Category: Uncategorized

In his presentation, entitled ‘Enabling Real-Time Systems & AI at the Edge’, Dr Dave Cavalcanti, Senior Engineer at Intel, will examine the architectural and system-level fundamentals required to enable deterministic real-time computing in conjunction with AI applications on modern edge platforms.

He will focus on time-coordinated computing, mixed-criticality workloads and the convergence of computing and networking as key factors for next-generation cyber-physical systems.

The presentation will discuss hardware, software and networking features, such as Time-Sensitive Networking, benchmarking tools and their role in enabling AI-enhanced real-time systems across various vertical markets.

The speaker

Dave Cavalcanti is a senior engineer at Intel Corporation, with extensive experience in distributed networking systems, connectivity, industry standards and ecosystems. He also serves as chairman of the Avnu Alliance, an industry forum that promotes standards and certification programmes to enable deterministic real-time performance based on interoperable Time-Sensitive Networking (TSN) devices and converged networks.

He obtained a PhD in Computer Science and Engineering in 2006 from the University of Cincinnati, a Master’s degree in Computer Science and a Bachelor’s degree in Electronic Engineering from UFPE in Brazil. He has published over 50 peer-reviewed articles and holds more than 125 granted patents.

In the research project led by Professor Francisco Airton (UFPI), he and his students Iure Fé (PhD), José Miqueias (MSc) and Lucas Lopes (MSc) are developing analytical models based on Petri nets, which are used to mathematically represent any distributed system. They create a diagram describing the system and perform probabilistic calculations to obtain various metrics, using a specific tool for this purpose.

Airton believes that these models can also be interpreted as digital twins. Based on this, the group selects a specific information system — in the case of the PhD student, Kubernetes — and constructs Petri net models that represent the deployment of Kubernetes. After modelling this system, they integrate the model with software capable of running it and monitoring Kubernetes in real time.

The master’s students, meanwhile, use other monitored systems: one works with camera monitoring, and another with a drone simulator.

The team is preparing papers for this year’s SBRC, describing how the digital twin platform can outperform different types of autoscaling. In this case, the system tests different configurations and simulates ‘what-if’ scenarios, which allows the best option to be identified before applying it in the real environment.

In traditional client-server architecture, Kubernetes operates alongside servers and can connect to any client, including IoT devices. On these devices, a set of sensors generates data that varies depending on the context, such as the time of day or traffic flow. This variation in demand requires the Kubernetes deployment configuration to be adjusted dynamically. This is where K8s-DT comes into play, predicting the best new configuration to be implemented.

 

Oscar Mauricio Caicedo Rendon, one of the members of the international research network of INCT ICoNIoT, will release his book “Machine Learning for Network Management” on April 26, 2026. The book is being published by Springer Nature and can be purchased as a complete volume or by individual chapters.

“Machine Learning for Network Management” was written as a guide text for undergraduate students in last year,  master’s and doctoral students in computer science and related areas who are interested in studying the application of machine learning in networks, introducing them to the world of algorithms that are useful for solving network management problems.

Book structure

The book is divided into three parts. The first is conceptual, covering ML and network management, and presenting a chronology of ML in network management from the 1950s up to 2024/2025, when it reaches generative AI.

The second part deals specifically with network configuration, covering elements such as performance, security, configuration failures, and the definition of algorithms for each of these areas. Algorithms and code are presented, along with their computational complexity and use cases.

The third and final part presents, in greater depth, an example application of generative AI for network management. It introduces a problem that can be addressed using generative AI, aiming to provide a current, up-to-date case.

In addition to code examples, each chapter presents practical exercises and their solutions, as well as slides that professors can use and customize as they wish.

The book connects with several research efforts developed within ICoNIoT.

About the researcher

Oscar Mauricio Caicedo is a professor at the Faculty of Electronic Engineering and Telecommunications at Universidad del Cauca, Colombia, where he is a member of the Telematics Engineering Group. He earned his PhD at UFRGS under the supervision of researcher Lisandro Granville. When he returned to Colombia in 2015, he began working in the field of network management using machine learning algorithms with Professor Nelson Fonseca—well before the boom in generative AI that has been observed. 

In 2018, Prof. Caicedo was already publishing the first results of his research in ML for network management, and he continues to do so in 2026. His book draws on this entire experience of more than ten years of research, using reinforcement learning, deep reinforcement learning, federated learning, explainable artificial intelligence, and large/small language models, among others for solving problems in faults, configuration, performance, and security in communication networks.

