An automatic safe landing-site detection system is proposed for aircraft emergency landing based on visible information acquired by aircraft-mounted cameras. Emergency an unplanned event in response to situations. If, as usually the case, there no airstrip or airfield that can be reached unpowered aircraft, a crash ditching has carried out. Identifying critical survival of passengers and crew. Conventionally, pilot chooses visually looking at terrain through cockpit. The success this vital...

Condition-based maintenance is believed to be a cost-effective and safety-assured strategy for railroad track management. Implementation of the strongly relies on reliable complete condition data, deterioration models, efficient solvable mathematical models optimal scheduling. In practice, reliability inspection data often in question; therefore, collected need preprocessed before it used implement condition-based strategy. Reliable means accurate positioning noiseless parameter...

Temporal knowledge graph embedding (TKGE) models are commonly utilized to infer the missing facts and facilitate reasoning decision-making in temporal based systems. However, existing methods fuse information into entities, potentially leading evolution of entity limiting link prediction performance TKG. Meanwhile, current TKGE often lack ability simultaneously model important relation patterns provide interpretability, which hinders their effectiveness potential applications. To address...

Since January 1993, the authors have been working to refine and extend Sphinx-II technologies in order develop practical speech recognition at Microsoft. The result of that work has Whisper (Windows Highly Intelligent Speech Recognizer). represents significantly improved efficiency, usability, accuracy, when compared with system. In addition offers input capabilities for Microsoft Windows can be scaled meet different PC platform configurations. It provides features such as continuous...

Identifying high-risk patients is crucial for effective cardiovascular disease (CVD) prevention. It not known whether electronic health record (EHR)-based machine-learning (ML) models can improve CVD risk stratification compared with a secondary prevention score developed from randomised clinical trials (Thrombolysis in Myocardial Infarction Risk Score Secondary Prevention, TRS 2°P).We identified large system, including atherosclerotic (ASCVD), split into 80% training and 20% test sets. A...

Previous works trained the Transformer and its variants end-to-end achieved remarkable translation performance when there are huge parallel sentences available. However, these models suffer from data scarcity problem in low-resource machine tasks. To deal with mismatch between big model capacity of small training set, this paper adds BERT supervision on latent representation encoder decoder designs a multi-step algorithm to boost such basis. The includes three stages: (1) training, (2) (3)...

For the special characters, remote sensing image has higher requirements in content security: it desires not only encryption during storage and transmission for preventing information leakage, but also watermarking after illegal usage copyright protection or even source tracing. Therefore, this paper proposed to integrate based on orthogonal decomposition comprehensive security of image. By method, can achieve operation independence mergence; moreover, there is requirement selecting...

This paper discusses how to compute word-level confidence measures based on sub-word features for large-vocabulary speaker-independent speech recognition. The performance of measure using at word, phone and senone level is experimentally studied. A framework transformation function system proposed high estimation. In this system, discriminative training used optimize the parameters function. comparison baseline, experiments show that reduces equal error rate by 15%, with up 40% false...

Word level confidence measures are of use in many areas speech recognition. Comparing the hypothesized word score to a ‘filler’ model has been most popular measure because it is highly efficient, and does not require large amount training data. This paper explores an extension this technique which also compares scores words that commonly confused for it, while maintaining efficiency low demand The proposed method gives 39% relative false accept rate reduction over ‘filler’model baseline, at...

This paper presents translation-based knowledge graph completion method via efficient relation rotation (TransERR), a straightforward yet effective alternative to traditional models. Different from the previous models, TransERR encodes graphs in hypercomplex-valued space, thus enabling it possess higher degree of translation freedom mining latent information between head and tail entities. To further minimize distance, adaptively rotates entity with their corresponding unit quaternions,...

Incomplete utterance rewriting (IUR) aims to restore the incomplete with sufficient context information for comprehension. This paper introduces a simple yet efficient IUR method. Different from prior studies, we first employ only one-layer MLP architecture mine latent semantic between joint utterances task (MIUR). After that, conduct feature matrix predict token type and thus utterance. The well-designed network make our method significantly superior existing methods in terms of quality...

In this paper, a new fuzzy adaptive maneuvering target tracking algorithm based on current statistic model is proposed. How to track key problem of in clutter. Current statistical needs pre-define the value maximum accelerations targets. So it may be difficult meet all conditions. The Fuzzy inference combined with proposed cope problem. Given error and change last prediction, system on-line determines magnitude acceleration adapt different maneuvers. Furthermore, difficulties lies...

This paper proposes an image combination algorithm of active and passive security protection. The is realized by adjusting the protection information embedding according to Arnold transform's periodicity. By means this algorithm, can be extracted conveniently from image's ciphertext without decoding it still exists in plaintext after decryption. will introduce research background firstly, then, details processed images experimental results are presented. In end, future orientation pointed out.

This paper introduces Unified Language-driven Zero-shot Domain Adaptation (ULDA), a novel task setting that enables single model to adapt diverse target domains without explicit domain-ID knowledge. We identify the constraints in existing language-driven zero-shot domain adaptation task, particularly requirement for IDs and domain-specific models, which may restrict flexibility scalability. To overcome these issues, we propose new framework ULDA, consisting of Hierarchical Context Alignment...

Recent studies have highlighted the effectiveness of tensor decomposition methods in Temporal Knowledge Graphs Embedding (TKGE) task. However, we found that inherent heterogeneity among factor tensors significantly hinders fusion process and further limits performance link prediction. To overcome this limitation, introduce a novel method maps onto unified smooth Lie group manifold to make distribution approximating homogeneous decomposition. We provide theoretical proof our motivation are...

Recent research in zero-shot Relation Extraction (RE) has focused on using Large Language Models (LLMs) due to their impressive capabilities. However, current methods often perform suboptimally, mainly a lack of detailed, context-specific prompts needed for understanding various sentences and relations. To address this, we introduce the Self-Prompting framework, novel method designed fully harness embedded RE knowledge within LLMs. Specifically, our framework employs three-stage diversity...