Connected Systems Group Seminars

The group runs a bi-weekly seminar series during academic terms. Internal and external speakers give presentations on their work. For more information or to be added to the seminar mailing list, please contact Adam Noel or Subhash Lakshminarayana.

Here is a log of upcoming and previous seminars.

DATE/LOCATION	SPEAKER	TITLE	ABSTRACT
2024-06-05 1pm R0.14	Michael Barros (Essex)	TBA	TBA
2024-05-22 1pm R0.12	Alexandros Paspatis (Manchester Metropolitan University)	TBA	TBA
2024-05-08 1pm A2.05a/b	Nikita Pietrow (Manufacturing Technology Centre - MTC)	Transforming Manufacturing with Digital Twins	Digital twins have the potential to transform many sectors, ranging from energy networks to healthcare, by bringing together the physical world and its digital representation. The manufacturing industry is just one of the many areas that are being revolutionised through application of digital twins, by enabling data-driven decision making and improving production process efficiency and sustainability. However, there are many challenges for their development and adoption, including technical and cultural issues such as integration of legacy equipment and lack of trust between organisations. This talk will cover the past, present and future of digital twins for manufacturing, including the definition of a digital twin and its building blocks, examples of its application in manufacturing, challenges for its adoption, and outlook over the next 10 years.
2024-04-24 1pm MS.04	Pit Hofmann (TU Dresden)	Simulation of Microfluidic Molecular Communication using OpenFOAM	The convergence of molecular communication and microfluidics towards applications in the Internet of Bio-Nano Things (IoBNT) has attracted attention in the research community. Conceivable fields of application are in medicine and biology, e.g., targeted drug delivery, nanosurgery, or treatment of cancer cells. How can we imagine such applications? What are the challenges from a scientific point of view, and what is the current state-of-the-art? In the talk, the basics of microfluidic molecular communication will be briefly explained from a communication technology perspective, current research results will be presented, and possible applications will be given. The audience will also get an insight into the simulation of a microfluidic simulation, starting with particle-based simulations and moving up to the open-source software tool OpenFOAM for more complex simulations.
2024-04-03 11am D2.02	Sami AlSaadi (Warwick)	Deep learning approach based on CNN with a new regularization method for intrusion detection systems in SDN	Software-Defined Networking (SDN) introduces a paradigm shift from traditional networking by decoupling control from data management hardware. This architectural approach offers numerous benefits such as enhanced programmability, flexibility, adaptability, and elasticity, which are challenging to achieve in current network architectures. However, the adoption of SDN also exposes the network to various threats and attacks, which can compromise its vulnerable surfaces. Were SDN control to be compromised, an attacker could overwhelm the whole network by causing significant damage to an organisation's network. To mitigate such risks and protect the SDN environment, the deployment of a robust network intrusion detection system (NIDS) is crucial. A NIDS serves as a vital tool for identifying and thwarting malicious activities and anomalous traffic within the SDN network. Deep Learning (DL) has demonstrated promising capabilities in tackling a wide range of challenges, including text, audio, and computer vision tasks. While several studies have explored the application of DL techniques for NIDS, they are often limited in their effectiveness due to issues such as overfitting and the utilization of outdated datasets. Consequently, these limitations hamper the supervised DL model's ability to detect unseen attacks, thereby endangering the overall security of the SDN environment. To address these challenges and enhance network security, it is imperative to develop a comprehensive solution that addresses the security issues inherent in SDN. In this study, we propose a novel hybrid supervised learning approach that leverages 2-dimensional convolutional neural networks (2D-CNNs) to effectively distinguish between normal and attack traffic flows within the SDN environment. To mitigate the issue of overfitting and improve the detection capabilities of the NIDS, we introduce a new regularization technique called MReg. This utilizes mean deviation theory within the weight matrix to enhance the model's generalization capabilities and combat overfitting. We evaluate the proposed approach using the most recent InSDN dataset, specifically designed for evaluating NIDS techniques within the SDN environment. The evaluation results demonstrate that the MReg regularization method outperforms previous regularization techniques, effectively addressing the overfitting issue. Furthermore, a hybrid model employing the MReg regularization technique exhibits superior performance across all evaluation metrics compared to using a single supervised learning approach. These findings highlight the efficacy of our proposed approach in improving NIDS detection accuracy within the SDN context, thereby strengthening network security.
