Facial identity is subject to two primary natural variations: time-dependent (TD) factors such as age, and time-independent (TID) factors including sex and race. This study aims to address a broader problem known as v...
详细信息
The advent of ubiquitous computing devices in the Internet of Things (IoT) has resulted in an explosion of data. Traditional centralized machine learning models face challenges including limited bandwidth in wireless ...
详细信息
The aim of this paper is to analyze the implementation of intelligent lighting within the concept of smart energy based on the possibility of saving and efficient use of energy, which is largely based on non-renewable...
详细信息
By the emergence of the fourth industrial revolution,interconnected devices and sensors generate large-scale,dynamic,and inharmonious data in Industrial Internet of Things(IIoT)*** vast heterogeneous data increase the...
详细信息
By the emergence of the fourth industrial revolution,interconnected devices and sensors generate large-scale,dynamic,and inharmonious data in Industrial Internet of Things(IIoT)*** vast heterogeneous data increase the challenges of security risks and data analysis *** IIoT grows,cyber-attacks become more diverse and complex,making existing anomaly detection models less effective to *** this paper,an ensemble deep learning model that uses the benefits of the Long Short-Term Memory(LSTM)and the AutoEncoder(AE)architecture to identify out-of-norm activities for cyber threat hunting in IIoT is *** this model,the LSTM is applied to create a model on normal time series of data(past and present data)to learn normal data patterns and the important features of data are identified by AE to reduce data *** addition,the imbalanced nature of IIoT datasets has not been considered in most of the previous literature,affecting low accuracy and *** solve this problem,the proposed model extracts new balanced data from the imbalanced datasets,and these new balanced data are fed into the deep LSTM AE anomaly detection *** this paper,the proposed model is evaluated on two real IIoT datasets-Gas Pipeline(GP)and Secure Water Treatment(SWaT)that are imbalanced and consist of long-term and short-term dependency on *** results are compared with conventional machine learning classifiers,Random Forest(RF),Multi-Layer Perceptron(MLP),Decision Tree(DT),and Super Vector Machines(SVM),in which higher performance in terms of accuracy is obtained,99.3%and 99.7%based on GP and SWaT datasets,***,the proposed ensemble model is compared with advanced related models,including Stacked Auto-Encoders(SAE),Naive Bayes(NB),Projective Adaptive Resonance Theory(PART),Convolutional Auto-Encoder(C-AE),and Package Signatures(PS)based LSTM(PS-LSTM)model.
Recent advances in neural architecture search (NAS) techniques have provided effective solutions for managing high-dimensional data. These methods algorithmically automate the design process of NAS, aiming to identify...
Phase-resolved partial discharge (PRPD) measurement has been used for decades as a method of monitoring defects in electrically insulating materials. More recently, it has seen a renewed interest in the context of fla...
详细信息
Phase-resolved partial discharge (PRPD) measurement has been used for decades as a method of monitoring defects in electrically insulating materials. More recently, it has seen a renewed interest in the context of flash sintering, a novel ceramic densification process where the sample to be densified is subjected to an electric field in addition to the usual application of heat. In the context of flash sintering, the monitoring of partial discharge (PD) activity has shown that this activity increases when approaching the onset of the thermal runaway phenomenon leading to the quick densification of the material, and is influenced by environmental factors such as relative humidity or pressure. A new microcontroller-based PRPD measurement system architecture has recently been proposed as a means to explore this PD activity in further details. While PDRD measurement is traditionally carried out by comparing the measured partial discharge pattern to the waveform of the voltage applied to the device under test (DUT), we show in this work that expanding this bespoke measurement system to be able to simultaneously monitor the waveform of the current going through the DUT allows for the collection of data related to the electrical power transferred to the DUT during the process that displays peculiar features. In the present work, the DUT consists of a classical needle-plane setup. As pressure decreases down from atmospheric levels, the threshold voltage leading up to the apparition of discharges decreases following a trend similar to the classical Paschen curve. Additionally, the nature of the discharge activity transitions from low-amplitude, rapid-firing tightly packed trains of pulses to high-amplitude, longer-lasting and more spread out pulses. Simultaneous measurement of the discharges, applied voltage and current going through the DUT shows that this second type of discharge activity can be synchronous with an asymmetric, distorted current waveform having the same per
Orthogonal time frequency space (OTFS) is envisioned as a highly promising modulation technique due to its superior performance in high-mobility scenarios. Meanwhile, non-orthogonal multiple access (NOMA) stands out a...
详细信息
Model compression is one of the most popular approaches to improve the accessibility of Large Language Models (LLMs) by reducing their memory footprint. However, the gaining of such efficiency benefits often simultane...
详细信息
Model compression is one of the most popular approaches to improve the accessibility of Large Language Models (LLMs) by reducing their memory footprint. However, the gaining of such efficiency benefits often simultaneously demands extensive engineering efforts and intricate designs to mitigate the performance decline. In this work, we leverage (Soft) Prompt Tuning in its most vanilla form and discover such conventionally learned soft prompts can recover the performance of compressed LLMs. More surprisingly, we observe such recovery effect to be transferable among different tasks and models (albeit natural tokenizer and dimensionality limitations), resulting in further overhead reduction and yet, subverting the common belief that learned soft prompts are task-specific. Our work is fully orthogonal and compatible with model compression frameworks such as pruning and quantization, where we enable up to 8× compressed LLM (with a joint 4-bit quantization and 50% weight pruning compression) to match its uncompressed counterparts on popular benchmarks. We note that we are the first to reveal vanilla Parameter-Efficient Fine-Tuning (PEFT) techniques have the potential to be utilized under a compression recovery context, opening a new line of opportunities for model accessibility advancement while freeing our fellow researchers from the previously present engineering burdens and constraints. The code is available at https://***/zirui-ray-liu/compress-thenprompt. Copyright 2024 by the author(s)
With the increasingly complex blockchain technology environment and emerging security threats, the detection and prevention of vulnerabilities in blockchain smart contracts have become crucial for ensuring the healthy...
详细信息
Data selection can be used in conjunction with adaptive filtering algorithms to avoid unnecessary weight updating and thereby reduce computational overhead. This paper presents a novel correntropy-based data selection...
详细信息
暂无评论