In recent years, neural network-based black-box modeling of nonlinear audio effects has improved considerably. Present convolutional and recurrent models can model audio effects with long-term dynamics, but the models require many parameters, thus increasing the processing time. In this paper, we propose KLANN, a Koopman-Linearised Audio Neural Network structure that lifts a one-dimensional signal (mono audio) into a high-dimensional approximately linear state-space representation with nonlinear mapping, and then uses differentiable biquad filters to predict linearly within the lifted state-space. Results show that the proposed models match the high performance of the state-of-the-art neural models while having a more compact architecture, reducing the number of parameters by tenfold, and having interpretable components.

A document by KHRITISH SWARGIARY. Click on the document to view its contents.

This paper describes the process followed to implement a system to characterize the complex permittivity of materials in the 10-330 GHz frequency band. Firstly, the method used and the system's calibration process are shown, consisting of a double calibration TRL (Thru-Reflect-Line) and GRL (Gated-Reflect-Line). Subsequently, a smoothing technique is used to improve the accuracy of the results. Finally, a test is performed on quartz and glass fiber samples, showing that the results are quite reliable over the entire measured bandwidth.

Age-related cognitive impairment is generally characterized by gradual memory loss and decision-making difficulties. The aim of this study is to investigate multi-level support and suggest relevant helping means for the elderly with mild cognitive impairment as well as their caregivers as the primary end-users. This work reports preliminary results on an intelligent assistive system achieved through the integration of the Internet of Things, augmented reality, and adaptive fuzzy decision-making methods. The proposed system operates in different modes, including automated and semi-automated modes. The former helps the user complete their daily life activities by showing augmented reality messages or making automatic changes; while the latter allows manual changes after the real-time assessment of the user’s cognitive state based on the augmented reality serious game score. We have also evaluated the accuracy of the serious game score with 37 elderly participants and compared it with users’ paper-based cognitive test results. We further noted that there is an acceptable correlation between the paper-based test and users’ serious game scores. Moreover, we observed that the system response in

Advances in Internet of Things (IoT) technologies have resulted in a significant surge in the utilization of sensor devices across diverse domains for environmental sensing and monitoring. The applications of IoT sensor devices in environmental monitoring span a wide range, including the surveillance of biodiverse areas such as peatlands, forests, and oceans, as well as air quality monitoring, commercial agriculture, and the safeguarding of endangered species. This paper provides a long term evaluation of IoT sensors data quality in environmental monitoring networks, particularly focusing on peatland regions. IoT sensors have the capacity to provide high resolution spatiotemporal dataset in environmental monitoring networks. Sensor data quality plays significant role in increasing the adoption of IoT devices for environmental data gathering. However, due to the nature of deployment (i.e., in harsh and unfavourable weather conditions), coupled with the limitations of low-cost components, IoT sensors are prone to collection of erroneous data, also the nature of peatland ecosystems presents unique challenges in data quality assurance due to their complex and dynamic characteristics. This paper identifies specific challenges and issues related to IoT sensor data quality in different peatland ecotopes. These challenges include sensor placement and calibration, data validation and fusion, environmental interference, and the management of data gaps and uncertainties. To address these challenges, the paper presents and evaluates methods for improving data quality in peatland monitoring networks. These methods encompass advanced sensor calibration techniques, data validation algorithms, machine learning approaches, data processing and data fusion strategies.

 This paper provides a visual based conceptual model and framework to help explain the behavior of electrical, electromagnetic and gravitational phenomenon.  The model is conceptual only and the hypothetical particles and interactions presented are simply designed as a way to visualize and understand electromagnetic interactions.  Further research needs to be done exploring the possibility that the proposed hypothetical particle and interactions may actually exist.  Future improvements to the model may need to be made to help it conform to new scientific discoveries.  Examples are provided showing how the framework conforms to many of the behaviors of electricity, electromagnetism and radiation.  A hypothetical particle is introduced, referred to as a sub-photon by this paper.  The proposed framework relies on previous research in controversial physical models and assumptions such as repulsive gravity and a fractal universe.  Despite the reliance on controversial concepts, the model is intuitive in many respects, easy to visualize and not dependent on difficult mathematics.  The model can be used as a visual tool to help explain very complex interactions with minimal mental effort.  Whether or not the existence of the sub-photon particle and its interactions are later supported by scientific literature, the model can remain a useful way to visualize and understand basic electromagnetic interactions.  This paper includes predictions about celestial bodies and gravitational waves that can be used to test the model in future research.Â