2024-04-03 11am D2.02	Yulin Wang (Warwick)	Machine Learning-Based Optical Fibre Channel Modelling	Channel modelling has a tremendous impact on optical communications by providing a mathematical representation of the optical transmission medium and its effects on the signal, which is essential for understanding, designing, optimising, and simulating optical communication systems. Conventional channel modelling is conducted through the split-step Fourier method (SSFM), a technique that simulates the channel by resolving the Nonlinear Schrödinger equation (NLSE). However, SSFM requires a substantial number of iteration steps, which implies extremely high computational complexity. Owing to such high computational complexity, Machine Learning-based models have been introduced to directly approximate the optical fibre channel, and these techniques can be mainly divided into data-driven and principle-driven. ML-based channel modelling has been widely used in End-to-end deep learning (E2EDL), Digital Twins (DT) and advanced study in fibre nonlinearities. In this seminar, the background and overview of fibre channel modelling will be introduced first. Subsequently, how data-driven and principle-driven models work in optical fibre will be explained, whilst the respective merits and drawbacks will be assessed. Concluding, the challenges and research directions for ML-based channel modelling will be discussed.
2024-03-20 11am A2.05b	Dr Yansha Deng (King's College London)	Task-oriented Semantics-aware Communications in 6G Era	Inspired by Shannon’s classic information theory, Weaver and Shannon proposed a more general definition of a communication system involving three different levels of problems, namely, (i) transmission of bits (the technical problem); (ii) semantic exchange of transmitted bits (the semantic problem); and (iii) effect of semantic information exchange (the effectiveness problem). The first level of communication, which is the transmission of bits, has been well studied and realized in conventional communication systems based on Shannon’s bit-oriented technical framework. However, with the massive deployment of emerging devices, including Extended Reality (XR) and Unmanned Aerial Vehicles (UAVs), diverse tasks with stringent requirements pose critical challenges to traditional bit-oriented communications, which are already approaching the Shannon physical capacity limit. This imposes the Sixth Generation (6G) network towards a communication paradigm shift to semantic level and effectiveness level by exploiting the context of data and its importance to the task. An explicit and systematic communication framework incorporating both semantic level and effectiveness level has not been proposed yet. Thus, my talk will discuss our recent results related to task-oriented communications for future 6G wireless networks, where I will focus on task-oriented and semantics-aware communication solutions for the virtual reality data type and control and command data.
2024-03-20 11am A2.05b	Dr Dadi Bi (King's College London)	Real-Time Signal Processing via Chemical Reactions for Microfluidic Molecular Communication Systems: Design, Analysis, and Prototype	Molecular communication (MC) is an emerging interdisciplinary field that explores the exchange of information using chemical molecules, mimicking the communication processes found in biological systems. MC research holds immense potential in emerging applications, such as medicine and biosensing, where traditional electromagnetic-based communications would be either unsafe or impractical. While MC theory has had major developments in recent years, more practical aspects in designing components capable of MC functionalities remain less explored. In this talk, we will share our work on joint design of chemical reactions and microfluidic systems to achieve various signal processing functions for MC. We will start from introducing a five-level architecture for describing a microfluidic circuit and identify some components at each level, which demonstrates the signal processing capabilities of chemical-reaction-based microfluidic circuits and facilitates new circuit design. We will then show how this architecture can be used to realize concentration shift keying (CSK) modulation-demodulation functionalities and how communication theory can be applied to predict the outputs of microfluidic circuits. Finally, we will share our microfluidic molecular communication (MIMIC) testbed. We will not only introduce the chemical selection, platform implementation, and experimental methods, but also show and discuss the key insights from experimental results.
