On July 1st 2024 at 16:00, Panagiotis Gonidakis will defend their PhD entitled “DATA- AND LABEL-EFFICIENT DEEP LEARNING FOR MEDICAL IMAGE ANALYSIS APPLICATION TO LUNG NODULE DETECTION ON THORACIC CT”.

Everybody is invited to attend the presentation in room D.0.03, or digitally via this link.

Convolutional neural networks (CNNs) have been widely used to detect and classify various objects and structures in computer vision and medical imaging. Access to large sets of annotated data is commonly a prerequisite for achieving good performance. In medical imaging, acquiring adequate amounts of labelled data can often be time consuming and costly. Therefore, reducing the need for data and in particular associated annotations, is of high importance for medical imaging applications. In this work we investigated whether we can lower the need of annotated data for a supervised learning classification problem.  

We chose to tackle the problem of lung nodule detection in thoracic computed tomography (CT) imaging, as this widely investigated application allowed us to benefit from publicly available data and benchmark our methods. We designed a 3D CNN architecture to perform patch-wise classification of candidate nodules for false positive reduction. Its training, testing and fine-tuning procedure is optimized, we evaluated its performance, and we compared it with other state-of-the-art approaches in the field.  

Next, we explored how data augmentation can contribute towards more accurate and less data-demanding models. We investigated the relative benefit of increasing the amount of original data, with respect to computationally augmenting the amount of training samples. Our result indicated that in general, better performance is achieved when increasing the amount of unique data samples, or augmenting the data more extensively, as expected. Surprisingly however, we observed that after reaching a certain amount of training samples, data augmentation led to significantly better performance compared to adding unique samples. Amongst investigated augmentation methods, rotations were found to be most beneficial for improving model performance. 

Following, we investigated the benefit of combining deep learning with handcrafted features. We explored three fusion strategies with increasing complexity and assessed their performance for varying amounts of training data. Our findings indicated that combining handcrafted features with a 3D CNN approach significantly improved lung nodule detection performance in comparison to an independently trained CNN model, regardless of the fusion strategy. Comparatively larger increases in performance were obtained when less training data was available. The fusion strategy in which features are combined with a CNN using a single end-to-end training scheme performed best overall, allowing to reduce training data by 33% to 43%, while maintaining performance. Among the investigated handcrafted features, those that describe the relative position of the candidate with respect to the lung wall and mediastinum, were found to be of most benefit. 

Finally, we considered the case in which abundant data is available, but annotations are scarce, and investigated several methods to improve label-efficiency and their combined effect. We proposed a framework that utilizes both annotated and unannotated data, can be pretrained via self-supervision, and allows to combine handcrafted features with learned representations. Interestingly, the improvements in performance derived from the proposed learning schemes were found to accumulate, leading to increased label-efficiency when these strategies are combined. We observed a potential to decrease the amount of annotated data up to 68% when compared to traditional supervised training, while maintaining performance.  

Our findings indicate that the investigated methods allow considerable reduction of data and/or annotations while maintaining model performance for lung nodule detection from CT imaging. Future work should investigate whether these results generalize to other domains, such that more applications that face challenges due to a shortage of annotated data may benefit from the potential of deep learning.

On March 31 2022 at 16.00 Tobias Birnbaum will defend his PhD entitled “A generic source coding. Methodology and architecture of dynamic holograms”.

Everybody is invited to attend the presentation live in room D.2.01 or online via https://us02web.zoom.us/j/84066799755?pwd=YkVseXFaVmlEQVM4ZU1leDBNdFN0Zz09#success

