReye embolization coils between 1 January 2014 and 31 December 2018. Available clinical follow-up for these patients was 1127 ± 719 days. Indications for use, adverse events, and procedural success rates were identified via automated extraction of structured data along with review of available unstructured data. The overall technical success rate was 96.7%, and the safety events rate was 5.3% with 18 major adverse events in 17 patients. The calculated technical success and safety rates met pre-established performance goals (≥ 85% for technical success and ≤ 12% for safety), highlighting the relevance of this surveillance method.ConclusionsGenerating RWE from RWD requires careful planning and execution. The process described herein provided valuable longitudinal data for PMS of real-world device safety and performance. This cost-effective approach can be translated to other medical devices and similar RWD database systems.

Performance evaluation and analysis of public-key schemes in embedded IoT devices

The Internet of Things (IoT) is considered one of the emerging technologies that have attracted widespread attention from industry and academia as a result of their ability to use them in many applications, including military, healthcare, and industrial control. The inherent vulnerabilities in communications of IoT networks, which consist of resource-constrained devices, pose significant security challenges. To protect sensitive data exchanged within these networks, it is necessary to design and implement lightweight and efficient cryptographic schemes that consider resource-constrained IoT devices. This paper evaluates the performance of five well-known public-key schemes on three real-world IoT devices including IOTLAB-M3, Arduino-Zero, and Decawave DWM1001. The examined public-key schemes are categorized into two classes: integer factorization-based (RSA and Rabin schemes) and elliptic curve-based (ECIES, ECDH, and ECDSA schemes). By conducting comprehensive experiments, we compare these schemes in terms of computational time and energy consumption, providing valuable insights for selecting optimal public-key schemes in resource-constrained IoT environments.

Analysis of electronic properties and sensing applications in Graphene/BC3 heterostructures

Many research studies have been conducted into the applications of the heterostructures of graphene (Gr) and boron carbide (BC3) in real-world devices after they were synthesized successfully by researchers. Thanks to their unique electronic attributes and high surface-to-volume ratio, the above-mentioned two-dimensional nanosheets (NSs) such as have enjoyed the interest of many research groups and scientists. Within this piece of research, the electronic and structural attributes of pure Gr (PGr), pure BC3 (PBC3) and their in-plane heterostructures were investigated by employing the DFT along with the density functionals B3LYP and WB97XD. The results demonstrated that by increasing the concentration of the B-C, there was a gradual increase in the bandgap. Also, the structural stability of the NSs was good and the cohesive energy was favourable. In addition, the adhesion attributes of these NSs were investigated towards two toxic gasses, namely CO and SO2. Amongst the heterostructures, after exposure to CO and SO2, the adhesion energy of GBC3I was greater, which was approximately −0.487 eV and −0.229 eV, respectively. Following the adhesion of SO2 onto the surface of the NSs, apart from the PGr, there was a significant change in their electronic attributes such as conductance, workfunction, Fermi level, the LUMO, the HOMO and the bandgap. However, following the adhesion of CO, the above-mentioned attributed remained almost the same. Based on the NBO and Mulliken charge analysis, there was a charge transport from the gasses to the NSs. Despite the fact that the adhesion energies for CO were relative higher than the adhesion energies for SO2for the NSs, the analysis of the MEP maps, the charge transport and the electronic attributes demonstrated that the NSs can be used as promising nanosensors for the detection of SO2 rather than CO.

A robust hybridizable discontinuous Galerkin scheme with harmonic averaging technique for steady state of real-world semiconductor devices

Solving real-world nonlinear semiconductor device problems modeled by the drift-diffusion equations coupled with the Poisson equation (also known as the Poisson-Nernst-Planck equations) necessitates an accurate and efficient numerical scheme which can avoid non-physical oscillations even for problems with extremely sharp doping profiles. In this paper, we propose a flexible and high-order hybridizable discontinuous Galerkin (HDG) scheme with harmonic averaging (HA) technique to tackle these challenges. The proposed HDG-HA scheme combines the robustness of finite volume Scharfetter-Gummel (FVSG) method with the high-order accuracy and hp-flexibility offered by the locally conservative HDG scheme. The coupled Poisson equation and two drift-diffusion equations are simultaneously solved by the Newton method. Indicators based on the gradient of net doping and solution variables are proposed to switch between cells with HA technique and high-order conventional HDG cells, utilizing the flexibility of HDG scheme. Numerical results suggest that the proposed scheme does not exhibit oscillations or convergence issues, even when applied to heavily doped and sharp PN-junctions. Devices with circular junctions and realistic doping profiles are simulated in two dimensions, qualifying this scheme for practical simulation of real-world semiconductor devices.