2024-03-06 1pm A2.05a	Dr Joseph Anande (Warwick)	Enhanced Modelling for Multi-class Classification with MoSELA	This study introduces the Model Stack Ensemble Learning Architecture (MoSELA), an advanced framework designed to address the limitations of traditional Machine Learning (ML) methods in multi-class classification tasks, particularly those affected by data complexities such as imbalance, skewness, and high dimensionality. MoSELA leverages enhanced Ensemble Learning (EL) techniques to mitigate the computational inefficiencies and performance drawbacks observed in existing stack EL models. Our research focuses on the application of MoSELA for the prediction, detection, and mitigation of malicious patterns in cyber data, demonstrating its superior efficiency and effectiveness over optimized gradient boosting models. The proposed framework achieves up to a 26.7 times increase in training speed and an 89.66% reduction in loss, culminating in a 100% success rate in classification and zero misclassification cases. These results underscore MoSELA's potential in improving multi-class classification accuracy and handling diverse cyber threat patterns more optimally.
2024-02-21 1pm D2.02	Dr Son Dinh-Van (Warwick)	Rapid Beam Training at Terahertz Frequency with Contextual Multi-Armed Bandit Learning	Terahertz (THz) frequency technology holds great promise for enabling high data rates and low latency, essential for manufacturing applications within Industry 4.0. To achieve these, beam training is necessary to enable MIMO communications without the need for explicit channel state information (CSI). In this context, the Multi-Armed Bandit (MAB) algorithms are able to facilitate online learning and decision-making in beam training, eliminating the necessity for extensive offline training and data collection. In this paper, we introduce three algorithms to investigate the applications of MAB in beam training at Terahertz frequency: UCB, Loc-LinUCB, and Probing-LinUCB. While UCB builds upon the well-established Upper Confidence Bound algorithm, Loc-LinUCB and Probing-LinUCB utilize the location of the user equipment (UE) and probing information to enhance decision-making, respectively. The beam training protocols for each algorithm are also detailed. We evaluate the performance of these algorithms using data generated by the DeepMIMO framework, which simulates abrupt changes and various challenging characteristics of wireless channels encountered in realistic scenarios as UEs move. The results illustrate that Loc-LinUCB and Probing-LinUCB outperform UCB, showing the potential of leveraging contextual MAB for beam training in Terahertz communications.
2024-02-21 1pm D2.02	Yanghao Zhong (Warwick)	Detecting Power Module Thermal Resistance Change in Wind Turbine Converters Using Attention-based LSTM-Autoencoder Architecture	This study proposes a technique to monitor the gradual ageing of IGBT power modules in offshore wind turbine converters from the SCADA data using a fusion model of Autoencoder (AE) and attention-based Long-Short-Term Memory Neural Network (AT-LSTM). Power electronic converters in wind turbines operate in complex conditions, and the device junction temperatures are difficult to detect directly. The junction temperature variation is directly affected by the module's thermal resistance, which changes with ageing. Monitoring the ageing process based on external measurements is crucial, but the variability of operating conditions presents challenges that can be addressed using machine learning. This article uses simulated temperature results based on real SCADA wind speed data. It employs Deep Neural Network (DNN) and AT-LSTM neural networks to dynamically predict the junction temperature variations of power modules, with AE reconstruction error used to detect abnormal thermal resistances. The result of comparison with DNN is that AT-LSTM significantly improves the generalization ability of its usage. It also enhances the problem of low prediction accuracy of LSTM multi-time step. AT-LSTM combined with AE is more suitable and effective for monitoring IGBT's long-term ageing.