For wave phenomena, the holographic principle describes how, based upon light propagation laws and a recording of the amplitude and the phase of a wave front in one place – called a hologram – a wave front in another place can be obtained. The holographic principle can be applied to, among others any electro-magnetic wave. It has great impact on applications such as holographic microscopy, interferometry and non-destructive testing. Applied to visible light, holograms allow seamless observation of 3D content without any distortions or adversary effects such as mismatching visual cues. At sufficient space-frequency bandwidths, holograms become optically indistinguishable from reality and can be refocused at observation time. When those high-quality holograms became digitally accessible due to advances in processing power in recent years, manipulation, duplication, and computergeneration from purely synthetic content became feasible. Applied to macroscopic content, the most promising applications include preservation of cultural treasures, art, entertainment, educational purposes, medical imaging, surgical assistance, big data visualizations, and computer aided design. However, digital holograms can only convey as much information because of their large space-frequency bandwidths resulting in resolutions of several gigapixel. Thus compression becomes a necessity, especially for dynamic content. As holograms of visible light are based on the interference of diffracted coherent light, they look similar to the patterns visible on the surface of a pond, after throwing a hand full of pebbles into. In a numerical hologram, typically, each point in the scene influences every point in the hologram. Both facts together render signal characteristics of holograms conceptually very different from regular images and videos, and thus novel strategies to compress dynamic holograms need to be investigated.
This PhD thesis consists of several aspects necessary to design such strategies as well as a first proposition of a holographic video codec suitable for multipleindependently objects. Most contributions exploit heavily the concepts of spatial frequency (number of lines per unit length) and optical phase-space (also known as space-frequency or time-frequency domain). The novel contributions include: compression of static Fourier holograms based on wave atoms refinement of a STFT based static compresion scheme suited for all DH types a segmentation of holograms corresponding to scenes of multiple independently moving objects and resulting from it, a generic holographic motion compensation scheme for such scenes. From the latter an inter-frame compression strategy is derived and a generic video compression scheme is proposed. Further contributions concern, various contributions to subjective quality assessment of digital holograms a newly proposed versatile similarity measure for complex numbers and studies on speckle denoising of the back-propagated wave fields with the objective to find lowcomplexity algorithms with acceptable visual performance.

On Ferbruary 2 2021 at 15.00 Volodymyr Seliuchenko will defend his PhD entitled “Active pixels for high dynamic range and 3d imaging applications”.

Our world is being reshaped by machines which are getting closer to humans in perceptive and cognitive abilities enabling previously unimaginable applications Autonomous cars, mobile home assistant robots, drone delivery networks are just a few examples of the emerging disrupting technologies of this brave new world Accelerating trends in computational power availability fuel the evolution of artificial intelligence systems which become capable of digesting more and more information that, for systems interacting with the real world, must come from sensors These emerging mobile robotics applications rely heavily on the image and distance sensors to create awareness about their environment the quality of the sensory data, in most cases, determines the key performance parameters and system safety Real applications are often posing the sensory system challenging conditions pushing the sensor specifications to the limits and often calling for novel sensing and signal processing approaches
In this work, 2 D and 3 D image sensor systems, the key sensor components of mobile robotics, are discussed Firstly, quantum efficiency improvement methods and a method for dynamic range extension of 4 T image pixels that preserves 4 T pixel dark noise performance are proposed These quantum efficiency improvement methods and the dynamic range extension method can be applied to both 2 D and 3 D imaging Further, indirect Time of Flight 3 D image sensors are analyzed, and improved 3 D image sensors based on Current Assisted Photonic Demodulators are proposed Finally, a hybrid Time of Flight method that produces a time domain echo signal using photonic demodulator sensor is proposed and compared to direct Time of Flight methods.

On September 22nd 2025 at 16:00, Thibaut Vandervelden will defend their PhD entitled “RUST-BASED IOT NETWORKS: A NETWORK PROTOCOL AND SECURITY PERSPECTIVE”.

Everybody is invited to attend the presentation in room I.2.01 or online via this link.

The Internet of Things (IoT) continues to transform our interconnected world. Its rapid growth raises significant concerns about privacy and security. In recent years, numerous IoT botnets have exploited vulnerabilities in embedded devices to launch large-scale Distributed Denial of Service attacks. These attacks have caused significant disruption to internet services worldwide.