ECG classification via integration of adaptive beat segmentation and relative heart rate with deep learning networks

We propose a state-of-the-art deep learning approach for accurate electrocardiogram (ECG) signal analysis, addressing both waveform delineation and beat type classification tasks. For beat type classification, we integrated two novel schemes into the deep learning model, significantly enhancing its performance. The first scheme is an adaptive beat segmentation method that determines the optimal duration for each heartbeat based on RR-intervals, mitigating segmenting errors from conventional fixed-period segmentation. The second scheme incorporates relative heart rate information of the target beat compared to neighboring beats, improving the model’s ability to accurately detect premature atrial contractions (PACs) that are easily confused with normal beats due to similar morphology. Extensive evaluations on the PhysioNet QT Database, MIT-BIH Arrhythmia Database, and real-world wearable device data demonstrated the proposed approach’s superior capabilities over existing methods in both tasks. The proposed approach achieved sensitivities of 99.81% for normal beats, 99.08% for premature ventricular contractions, and 97.83% for PACs in beat type classification. For waveform delineation, we achieved F1-scores of 0.9842 for non-waveform, 0.9798 for P-waves, 0.9749 for QRS complexes, and 0.9848 for T-waves. It significantly outperforms existing methods in PAC detection while maintaining high performance across both tasks. The integration of aforementioned two schemes into the deep learning model improved the accuracy of normal sinus rhythms and arrhythmia detection.

From Design to Device: Challenges and Opportunities in Computational Discovery of p -Type Transparent Conductors

A high-performance p-type transparent conductor (TC) does not yet exist but could lead to advances in a wide range of optoelectronic applications and enable new architectures for, e.g., next-generation photovoltaic (PV) devices. High-throughput computational material screenings have been a promising approach to filter databases and identify new p-type TC candidates and some of these predictions have been experimentally validated. However, most of these predicted candidates do not have experimentally achieved properties on par with n-type TCs used in solar cells and therefore have not yet been used in commercial devices. Thus, there is still a significant divide between transforming predictions into results that are actually achievable in the laboratory and an even greater lag in scaling predicted materials into functional devices. In this perspective, we outline some of the major disconnects in this materials discovery process—from scaling computational predictions into synthesizable crystals and thin films in the laboratory to scaling laboratory-grown films into real-world solar devices—and share insights to inform future strategies for TC discovery and design. Published by the American Physical Society 2024

Conditions Leading to Ketene Formation in Vaping Devices and Implications for Public Health.

The outbreak of e-cigarette or vaping use-associated lung injury (EVALI) in the United States in 2019 led to a total of 2807 hospitalizations with 68 deaths. While the exact causes of this vaping-related lung illness are still being debated, laboratory analyses of products from victims of EVALI have shown that vitamin E acetate (VEA), an additive in some tetrahydrocannabinol (THC)-containing products, is strongly linked to the EVALI outbreak. Because of its similar appearance and viscosity to pure THC oil, VEA was used as a diluent agent in cannabis oils in illicit markets. A potential mechanism for EVALI may involve VEA's thermal decomposition product, ketene, a highly poisonous gas, being generated under vaping conditions. In this study, a novel approach was developed to evaluate ketene production from VEA vaping under measurable temperature conditions in real-world devices. Ketene in generated aerosols was captured by two different chemical agents and analyzed by gas chromatography-mass spectrometry (GC-MS) and liquid chromatography with tandem mass spectrometry (LC-MS/MS). The LC-MS/MS method takes advantage of the high sensitivity and specificity of tandem mass spectrometry and appears to be more suitable than GC-MS for the analysis of large batches of samples. Our results confirmed the formation of ketene when VEA was vaped. The production of ketene increased with repeat puffs and showed a correlation to temperatures (200 to 500 °C) measured within vaping devices. Device battery power strength, which affects the heating temperature, plays an important role in ketene formation. In addition to ketene, the organic oxidant duroquinone was also obtained as another thermal degradation product of VEA. Ketene was not detected when vitamin E was vaped under the same conditions, confirming the importance of the acetate group for its generation.

Using ranging for collision-immune IEEE 802.11 rate selection with statistical learning

Appropriate data rate selection at the physical layer is crucial for Wi-Fi network performance: too high rates lead to loss of data frames, while too low rates cause increased latency and inefficient channel use. Most existing methods adopt a probing approach and empirically assess the transmission success probability for each available rate. However, a transmission failure can also be caused by frame collisions. Thus, each collision leads to an unnecessary decrease in the data rate. We avoid this issue by resorting to the fine timing measurement (FTM) procedure, part of IEEE 802.11, which allows stations to perform ranging, i.e., measure their spatial distance to the AP. Since distance is not affected by sporadic distortions such as internal and external channel interference, we use this knowledge for data rate selection. Specifically, we propose FTMRate, which applies statistical learning (a form of machine learning) to estimate the distance based on measurements, predicts channel quality from the distance, and selects data rates based on channel quality. We define three distinct estimation approaches: exponential smoothing, Kalman filter, and particle filter. Then, with a thorough performance evaluation using simulations and an experimental validation with real-world devices, we show that our approach has several positive features: it is resilient to collisions, provides near-instantaneous convergence, is compatible with commercial-off-the-shelf devices, and supports pedestrian mobility. Thanks to these features, FTMRate outperforms existing solutions in a variety of line-of-sight scenarios, providing close to optimal results. Additionally, we introduce Hybrid FTMRate, which can intelligently fall back to a probing-based approach to cover non-line-of-sight cases. Finally, we discuss the applicability of the method and its usefulness in various scenarios.