2023-11-29 1pm R0.12	Sajjad Maleki (Warwick)	The Impact of Load Altering Attacks on Distribution Systems with ZIP Loads	Load-altering attacks (LAAs) pose a significant threat to power systems with Internet of Things (IoT)-controllable load devices. This research examines the detrimental impact of LAAs on the voltage profile of distribution systems, taking into account the realistic load model with constant impedance Z, constant current I, and constant power P (ZIP). We derive closed-form expressions for computing the voltages of buses following LAA by making approximations to the power flow as well as the load model. We also characterize the minimum number of devices to be manipulated in order to cause voltage safety violations in the system. We conduct extensive simulations using the IEEE-33 bus system to verify the accuracy of the proposed approximations and highlight the difference between the attack impacts while considering constant power and the ZIP load model (which is more representative of real-world loads).
2023-11-29 1pm R0.12	Junqiu Wang (Warwick)	Ultra-Massive MIMO Integrated Sensing, Communication And Edge Computing at Terahertz Band with Hardware Impairment	Spectrum scarcity has become one the main bottlenecks of wireless communications systems as entering future sixth-generation (6G). Thus, utilizing the THz frequencies (0.1-10 THz) that has much larger bandwidths than the mmWave is expected to be a key technology to mitigate this severe challenge. In the Thz communications, Tb/s data rates could be readily achieved via its hundreds of GHz bandwidth, significantly facilitating the whole system's efficiency without dramatic change of the physical-layer architecture. Since applications at THz have been proved powerful in both communications and radar, the performance of current integrate radar sensing and communications functionalities or joint radar-communications (JRC) architectures might be greatly enhanced to further facilitate spectrum sharing, improve resource usage and pencil beamforming as well as reduce hardware cost. Also, the user terminal can offload its radar sensing information to the BS for edge computing, which provides highly accurate and shorter delayed sensing result. In view of the emerging new ideas and technologies, our work focuses on an ultra-massive MIMO structure for integrated sensing, communication and edge computing (ISCAC) at THz band considering the effect of I/Q imbalance and phase noise. The Informer deep learning model of Thz channel estimation based on basis equation method and tensor decomposition is proposed. After the estimation, we propose the design algorithm for optimizing the hybrid analog and digital beamforming to further enhance the ISCAC's performance.
TBA (re-schedule)	Prof Siraj Ahmed Shaikh (Swansea)	Economics of Critical National Infrastructure Protection and Security	This talk will introduce a recently launched initiative to explore the economics of security in critical national infrastructure (CNI) systems. The aim is to address both (1) macroeconomics, in terms of incentives and penalties around CNI security. This includes how decisions of operators, suppliers and governments affect key measures of risks, incidents and disruptions. This extends to understanding CNI supply chains (in terms of self-sufficiency and national security) as part of strategic industries, and where may tax breaks, subsidies and protection from foreign competition be effective; and (2) microeconomics, studying the impact of digital and physical lock-ins, externalities and asymmetries (arising out of technologies and vendor-customer relationships), alongside organisational models of investment returns on spending on raising security awareness, and improving security behaviours and culture.
2023-10-30 11am MS.04	Dr Jason Avramidis (OakTree Power)	Unlocking Micro-Flexibility from Residential Energy Sources: Academic Theory, Industrial Reality, and UK's experience	Research on unlocking domestic flexibility has exploded in the last 20 years. However, research efforts are mostly one-sided, adopting biased customer-only or DSO-only viewpoints and losing sight of the real-life constraints that are prevalent. The academia-industry gap thus continues to grow. This talk will give an academic overview of different viewpoints for designing domestic flex markets: customer-driven, grid-driven, and collaborative. It then evaluates the realism of common academic assumptions, and discusses areas requiring further investigation. It finally presents the current state of domestic flex in the UK, the approach of DSOs and aggregators, and their successes so far.