Ulcerative colitis is a chronic disease characterized by bleeding and ulcers in the colon. Currently, the gastroenterologist reviews the colonoscopy video to assess the disease severity using an endoscopic score. This task is time-consuming and does not consider the size and the number of lesions. Consequently, automatic detection methods were proposed enabling fine-grained assessment of lesion severity. However, they depend on the quality of the training set, and its specificity to the application context. To suit the local clinical setup, we opted for an internal training dataset containing only rough bounding box annotations around lesions. Color information is the primary indicator used by specialists to recognize the lesions. Thus, we propose to use linear models in suitable color spaces to detect lesions. We introduce an efficient sampling scheme for exploring the set of linear classifiers and removing trivial models i.e. those showing zero false negative or positive ratio. Using bounding boxes leads to ex- aggerated false negative/positive ratios due to mislabeled pixels, especially in the corners, resulting in decreased modelsâ\euro™ accuracy. Therefore, we propose to evaluate the model sensitivity on the annotation level instead of the pixel level. Our sampling strategy can eliminate up to 25% of trivial models. Despite the limited annotationsâ\euro™ quality, the detectors achieved good performance (93% specificity/89% sensitivity for bleeding and 57% specificity/83% sensitivity for ulcers). The best models exhibit low variability when tested on a small subset of endoscopic images. However, the inter-patient model performance was variable suggesting that appearance normalization is critical in this context.

Named Entity Recognition (NER) task aims at automatically recognising and classifying named entities in a given natural language input. Majority of the studies related to Named Entity Recognition are focused on English language. Named Entity Recognition in Indian languages is challenging due to the complex grammar and morphology of Indian languages and the scarcity of good quality labelled data. The dissimilarity between the literary and spoken versions of these languages is also a big challenge regarding usability of NER models. Kannada is such an Indian language for which the task of NER is still an active area of research. Usage of Deep learning, especially Transfer learning using Transformer models has drastically improved performance of many NLP tasks at scale. However, Transfer learning still requires considerable amount of data for the required task. In case of low resource language like Kannada, very few labelled datasets are available publicly and creating one from scratch is expensive in terms of time and labor. Active learning (AL) aims to tackle the labelled data acquisition problem by having the learning model and an Oracle to cooperate. Active Learning iteratively builds an optimally labelled and sufficiently large dataset.  This study focuses on Named Entity Recognition in Kannada language. We explore the application of Active Learning technique to Named Entity Recognition problem in the low-resource language Kannada. Results show that AL can be used to boost a multilingual models performance in fine-tuning for NER. We also try to mitigate the gap between formal and colloquial dialects of Kannada in NER datasets.

High-Performance Computing (HPC) systems are designed for large-scale processing and complex dataset analysis leveraging scalability, efficiency, and parallelism, often integrating specialized hardware structures such as Vector Processing Units (VPUs). As these systems have grown in complexity and scale, their vulnerability to errors and failures has become an important and complex issue in the HPC world. Our research addresses this challenge by exploring and implementing advanced fault tolerance techniques inside the Vitruvius+ architecture, a partial out-of-order Vector Processing Unit. To the best of our knowledge, this is the first full RTL-level implementation of instruction replication in an HPC-class vector processor for reliability. Specifically, we investigate the integration and interaction of redundancy mechanisms inside the most sensitive architectural units, obtaining a reduction of 75% in non-silent faults causing system failure, proven by an extensive fault injection simulation campaign, with a hardware overhead of only 7.5% and a negligible variation in clock frequency.