To learn more, visit the book’s page on the Springer Nature website

 

At the beginning of the year, the OpenRAN@Brasil Programme announced the results of the Call for 5G Open RAN Applications and Hosts, an initiative aimed at promoting innovative solutions based on open radio access networks. The selected proposals aim to boost the development of 5G Open RAN applications in areas of strategic importance to the country.

Of the six proposals selected, three are coordinated by researchers from INCT ICoNIoT. They are: GreenRAN: Sustainable Open RAN for Smart Agriculture and Campuses/Coordinator: Eduardo Cerqueira (UFPA); OpenHealth5G: Open 5G Networking for Digital Health/Coordinator: Juliano Wickboldt (UFRGS); and Sentinel-5GO: Collaborative Security and Environmental Monitoring in 5G Open RAN Networks/Coordinator: Augusto Neto (UFRN).

A fourth proposal, entitled CampusRAN: Intelligent Campus with Open Radio Access Network and coordinated by Dianne Scherly (UFF), involves the participation of ICoNIoT researchers Igor Moraes and Diogo Mattos, both from UFF.

Learn more about the proposals:

GreenRAN: Sustainable Open RAN for Smart Agriculture and Campuses

The proposal “GreenRAN: Sustainable Open RAN for Smart Agriculture and Campuses‘ offers the ORAN-Green Solution, which will be systematically and extensively evaluated based on two applications developed in different use cases, in the verticals ’Agro 4.0 and Rural Connectivity‘ (primary) and ’Smart Cities and Campuses” (secondary). These applications will use 5G connectivity provided by the Green-5G Network, hosted on the UFPA campus in Belém.

OpenHealth5G

OpenHealth5G, a partnership between UFRGS, PUC-RS, Federal University of Health Sciences, and UNISINOS, proposes the development and testing of innovative applications in Digital and Connected Health, exploring the potential of open 5G and Open RAN network technologies. Prioritised use cases include emergency telehealth, which will enable remote care in critical situations with support for connected medical devices and sensors, and immersive medical education, which will use augmented and virtual reality resources to train students and healthcare professionals.

CampusRAN

The objective of this project is to implement a smart campus at UFF for the development and testing of 5G Open RAN applications. Two applications will be developed, Monitoring Dashboard and Gamified Connectivity Application, which involve the following services: indoor location of UEs, video monitoring for campus community safety and environmental control, indoor and outdoor space sensing, and Internet access for the academic community.

National expansion of the testbed

In addition to selecting applications, the call also approved new host institutions that will receive the OpenRAN@Brasil Programme islands. The increase in the number of hosts strengthens the national expansion of the testbed, allowing for greater infrastructure capillarity and expanding access for researchers in the North, Northeast, and South regions of the country. In the Northeast region, UFRN will be awarded the island, also under the coordination of Professor Augusto Neto.

Research infrastructure

The projects approved by OpenRAN@Brasil are aligned with UNICAMP’s goal of developing and implementing research infrastructure, as well as the associated control tools to enable experimentation in Europe and Brazil, in order to promote experimental research at the point of convergence between optical and wireless networks. Also in 2016, UNICAMP developed the FUTEBOL project with UFRGS, UFMG, UFES, UFCE, and European universities. The project’s objective was to develop and implement research infrastructure, as well as the associated control tools to enable experimentation in Europe and Brazil, in order to promote experimental research at the point of convergence between optical and wireless networks.

Open innovation

One of the great potentials of OpenRAN@Brasil is to implement a testbed environment that opens up many possibilities for research, as it is open. Such testbed environments usually have a cost that can deter researchers with fewer resources. The approved applications will explore the potential of the Open RAN ecosystem, which, in addition to experimentation, will enable technological validation and the generation of applied knowledge in different contexts of use, aligned with national demands for innovation and digital transformation.

The programme is run by RNP, CPQD, Inatel and the Eldorado Institute. It encourages collaboration between industry and academia and seeks to promote collaborative development models, without limiting itself to open source. It seeks to meet the demands of service providers and users of private networks, promote the innovation ecosystem through experimentation and demonstration space, promote application scenarios (public and private networks) and support human resources training.

The 44th edition of the Brazilian Symposium on Computer Networks and Distributed Systems (SBRC) will be held from 25 to 29 May 2026 in person.

The mini-courses approved for the symposium were recently announced: there are five mini-courses, four of which include researchers from ICoNIoT.