Fabrication of Ni-Polyaniline/Graphene Oxide Composite Electrode with High Capacitance and Water Splitting Activity.

The Ni-PANI@GO composite electrode was fabricated via cost effective electrodeposition technique. According to the XRD, FTIR, Raman, SEM, and XPS analyses revealed that the nickel doped PANI@GO composite has been fabricated on the surface of the nickel foam. Addition of nickel significantly enhanced interaction between graphene with PANI leading to higher degree of polyaniline doping though imine groups. Electrochemical investigation revelated the significant performance of the Ni-PANI@GO composite electrode, boosting an impressive capacitance of 4480 F/g at 40 A/g, surpassing previous Ni-foam-based binder-free electrodes. Notably, Ni-PANI@GO electrode displayed excellent catalytic activity in both oxygen evolution reaction (OER) and hydrogen evolution reaction (HER), generating a considerable volume of the gas bubbles at relatively modest overpotentials of 279 mV and 244 mV respectively. This event allows for the achievement of 20 mA cm-2 current density. Furthermore, in the laboratory-scale water electrolyzer, a low cell voltage of 1.72 V was achieved, facilitating a water-splitting current density of 20 mA cm-2. This study underscores the premising potential for the real-world device's application of the versatile Ni-PANI@GO composite electrode.

Dependency-aware action planning for smart home.

How can a smart home system control a connected device to be in a desired state? Recent developments in the Internet of Things (IoT) technology enable people to control various devices with the smart home system rather than physical contact. Furthermore, smart home systems cooperate with voice assistants such as Bixby or Alexa allowing users to control their devices through voice. In this process, a user's query clarifies the target state of the device rather than the actions to perform. Thus, the smart home system needs to plan a sequence of actions to fulfill the user's needs. However, it is challenging to perform action planning because it needs to handle a large-scale state transition graph of a real-world device, and the complex dependence relationships between capabilities. In this work, we propose SmartAid (Smart Home Action Planning in awareness of Dependency), an action planning method for smart home systems. To represent the state transition graph, SmartAid learns models that represent the prerequisite conditions and operations of actions. Then, SmartAid generates an action plan considering the dependencies between capabilities and actions. Extensive experiments demonstrate that SmartAid successfully represents a real-world device based on a state transition log and generates an accurate action sequence for a given query.

Stateful black-box fuzzing for encryption protocols and its application in IPsec

Owing to the rapid development of information security technology, the security analysis of encryption protocols has received widespread attention. In this paper, we propose a stateful black-box encryption protocol fuzzing method to analyze the security of real-world black-box encryption protocol programs and devices. This method does not rely on the source code but uses captured packets as input, performs state selection based on a Monte Carlo tree search algorithm, and processes the encryption/decryption conversion of mutant test cases based on the intermediate mapper. It sends test cases and collects responses in interactive communication with the tested program and dynamically optimizes the corpus based on the collected state information. Based on this method, we develop SBEPFuzz and primarily analyze IPsec. We evaluate SBEPFuzz on six widely used IPsec implementations. The experimental results show that SBEPFuzz achieves higher code coverage, and can discover more protocol state sequences and vulnerabilities. Furthermore, we discover four anomalies, including malformed packets triggering service crashes and abnormal interactions leading to plaintext ID leakage, which also reflect the differences in details among different implementations.

IoT-SecNet: A Trust-based Framework for Enhancing Security in IoT Communications and Network Layer Protocols

The Internet of Things (IoT) integrates cloud computing, big data, and data mining to become a vital 21st-century technology. IoT uses hardware, software, and communication protocols to connect and share data amongst devices. However, IoT’s increasing usage raises security concerns, especially due to its ubiquitous communication infrastructure. IoT ecosystems depend on communication protocols to connect hardware and application software and enable real-world device deployment. Due to their openness, these protocols are vulnerable to DoS, DDoS, and buffer overflow attacks despite their importance. Thus, a complete framework to protect these protocols at the communication and network layers is needed. This study proposes an IoT-SecNet (Secure Network), a Trust-based Framework to improve security across IoT communication protocols. This framework verifies sensor data and sends request packets for processing, ensuring security and effective traffic management across IoT layers. Through experimental assessments, we show a 90% accuracy rate in identifying assaults and compare it to existing security solutions. Our architecture protects against threats by maintaining high communication layer protocol throughput and low delay.