2023-10-18 1pm S0.18	Dr Hua Yan (Warwick)	Overhead Line Sagging Monitoring Using 5G Signals	All overhead lines in the GB transmission network must maintain statutory clearances to ground, roads, and other objects. To maintain these clearances the line sag needs to be monitored. Also, if the line sag can be monitored easily and with great frequency (dynamically), it is possible to provide valuable inputs to the thermal rating of the overhead line. Current methods use either sensors installed on the line to directly measure temperature/sag or weather stations nearby to indirectly calculate temperature/sag. Sensor installation generally requires circuit outage and demands highly skilled engineers as well as access to land where the lines and towers are located. The access is sometimes very restricted and hence brings many challenges when apply. Although sensors can achieve high accuracy of monitoring, the cost is high. Weather stations do not require sensor installation but have similar limitations of installing equipment on the tower and have relative low accuracy. In this talk, a new method by exploiting the 5G signals to directly monitor and measure the line sagging is introduced.
2023-10-04 1pm S0.18	Nadezhda Briantceva (KAUST, Saudi Arabia)	Mobility in Molecular Communication	Join Nadezhda Briantceva as she talks about the ever-evolving field of Molecular Communication (MolCom), where various vital subjects meet, including but not limited to biology, mathematics, and computer science. In this talk, she breaks down the concept of mobility in MolCom, explaining how the movement of system components such as transmitter (TX) and receiver (RX) affects the MolCom networks. Along with that, she will mention the problems where networks must navigate complex structure environments. For example, in targeted drug delivery systems, where the drug molecules must go through the biological tissues. Here, the role of anomalous diffusion will be introduced, a phenomenon that takes center stage in such settings. Through easy-to-follow explanations, she will detail how this kind of unpredictable movement impacts the communication capabilities of the whole MolCom system. Finally, she ties it all together with a look at how numerical approaches are being used to solve the challenges presented by complex channel geometries in MolCom systems. By the end of this session, attendees will have gained insight into the interplay between these significant topics and the potential they hold for the future of healthcare and technology.
2023-09-19 1pm A0.23	Dr Maximilian Schäfer (Erlangen, Germany)	The Internet of BioNanoThings - Applications, Components and Future Directions	Molecular communication (MC) is an emerging field of research at the intersection of natural science and engineering and it should enable the communication between nano machines and facilitate interaction with biological systems. In the future, in-body devices are expected to communicate with each other by means of MC and are connected to the Internet to form the Internet of BioNanoThings (IoBNT). In contrast to existing Internet of Things (IoT) devices for external health monitoring, IoBNT devices form an in-body communication network that enables localized diagnosis and personalized treatment on the organ or even single cell level. The successful realization of the IoBNT would be an important step towards many innovative applications in the medical sector and especially in the field of health monitoring and personalized treatment. However, the successful realization of the IoBNT requires interdisciplinary research concepts with a particular focus on realizable systems. This talk introduces several envisioned applications of the IoBNT and discusses the basic challenges towards its realization. A particular focus is on the identification and development of practical components to be used as communication nodes. Finally, the talk gives specific examples for the interdisciplinary research on the IoBNT at FAU Erlangen-Nürnberg.
2023-06-28 11am S0.20	Prof Justin Coon (Oxford)	Some Recent Mathematical Results on Graph Compression	Many datasets found in practical applications contain correlation structures that can be represented using a graph formalism. Consequently, the interest in graph compression has grown in recent years. In this talk, I will provide a brief overview of graph compression as viewed through an information theoretic lens, beginning with a review of basic concepts related to lossless and lossy compression. I will then focus on some recent results pertaining to random geometric graphs and the stochastic block model.