Check out the courses and participating researchers:

Mini-course 1
Attacks on Federated Learning: Practical Impacts and Mitigation Strategies, Helio N. Cunha Neto (UERJ), Carlos Henrique Nunes (UERJ), Raphael Ortolan (UERJ), Ian Vilar Bastos (UERJ), Evandro Luiz Cardoso Macedo (UERJ), Rafaela Correia Brum (UERJ), Alexandre Sztajnberg (UERJ), Diogo Menezes Ferrazani Mattos (UFF).

Mini-course 2
Federated Vehicle Learning: From Theory to Practice, Lucas Airam C. de Souza (UFRJ), Guilherme Thomaz (UFRJ), Mateus da Silva Gilbert (UFRJ), Vinicius Avena (UFRJ), Felipe Gomes Táparo (UFRJ), João Sobrinho (UFRJ), Fernando Silva (UFRJ), Nadjib Achir (INRIA Saclay), Miguel Elias Mitre Campista (UFRJ), Luis Henrique Maciel Kosmalski Costa (UFRJ).

Mini-course 3
Intelligent Agents for Computer Network Configuration: From Theory to Practice with LLM, SLM, RAG and Agentic AI, William Lima Reiznautt (UNICAMP), Eduardo Coelho Cerqueira (UFPA), Diogo Maciel Cunha (UNICAMP), Leandro Villas (UNICAMP), Antonio Alfredo Ferreira Loureiro (UFMG), Denis Rosário (UFPA), Allan M. de Sousa (UNICAMP), Nelson Fonseca (UNICAMP).

Mini-course 5
Building Digital Twin Systems: A Middleware-based Approach, André Gustavo Almeida (IFRN), Lucas Pereira (UFRN), Thais Vasconcelos Batista (UFRN), Everton Cavalcante (UFRN), Flavia Delicato (UFF), Rebeca Mota (UFF).

He will present the lecture ‘Generative AI-empowered 3D human pose tracking’

Abstract: In recent years, 3D human activity recognition and tracking has become an important topic in human-computer interaction. To preserve the privacy of users, there is considerable interest in techniques without using a video camera. In this talk, we first present RFID-Pose, a vision-assisted 3D human pose estimation system based on deep learning (DL), as well as its variations with enhanced generalizability to unseen test subjects and environments. The performance of DL models depends on the availability of sufficient high-quality radio frequency (RF) data, which is more difficult and expensive to collect than other types of data. To overcome this obstacle, in the second part of this talk, we present generative AI approaches to generate labeled synthetic RF data for multiple wireless sensing platforms, such as WiFi, RFID, and mmWave radar, including a conditional Recurrent Generative Adversarial Network (R-GAN) approach and diffusion/latent diffusion based approaches. Finally, we propose a novel framework that leverages latent diffusion transformers to synthesize high quality RF data. This synthetic data augments limited datasets, enabling the training of a subject-adaptive transformer-based kinematics predictor to estimate 3D poses with temporal smoothness from RFID data. Additionally, we introduce a latent diffusion transformer with cross-attention conditioning to accurately infer missing joints in skeletal poses, completing full 25-joint configurations from partial (i.e., 12-joint) inputs. This is the first method to detect 20+ distinct skeletal joints using Generative-AI for RF sensing-based continuous 3D human pose estimation (HPE).

Bio: Shiwen Mao is a Professor and Earle C. Williams Eminent Scholar and Director of the Wireless Engineering Research and Education Center at Auburn University. Dr. Mao’s research interest includes wireless networks, multimedia communications, RF sensing and IoT, smart health, and smart grid. He is the editor-in-chief of IEEE Transactions on Cognitive Communications and Networking, a member-at-large on the Board of Governors and the Technical Committee Board Director of IEEE Communications Society, and Vice President of Technical Activities of IEEE Council on Radio Frequency Identification (CRFID). He is a co-recipient of several technical and service awards from the IEEE. He is a Fellow of the IEEE.

He will present the lecture ‘Generative AI-empowered 3D human pose tracking’

Abstract: In recent years, 3D human activity recognition and tracking has
become an important topic in human-computer interaction. To preserve the privacy of users, there is considerable interest in techniques without
using a video camera. In this talk, we first present RFID-Pose, a
vision-assisted 3D human pose estimation system based on deep learning
(DL), as well as its variations with enhanced generalizability to unseen
test subjects and environments. The performance of DL models depends on
the availability of sufficient high-quality radio frequency (RF) data,
which is more difficult and expensive to collect than other types of data.
To overcome this obstacle, in the second part of this talk, we present
generative AI approaches to generate labeled synthetic RF data for
multiple wireless sensing platforms, such as WiFi, RFID, and mmWave radar, including a conditional Recurrent Generative Adversarial Network (R-GAN) approach and diffusion/latent diffusion based approaches. Finally, we propose a novel framework that leverages latent diffusion transformers to synthesize high quality RF data. This synthetic data augments limited
datasets, enabling the training of a subject-adaptive transformer-based
kinematics predictor to estimate 3D poses with temporal smoothness from
RFID data. Additionally, we introduce a latent diffusion transformer with
cross-attention conditioning to accurately infer missing joints in
skeletal poses, completing full 25-joint configurations from partial
(i.e., 12-joint) inputs. This is the first method to detect 20+ distinct
skeletal joints using Generative-AI for RF sensing-based continuous 3D
human pose estimation (HPE).