LLMEffiChecker:Understanding and Testing Efficiency Degradation of Large Language Models

Large Language Models (LLMs) have received much recent attention due to their human-level accuracy. While existing works mostly focus on either improving accuracy or testing accuracy robustness, the computation efficiency of LLMs, which is of paramount importance due to often vast generation demands and real-time requirements, has surprisingly received little attention. In this paper, we make the first attempt to understand and test potential computation efficiency robustness in state-of-the-art LLMs. By analyzing the working mechanism and implementation of 20,543 public-accessible LLMs, we observe a fundamental property in LLMs that could be manipulated in an adversarial manner to reduce computation efficiency significantly. Our interesting observation is that the output length determines the computation efficiency of LLMs instead of the input, where the output length depends on two factors: an often sufficiently large yet pessimistic pre-configured threshold controlling the max number of iterations and a runtime generated end of sentence (EOS) token. Our key motivation is to generate test inputs that could sufficiently delay the generation of EOS such that LLMs would have to go through enough iterations to satisfy the pre-configured threshold. We present LLMEffiChecker , which can work under both white-box setting and black-box setting. In the white-box scenario, LLMEffiChecker develops a gradient-guided technique that searches for a minimal and unnoticeable perturbation at character-level, token-level, and structure-level. In the black-box scenario, LLMEffiChecker employs a causal inference-based approach to find critical tokens and similarly applies three levels of imperceptible perturbation to them. Both the white-box and black-box settings effectively delay the appearance of EOS, compelling these inputs to reach the naturally-unreachable threshold. To demonstrate the effectiveness of LLMEffiChecker , we conduct a systematic evaluation on nine public-available LLMs: Google T5, AllenAI WMT14, Helsinki-NLP translator, Facebook FairSeq, UNICAMP-DL translator, MarianMT, Google FLAN-T5, MBZUAI LaMini-GPT and Salesforce CodeGen. Experimental results show that LLMEffiChecker can increase on average LLMs’ response latency and energy consumption by 325% to 3244% and 344% to 3616%, respectively, by perturbing just one character or token in the input sentence. Our case study shows that inputs generated by LLMEffiChecker significantly affect the battery power in real-world mobile devices ( i.e. , drain more than 30 times battery power than normal inputs).

Decentralised Global Service Discovery for the Internet of Things.

The Internet of Things (IoT) consists of millions of devices deployed over hundreds of thousands of different networks, providing an ever-expanding resource to improve our understanding of and interactions with the physical world. Global service discovery is key to realizing the opportunities of the IoT, spanning disparate networks and technologies to enable the sharing, discovery, and utilisation of services and data outside of the context in which they are deployed. In this paper, we present Decentralised Service Registries (DSRs), a novel trustworthy decentralised approach to global IoT service discovery and interaction, building on DSF-IoT to allow users to simply create and share public and private service registries, to register and query for relevant services, and to access both current and historical data published by the services they discover. In DSR, services are registered and discovered using signed objects that are cryptographically associated with the registry service, linked into a signature chain, and stored and queried for using a novel verifiable DHT overlay. In contrast to existing centralised and decentralised approaches, DSRs decouple registries from supporting infrastructure, provide privacy and multi-tenancy, and support the verification of registry entries and history, service information, and published data to mitigate risks of service impersonation or the alteration of data. This decentralised approach is demonstrated through the creation and use of a DSR to register and search for real-world IoT devices and their data as well as qualified using a scalable cluster-based testbench for the high-fidelity emulation of peer-to-peer applications. DSRs are evaluated against existing approaches, demonstrating the novelty and utility of DSR to address key IoT challenges and enable the sharing, discovery, and use of IoT services.

Layer Compression of Deep Networks with Straight Flows

Very deep neural networks lead to significantly better performance on various real tasks. However, it usually causes slow inference and is hard to be deployed on real-world devices. How to reduce the number of layers to save memory and to accelerate the inference is an eye-catching topic. In this work, we introduce an intermediate objective, a continuous-time network, before distilling deep networks into shallow networks. First, we distill a given deep network into a continuous-time neural flow model, which can be discretized with an ODE solver and the inference requires passing through the network multiple times. By forcing the flow transport trajectory to be straight lines, we find that it is easier to compress the infinite step model into a one-step neural flow model, which only requires passing through the flow model once. Secondly, we refine the one-step flow model together with the final head layer with knowledge distillation and finally, we can replace the given deep network with this one-step flow network. Empirically, we demonstrate that our method outperforms direct distillation and other baselines on different model architectures (e.g. ResNet, ViT) on image classification and semantic segmentation tasks. We also manifest that our distilled model naturally serves as an early-exit dynamic inference model.