2023-06-14 1pm H0.60	Syed Bilal Tirmizi (Warwick)	Hybrid Satellite–Terrestrial Networks toward 6G: Key Technologies and Open Issues	Current terrestrial systems may not be sufficient to provide services that require higher data rates with reliability and up to date QoS standards, especially in the harsh oceanic geographic regions critical for maritime communication networks (MCNs), as they suffer from coverage and capacity issues. Additionally, owing to economic and geographic constraints, terrestrial networks are mainly deployed in developed areas, such as urban areas. There are still large numbers of people and devices that remain unconnected even after construction of the 5G network. With the deployment of the 5G network, research on the 6G wireless network comes into focus to overcome these unsolved challenges. The International Telecommunication Union (ITU) has built the Network 2030 group for developing the next-generation wireless network. China has established a project to study the 6G wireless network for 2030 and beyond. In the development targets of the 6G network, peak data rate is expected to reach 100 Gb/s to 1 Tb/s, which is 10 to 100 times higher than that of the 5G network. Latency is expected to decrease to 0.1 ms, which is a tenth of that of the 5G net- work. Additionally, other targets include higher positioning accuracy, higher energy efficiency, extreme reliability, larger connectivity density, and longer battery life. In the 6G white paper, it has been proposed that the future wireless network must be able to seamlessly interface with terrestrial and satellite networks. Non-terrestrial networks and hybrid networks will play an important role in achieving this.
2023-06-14 1pm H0.60	Ula Hijjawi (Warwick)	Automated Solar Cell Defect Detection Using Artificial Intelligence	The installation of photovoltaic (PV) plants has led to the exponential growth of solar cell deployment worldwide. The number of sites generating electricity from solar PV sites in the UK boomed from 28,958 in 2010 to 1,048,328 by the end of 2020. However, defects in solar modules may degrade the efficiency of the solar module by disseminating power, or even consuming power stored in the battery bank. Manual inspection is a time-consuming task and requires intensive analysis efforts of images captured by remote cameras. Therefore, it is crucial to identify a set of automated defect detection approaches for predictive maintenance and condition monitoring of PV modules. Recent state-of-the-art research has focused on Artificial Intelligent (AI) techniques for condition monitoring of solar cells, particularly, for accurately detecting and localising manufacturing as well as after-installation defects. In this research, novel AI-based classification algorithms are investigated and developed to perform secure, accurate and robust automated defect detection for solar cells. Moreover, other challenges related to data scarcity of defective solar modules are discussed throughout data augmentation and unsupervised learning methods.
2023-05-17 1pm H0.60	Wenxiu Hu (Warwick)	Non-coherent signal detection methods for wireless ultraviolet communications	Two types of no-coherent detection methods to counteract the ISI effect of wireless ultraviolet communications will be introduced: optimal weighted non-coherent detection (OWNCD) scheme and high dimensional non-coherent detection (HDNCD) scheme, both of which do not rely on the exact channel statement information (CSI), but devote to the signals’ geometric features from the received UV signals. For OWNCD, we linearly combine these extracted features with optimal wights that minimizes theoretical BER. For HDNCD, we construct a new set using these features, deduce the theoretical detection surface, and prove its superiority to OWNCD.
2023-05-17 1pm H0.60	Dr Ibrahim Isik (Warwick)	Biophysical Model for Signal-Embedded Droplet Soaking into 2D Cell Culture	Using agar plates hosting a 2D cell population stimulated with signaling molecules is crucial for experiments such as gene regulation and drug discovery in a wide range of biological studies. In this paper, a biophysical model is proposed that incorporates droplet soaking, diffusion of molecules within agar, cell growth over an agar surface, and absorption of signaling molecules by cells. The proposed model describes the channel response and provides valuable insights for designing experiments more efficiently and accurately. The molecule release rate due to droplet soaking into agar, which is characterized and modeled as the source term for the diffusion model, is derived. Furthermore, cell growth is considered over the surface, which dictates the dynamics of signaling molecule reactions and leads to a variable boundary condition. As a case study, genetically-modified 𝐸. 𝑐𝑜𝑙𝑖 bacteria are spread over the surface of agar and Isopropyl-beta-D thiogalactopyranoside (IPTG) is considered as a signaling molecule. IPTG droplets are dropped onto the bacteria-covered agar surface. The parameters for the IPTG molecule release rate as a diffusion source into the agar are estimated from this experiment. Then, a particle-based simulator is used to obtain the spatio-temporal profile of the signaling molecules received by the surface bacteria. The results indicate that the number of molecules reacting with or absorbed by bacteria at different locations on the surface could be widely different, which highlights the importance of taking this variation into account for biological inferences.