Bio: Shiwen Mao is a Professor and Earle C. Williams Eminent Scholar and
Director of the Wireless Engineering Research and Education Center at
Auburn University. Dr. Mao’s research interest includes wireless networks,
multimedia communications, RF sensing and IoT, smart health, and smart
grid. He is the editor-in-chief of IEEE Transactions on Cognitive
Communications and Networking, a member-at-large on the Board of Governors and the Technical Committee Board Director of IEEE Communications Society, and Vice President of Technical Activities of IEEE Council on Radio Frequency Identification (CRFID). He is a co-recipient of several technical and service awards from the IEEE. He is a Fellow of the IEEE.

Network configuration is essential to ensure the reliable and secure operation of computer systems. This process ranges from hardware setup—such as routers, switches, and firewalls—to the design of protocols, IP addressing schemes, and the definition of routing and security policies. The goal is to ensure that the network behaves as expected.

Historically, network configuration has been performed manually, requiring specialized expertise and the development of domain-specific languages. However, as networks grow in size and complexity, manual configuration becomes increasingly inefficient and error-prone.

To address the challenge of complexity, network configuration has become a central pillar in automation frameworks. The most recent major proposal for network automation, as Nelson Fonseca explains, is the Zero-touch Network & Service Management (ZSM) paradigm. It encompasses a set of advanced capabilities, including self-healing, self-monitoring, self-optimization, and, crucially, self-configuration. These automation concepts are part of a broader field that includes autonomic networks and cognitive networks, which have been discussed for over 20 years and are closely related to zero-touch networking.

Within the network management framework, the path toward self-configuration involves specifying the network through intentions (intent-driven networks). These intentions are expressed in natural language. As a result, a network administrator—or even a non-expert—can simply state what they want to be implemented in the network.

This relates to the concept of programmable networks, where high-level APIs allow users to program their networks by specifying how they should behave to support applications. The major challenge in this context is translating these natural language intentions into syntactically and semantically valid network configuration specifications.

Intent-driven self-configuration based on Large Language Models (LLMs) has demonstrated significant potential. However, the use of traditional LLMs presents barriers such as high computational cost and intensive resource consumption. LLMs are typically cloud-based, requiring substantial infrastructure and energy usage.

To overcome these limitations, a lightweight approach based on a Small Language Model (SLM) was presented. Unlike LLMs, SLMs can be deployed locally (on-premise), with lower energy consumption.

This innovative approach employs a fine-tuned SLM built on an agent-based architecture. Fine-tuning is essential to provide the SLM with domain-specific knowledge, enabling it to generate appropriate network configurations.

By leveraging parameter-efficient techniques, this framework enables the rapid translation of configuration requests—expressed in natural language—into valid network configurations. The ability to operate entirely on-premise ensures efficiency, accuracy, and privacy.

This strategy points to a safe and practical path toward automated, intent-driven self-configuration in next-generation systems.

Discover the work of researcher Marcelo Fernandes from the Federal University of Rio Grande do Norte (UFRN)

The intersection of Artificial Intelligence (AI), the Internet of Things (IoT), and Edge Computing is reshaping industry and healthcare. The work of researcher Marcelo Fernandes, from the Federal University of Rio Grande do Norte (UFRN), who has been working with AI for 25 years, explores this integration. He is currently involved in three thematic areas within INCT ICoNIoT: Industrial IoT, Edge Computing, and Healthcare.

Marcelo Fernandes

Marcelo explains that one of the major obstacles to deploying machine learning (ML) on IoT devices is model size. Although ML models for IoT are not language models and are smaller than Large Language Models (LLMs), they can still be too large to fit on resource-constrained devices. The proposed solution involves compressed models. In collaboration with Professor H. T. Kung from Harvard University, Marcelo Fernandes developed two innovative aware compression techniques. The compression is considered “aware” because the model not only learns to perform its task but also learns to compress itself. This occurs during training, where the model is forced to use fewer bits and to self-prune by removing nodes based on their importance. Other techniques reduce model size after training, but this often leads to a drop in accuracy.