Green manufacturing has become a pressing issue in recent years, driven by the acknowledgment of its long-term social and economic benefits, increasing demand for environmentally friendly products, and expanding regulations. One of the key approaches to attaining sustainability in green manufacturing is to design long-lasting products. Nevertheless, the traditional approaches have faced several unique challenges in designing a product with an extended lifetime, such as costly and time-consuming procedures, as well as the noisy, sparse, insufficient or incomplete data. This paper proposes a novel framework to tackle these issues from a data-driven perspective. Specifically, the proposed method employs Bayesian optimization with Monte-Carlo acquisition functions to take into account both product design factors and degradation signals and incorporate the inherent uncertainty in the modeling of underlying degradation processes and prediction of product lifetimes. A series of simulation studies are presented to assess the performance of the proposed method. A case study on the Lithium-ion battery dataset is further conducted, which demonstrates the advantages of the proposed method over existing benchmark approaches.

We experimentally show that: in the same classical wave experiment, (1) the light is photons; blob:https://www.techrxiv.org/2498c0ca-197a-4f4a-a278-2d1886734983 (2) near the double slit, although we assume the photons have the same frequency and phase, the photons do not produce the interference pattern and thus, we referrer the states of the photons are semi-coherent, i.e., non-coherent; (3) near and on the screen, the photons distribute as waves and produce the interference pattern and thus, the states of the photons landed on the screen are coherent. blob:https://www.techrxiv.org/ddc7b64b-d226-49db-a98f-1202d997dc4a blob:https://www.techrxiv.org/1fcdd9ed-f1e4-4947-ac51-7d66f2894f51 We referred it as Photon Waves (or Photowave) phenomena. The mystery is how the semi-coherent photons evolve to the coherent photons, i.e., how the non-interference patterns evolve to the interference patterns.

Realistic full-body avatars play a key role in representing users in virtual environments, where they have been shown to considerably improve body ownership and presence. Driven by the growing demand for realistic virtual humans, extensive research on scanning-based avatar reconstruction has been conducted in recent years. Most methods, however, require complex hardware, such as expensive camera rigs and/or controlled capture setups, thereby restricting avatar generation to specialized labs. We propose WILDAVATARS, an approach that empowers even non-experts without access to complex equipment to capture realistic avatars in the wild. Our avatar generation is based on an easy-to-use smartphone application that guides the user through the scanning process and uploads the captured data to a server, which in a fully automatic manner reconstructs a photorealistic avatar that is ready to be downloaded into a VR application. To increase the availability and foster the use of realistic virtual humans in VR applications we will make WILDAVATARS publicly available for research purposes.

The world of artificial intelligence and deep learning is constantly evolving, with many pioneers and researchers utilizing frameworks such as TensorFlow and PyTorch to expedite their research. While these pre-built frameworks offer a swift execution process and alleviate the burden of pure programming, they may also result in a superficial grasp of the intricate mathematical mechanics involved in deep learning. This lack of deep understanding can impede the optimization process and hinder the achievement of optimal performance in developing deep networks. To overcome these challenges, the objective of this paper is to simplify, clarify, and remove obstacles to the mechanics of deep learning networks, streamlining the development process for researchers. In this paper, Our exploration of these networks will include explaining the derivative of various methods and activation functions, providing a deeper insight into the topic .