Many of these security vulnerabilities come from the use of memory-unsafe programming languages, such as C and C++. Programming languages with built-in safety features can mitigate these vulnerabilities. Since its first stable release in 2015, Rust has emerged as a popular choice for system programming, gaining popularity due to its unique combination of memory safety guarantees while keeping its performance comparable to the one obtained with traditional programming languages. Developers have successfully deployed Rust across diverse domains, including Operating Systems (OSs), web services, and embedded devices.

For IoT devices, two software components are particularly important: the OS and the network stack enabling communication. We provide an evaluation of OSs and frameworks for embedded devices available in Rust and examine their suitability for various IoT applications. We also investigate the feasibility and advantages of implementing a complete network stack in Rust for resource-constrained embedded devices. We built on the smoltcp library and extended it with a Rust implementation of IPv6 over Low-power Wireless Personal Area Networks (6LoWPAN). We studied and implemented the IPv6 Routing Protocol for Low-Power and Lossy Networks (RPL). We developed and applied an evaluation methodology to check compliance with the standard and to verify interoperability with existing implementations. Doing so, we solved several issues of the current RPL implementation in ContikiNG. Our research also evaluates the effectiveness of 6LoWPAN generic header compression for use in RPL networks. The results demonstrate that this compression format significantly reduces energy consumption for IEEE 802.15.4-based networks that offer low data rates. For higher data rates, the benefit of this compression format diminishes.

Alongside contributing to networking components in Rust, we also explored security primitives. We evaluated the performance of different Keccak sponge constructions, cryptographic primitives that can serve as building blocks for various security protocols. Our findings reveal that Keccak sponge constructions with lower capacity parameters achieve significantly better efficiency on 32-bit microcontrollers commonly used in IoT devices. Building on these insights, we design and implement two novel symmetric-key-based authentication protocols. One tailored for wireless sensor networks and the other for fog based networks. We demonstrate that our protocol provides robust protection against known attacks while maintaining low computational and communication overhead.

Two guest lecturers of the Hanoi University of Science and Technology (HUST) will be presenting their work On Friday 05/08 at 10 AM, in room K.4.52. You are all cordially invited to attend.

ANSA (Advanced Networks and Smart Applications) is the research group composed of 08 professors and researchers, and quite large number of master students, undergraduates, and several PhD students.

The research group has the main interest focusing at advanced communication technologies and their applications in various fields, e.g. smart city, intelligent transportation systems, smart agriculture, environment monitoring and management,…

Several current topics includes cloud computing, edge computing, SDN-based architecture for task scheduling in edge-cloud computing, Quality of Experience, Network Security, Virtual Reality, Internet of Things (Zigbee, LoRa, NB-IoT), IoV (cellular-based and DSRC-based V2X). 

Presenter: Prof. Thanh Nguyen, Dean of School of Electrical and Electronic Engineering (SEEE) and a senior member of ANSA research group. He will give an introduction of the lab and interesting topics. 

Presenter: Prof. Nguyen Huu Thanh

School of Electrical and Electronic Engineering

Hanoi University of Science and Technology

—

The project aims to provide an Internet of Things (IoT) based parking monitoring and management solution for city-owned parking fields on streets. The system will be an integral part of the Intelligent Transportation System, in which the technology of AI, big data, IoT will be applied to provide innovative services related to transportation and traffic management and enable the user to be better informed and make “smarter” use of transport.

After nearly 3 years of running the project, various outcomes have been achieved. They include the design and implementation of detector devices, smart IoT gateways supporting various communication interfaces (LoRa, 4G, Wifi, Ethernet). Other alternative solution includes cameras and a lightweigh algorithm to detect and quantify free parking lots. Several theoretical results in Vehicular Fog Computing topic, published in a Q1 journal paper, will be presented. 

Presenter: Dr. Phùng Kiều Hà

School of Electrical and Electronic Engineering

Hanoi University of Science and Technology

Arno Hemelhof finalized his first tape-out in InP Teledyne technology, paving the path towards sustainable IC for 6G communication