2023-05-03 1pm H0.60	Kusuma Wardana (Warwick)	Tiny Machine Learning for a Low-cost Air Quality Monitoring Device	Tiny machine learning (tinyML) is a cutting-edge field of artificial intelligence. This paradigm brings machine learning (ML) algorithms to resource-constrained devices, such as microcontrollers. However, microcontrollers have limited memory capacity and computational capabilities. Thus, effective deployment of tinyML models requires a thorough understanding of hardware, software, algorithms, and applications. Regardless of their limited performance, microcontrollers can gather physical environment data through sensors and perform decisions based on ML algorithms. Recent research has also demonstrated the feasibility of low-cost sensor nodes for air quality monitoring systems. This emerging sensor-based air quality monitoring field can provide high-density spatiotemporal pollution data, supplementing the established methodology with more precise and expensive devices. In many applications, low-cost air quality monitoring devices leverage the capabilities of microcontrollers. With the immense volume of spatiotemporal data collected by low-cost air quality devices, there is a better opportunity to apply tiny machine learning techniques in air quality research.
2023-05-03 1pm H0.60	Mahir Taher (Warwick)	Modelling neuronal trafficking to be verifiable	Dementia caused by Alzheimer’s disease is an insidious process, and damage to the brain can often go unnoticed until cognitive impairment manifests itself in an obvious and diagnosable manner (e.g., memory loss). Whilst everyone ages and whilst ageing is a risk factor for Alzheimer’s, dementia is not a necessary outcome of growing old. The progression of Alzheimer’s disease is not well characterised, and whilst there currently isn’t a cure for Alzheimer’s, there are mitigating therapeutics which are more effective when implemented sooner rather than later through early detection. The packaging and transport of cargos within neurons are amongst the first mechanisms to be perturbed by Alzheimer's. Unfortunately, how this neuronal trafficking degrades over a realistic time course (years) remains unclear. Modelling this trafficking aims to inform how neurons fail to communicate important cargos, and making this model trainable and verifiable with experimental data is quit essential for its confidence.
2023-03-15 1pm H0.60	Dr Michael Schmuker (Hertfordshire)	Decoding sub-second gas dynamics with metal-oxide sensors	Many animals have a remarkable ability to track a scent and locate its source. For example, male moths can locate mating partners across large distances on the basis of minute quantities of pheromone that the female releases into the wind. Since odors in natural environments are mostly dispersed via turbulent air currents, there is no useful odor gradient that could be used for navigation. Instead, animals have to infer information from brief and intermittent odor encounters. To achieve this, their olfactory system has evolved to detect odor pulses on time scales in the range of tens of milliseconds. This suggests that low detection latency and rapid odor recognition are required for successful odor search using robots. However, metal-oxide (MOx) sensors, which are often used in robotic tasks, typical exposure times range from tens of seconds to minutes. In my talk I will show how sub-second precision reporting of gas concentration changes can be achieved with off-the-shelf MOx sensors though signal processing and optimisation of sensor periphery. We explore how fast MOx-responses enable robotic gas-based searching algorithms inspired by animal behaviour.