With the Internet of Things (IoT) generating vast amounts of data, privacy breaches have become increasingly prevalent, exposing individuals to serious risks such as identity theft and life-threatening situations. This research addresses the challenge of identifying cybersecurity threats and vulnerabilities leading to privacy breaches, as evidenced by recent cyber-attacks on Australian Medibank, Optus, and hospital networks. We propose a machine learning (ML)-based approach to distinguish between legitimate and rogue privacy policies, defining fundamental concepts of privacy, security, and access control in the context of personal, confidential, and sensitive information breaches. Our methodology introduces zero-privacy (ZP) and binary question-answer (QA) models to discern legitimate versus illegitimate actions or interests within privacy policies. Our experiments utilise natural language processing (NLP)-based ML models to analyse the linguistics of privacy policies. In experiments conducted on a dataset from the top 100 Forbes-listed companies, including 67 rogue policies, our privacy classification approach demonstrates reliability, accurately distinguishing between legitimate and rogue policies. With a dataset split of 90% for training and 10% for testing, our model achieves accuracy and precision exceeding 94% and 91%, respectively. Additionally, we evaluate the probability of ZP occurrences in organisations’ privacy and service-level agreements, revealing significant privacy breach risks. Through case studies utilising our proposed binary QA model, we underscore the urgent need for enhanced privacy measures across various organisations’ policies. Introducing a novel approach to access control, we specify permissions under conditions of legitimate and rogue privacy policies, exemplifying the applicability of our proposed access control mechanism through security policy modelling.

Polyphone disambiguation has long been a crucial problem for Chinese text-to-speech (TTS) systems. Researchers have greatly improved the performance of neural disambiguation models by leveraging pre-trained BERT models. However, the vast computation of BERT makes its deployment on edge devices extremely hard. To mitigate this problem, we present a hybrid model compression strategy incorporating both network pruning and knowledge distillation for polyphone disambiguation. In contrast to previous methods that compressed the model only through knowledge distillation, our approach can produce compact alternatives by directly pruning an existing BERT, thus remarkably reducing the training cost. Besides, we also provide a new dataset for common Chinese polyphonic characters (CCPC) to facilitate future research. Experimental results demonstrate that our approach can offer efficient disambiguation models with highly comparable performance to the BERT-based baseline.

This study proposes a decentralized collision avoidance method for multiple ships and utilizes information exchange between vessels. The method is formulated with a hierarchical algorithm architecture and communicates with other ships using the route exchange concept to calculate collision-free trajectories. The hierarchical architecture consists of mid-level and reactive trajectory planners and considers navigational rules, e.g. the COLREG, and both static and dynamic obstacles. Dynamic obstacles represent the ships in the scenarios and the algorithm accounts for both cooperative and non-cooperative vessels. The mid-level planner is used for finding effective control actions by combining the A* algorithm with a proactive collision avoidance optimizer based on the MPC. The reactive planner uses the Informed SB-MPC algorithm and serves as a final barrier for finding solutions to problematic scenarios where the mid-level planner solutions are infeasible. The main motivations of the study include 1) to utilize ship-to-ship information exchange to convey intentions and to solve collision avoidance problems and 2) to prevent autonomous ships to initiate aggressive maneuvers in close-range encounters by calculating and applying proactive control actions in advance. The performance of the method is validated via simulations of 22 Imazu problem cases and a case scenario defined in the Trondheim Fjord. The results support that the algorithm can generate collision-free trajectories considering the COLREG rules 8, 13-18 and maintains a minimum safety distance. Finally, the hierarchical algorithm demonstrates computationally efficient run times to be used in real-time applications.

This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.

Free space optical system (FSO) is a wireless communication system which offers unlicensed spectrum with low cost of deployment. Losses caused by atmospheric turbulence are an enormous challenge for FSO systems. These losses are modeled by Gamma-Gamma model. The Gamma-Gamma model is used to represent moderate to strong turbulence. This model considers wind velocity, but does not consider Rainfall parameter for the calculation of refractive index structure parameter Cn2. In this research paper, we have considered effect of heavy rainfall as well as rms value of strong wind velocity which exist in hilly terrains. This modified model for the proposed spectrum-sliced wavelength division multiplexing (SS-WDM) FSO communication link incorporates differential phase shift keying (DPSK) between the popular high altitude Nilgiris Mountains from the state of Tamilnadu, India. Simulation of conventional model and the modified model by considering rainfall parameter as well as wind velocity has been demonstrated using meteorological data. The results in terms of Eye diagrams, and performance metrics such as Bit error rate (BER), log (BER), Q-factor, and Maximum eye height indicate that the modified model is more accurate for maintaining foolproof and efficient FSO communication link.