The two techniques proposed by the researchers are:

• Iterative Aware Compression Based on Quantization Followed by Pruning
• Iterative Aware Compression Based on Pruning Followed by Quantization

The researchers’ first publication on this topic appeared in 2021 (doi: 10.1109/IJCNN52387.2021.9534430). Since then, Fernandes has been testing these techniques with a focus on various applications. They are now being applied to IoT devices for the first time—this direction began when Marcelo Fernandes started working with ICoNIoT. He is currently supervising a PhD student, Mateus Golbarg, and a Master’s student, Vitor Fidelis Freitas, at UFRN, who are also working within this

The initiative aims to encourage bicycle use in São Paulo by rewarding active users

Bike SP Project arose from the need to make urban mobility more efficient, inclusive, and environmentally sustainable. Its main idea is to encourage more people to use bicycles as a means of transportation in São Paulo, choosing them for daily commutes between home and work, school, university, and other routine destinations. To strengthen this incentive, the project provides cyclists with credits that can be used in public transportation (buses, trains, and the metro).

Researcher Fabio Kon, a member of ICoNIoT from IME-USP, is part of the team coordinating the project. He explained in detail how the initiative works and the technologies involved.

The core idea is to use an application (initially developed for Android) to track users’ bicycle trips and understand their behavior, generating insights that can inform the development of public policies. The project is not limited to producing data for a specific policy area; rather, it aims to serve as a general framework demonstrating how scientific research can support public policy for improving urban mobility in large cities.

How does it work?

During each trip, the app collects the user’s location every 30 seconds. Data from the phone’s accelerometer are used to verify whether the trip is actually being made by bicycle, which is determined through AI algorithms. In addition, GPS coordinates are essential: each segment of the trip has its average speed analyzed to confirm that the user is indeed cycling.

It is crucial for the system to verify that trips are genuinely made by bicycle, as the collected data determine the user’s compensation. Each week, a reward value per kilometer is randomly defined. At the end of the week, the system calculates the total distance traveled and multiplies it by the reward value to determine the amount each user receives.

Not all bicycle trips qualify. The project focuses on collecting data from essential daily trips, such as commuting between home and work, school, university, or public transport terminals. Users register these key locations, which are converted into latitude and longitude coordinates, and the system calculates the distances between them.

One of the project’s main goals is to understand how much financial incentive is needed to effectively encourage people to adopt cycling for essential daily travel. The reward values will be proposed based on these analyses.

Generated Data

The data collected by the Bike SP system will form a large dataset, which will be analyzed by a multidisciplinary team of statisticians, economists, and computer scientists.

The application is capable of collecting detailed data for each city block traveled by users. This enables the system to generate unprecedented insights into urban mobility in São Paulo. The resulting dataset will be unique and directly useful for city planning, particularly for expanding cycling infrastructure. Currently, São Paulo has around 740 km of bike lanes and paths, and the goal is to triple this network within ten years.

By revealing where cyclists travel and the challenges they face, the dataset can support decisions about where to expand bike lanes. Today, many cycling paths in São Paulo function like isolated “islands.” Because they are disconnected, cyclists are often forced to leave the bike lane and rejoin it later, which increases their exposure to traffic and reduces safety. Due to these safety concerns, women—who tend to be more cautious in traffic—are significantly underrepresented among cyclists in the city.

Technologies Involved

The Bike SP system communicates with the SPTrans system, informing the municipal system of the amount to be paid to each cyclist. In addition, the system integrates with several other applications.

For example, the TomTom API calculates distances between registered locations, ensuring routes are suitable for bicycles. Crashlytics is used to report errors in the app during usage, allowing system managers to address issues. Firebase Cloud Messaging (FCM) enables notifications to be sent simultaneously to all users. AppCheck helps prevent fraudulent trip registrations by verifying, through cryptographic methods, that the app in use is authentic, checking its digital signature.

Artificial Intelligence

Artificial intelligence is used in the project to determine whether a journey was actually made by bike: machine learning is applied to analyse the mobile phone’s accelerometer. In addition, an analytics dashboard is currently being developed to help the project team make sense of the 30,000 journeys. It will feature a dashboard for the organised visualisation of this data. In this analysis system, clustering techniques will be applied to detect patterns and group users into categories – for example, those most affected by the incentive will be identified, making it possible to understand for whom (which user profile) specifically the public policy will be developed.