2023-03-01 1pm H0.60	Dr Leslie Kanthan (TurinTech)	Optimizing Signal Processing with evoML: A Machine Learning-Based Platform	Signal processing has become increasingly crucial in today's data-driven world. Machine learning approaches, in particular, offer more precise and efficient solutions to signal processing. The process of building a machine learning model involves selecting relevant features and engineering them to predict the behaviour of the target feature. However, the traditional machine learning model development can be lengthy and resource-intensive, taking months to go from conceptualization to deployment and requiring significant costs. evoML aims to streamline the machine learning model development process while adding additional functionalities such as multi-objective optimization and code optimization. This platform not only automates the development process but also provides the source code of the model, resulting in greater transparency compared to traditional black box models. This talk will delve into: (1) the convergence of machine learning and signal processing for improved outcomes; (2) the challenges of using machine learning for signal processing; (3) how evoML mitigates these challenges; (4) a demonstration of evoML with a relevant use case. Join us as we explore the potential of evoML for revolutionizing the field of signal processing.
2023-02-15 1pm H0.58	Yanahan Paramalingam (Warwick)	Communication in Biofilm	In nature, bacteria are often found in enclosed colonies called biofilms. Biofilm includes extracellular polymeric substance (EPS), which acts as a protective layer. Biofilms are an issue in healthcare since they contribute to antimicrobial resistance. Bacteria are able to chemically communicate cell-to-cell via extracellular signalling molecules called autoinducers. This process is called quorum sensing (QS) and relies on the production of and response to the autoinducers. Soluble signalling molecules are transported throughout the biofilm through both water channels and the EPS. One way proposed to disrupt biofilm is to interrupt the quorum sensing process. A greater understanding of molecule propagation, including that of autoinducers, could lead to strategies to improve the inhibition effectiveness of antimicrobial agents. Such improvements could increase our capability to disrupt and disperse the bacteria in biofilms.
2023-02-15 1pm H0.58	Chenguang Liu (Warwick)	Knowledge distillation based semantic communications for multiple users Deep learning has shown great potential in revolutionizing the traditional communications system	Many applications in communications have adopted deep learning techniques due to their powerful representation ability. However, the learning-based methods can be dependent on the training dataset and perform worse on unseen interference due to limited model generalizability and complexity. In this paper, we consider the deep semantic communications system with multiple users, where there is a limited number of training samples and unexpected interference. To improve the model generalization ability and reduce the model size, we propose a knowledge distillation based system with Transformers and fully connected neural networks. Transformer based encoder-decoder is implemented as the semantic encoder-decoder and fully connected neural networks are implemented as the channel encoder-decoder. Specifically, four types of knowledge transfer and model compression are analyzed. Several system and model parameters are considered, including the level of noise and interference, the number of interfering users and the size of the encoder and decoder. Numerical results demonstrate that knowledge distillation significantly improves the robustness and the generalization ability when applied to unexpected interference, and reduces the performance loss when compressing the model size.
2023-02-01 1pm S0.13	Dr Yongxu Zhu (Warwick)	Stochastic Geometry Analysis of Large Intelligent Surface-Assisted Millimeter Wave Networks	Reliable and efficient networks are the trend for next-generation wireless communications. Recent improved hardware technologies – known as Large Intelligent Surfaces (LISs) – have decreased the energy consumption of wireless networks, while theoretically being capable of offering an unprecedented boost to the data rates and energy efficiency (EE). In this paper, we use stochastic geometry to provide performance analysis of a realistic two-step user association based millimeter wave (mmWave) networks consisting of multiple users, transmitters and one-hop reflection from a LIS. All the base stations (BSs), users and LISs are equipped with multiple uniform linear antenna arrays. The results confirm that LIS-assisted networks significantly enhance capacity and achieve higher optimal EE as compared to traditional systems when the density of BSs is not large. Moreover, there is a trade-off between the densities of LIS and BS when there is a total density constraint. It is shown that the LISs are excellent supplements for traditional cellular networks, which enormously enhance the average rate and area spectral efficiency (ASE) of mmWave networks. However, when the BS density is higher than the LIS density, the reflected interference and phase-shift energy consumption will limit the performance of LIS-assisted networks, so it is not necessary to employ the LIS devices.