In this project, Multiview AI algorithms will be explored and implemented that have the ability to generate levels of uncertainty in their predictions in such a way that it is possible to quantify how reliable a given prediction is. This way, if a prediction is known to be unreliable, an expert can be notified to make the final decision. Multiview AI algorithms with reliability quantification will be tested in different scenarios relevant to ISP, IXP, and IoT systems, including anomaly detection and device attacks. Specifically, we will seek to combine specific protocol characteristics, statistics based on flows, contents, and packet-level behaviors, with the objective of building more efficient and discriminative multiview representations for supervised learning tasks.
The RedLLM-Integrator project proposes a solution to strengthen the capabilities of Security Operations Centers (SOCs) in Latin America through the progressive and ethical integration of explanatory artificial intelligence tools into their existing processes. Instead of training a new model from scratch – which implies high costs, technological dependence and advanced knowledge in AI infrastructure – the proposal focuses on leveraging already established architectures such as Retrieval-Augmented Generation (RAG) and pre-trained Large Language Models (LLMs) to support offensive simulation, identification of vulnerabilities and the generation of understandable recommendations in digital security environments.
Several research and technological development projects are based on Artificial Intelligence (AI) techniques, statistical models, data mining, among others. These approaches require data to support the implementation of technological solutions. In this context, this project aims to develop a platform for making data available for research and technological development. Additionally, this solution will be integrated with a cloud environment, allowing greater capacity and availability of computational resources (storage and processing). Therefore, the project will have the following activities: (I) Systematic review of privacy laws and implementation of anonymization techniques; (II) Development of an API for data reception; (IV) Perform data anonymization; (V) Make anonymized data available on an open platform; and, (VI) Implement the solution in a cloud environment.
The “Internet Laboratory in Cartagena Rural” is a pilot initiative of Aditum, ISA Impact, the Impact innovation Program of ISA Intercambio Eléctrica S.A. E.S.P. and Airband (Microsoft’s CSR program) with the purpose of closing the digital divide in Latin America, where more than 80 million people in rural areas lack access to fixed internet, drastically limiting their human development and in particular the possibilities of social mobility.
The “Internet Laboratory in Cartagena Rural” is our pilot in the community of Arroyo Grande, Cartagena de Indias, where we have managed to successfully connect more than 300 homes and various institutions, including 3 educational institutions, 4 health posts and nearly 30 businesses and enterprises. This deployment has allowed us to validate the technical and economic viability of our hypotheses, as well as their capacity to generate a tangible impact on the quality of life, education, economic opportunities and overcoming poverty in the benefited communities. Thanks to this, the internet laboratory in Cartagena Rural is positioned as a catalyst for sustainable and scalable digital inclusion in the region.
This project proposes to evaluate the feasibility, performance impact and best practices for the integration of post-quantum algorithms in the DNSSEC and TLS protocols, two fundamental pillars of Internet security. DNSSEC will be the main focus as it heavily employs public key cryptography for digitally signing DNS responses, making it an excellent environment to measure the practical effects of large-scale PQC adoption. The project will also investigate the use of post-quantum algorithms in the TLS handshake, which uses public key cryptography for key exchange and server authentication. The project methodology will combine computational simulations and the assembly of physical network prototypes, including Internet of Things (IoT) devices, to evaluate the impact of PQC in resource-constrained scenarios. Metrics such as latency, throughput, CPU, and memory consumption will be monitored to quantify the effects of adopting the new algorithms.
This project seeks to identify, map and analyze active malicious infrastructure in Latin America, with a focus on command and control (C2) servers linked to malware and cyber threats. Through a combination of automated techniques and manual analysis using tools such as Shodan and Censys, indicators of compromise (IOCs), fingerprints and network data will be collected to allow these servers to be located in five key countries in the region. The objective is to obtain a technical vision of the Latin American threat landscape, identify trends by country, ASN or ISP, and better understand the attack surface in our networks. The project will generate a replicable methodology, a technical report of findings and an open workshop for threat laboratories and technical communities in the region, promoting local research and monitoring capabilities. All work will be developed by the ZoqueLabs team, who have experience in digital threats, OSINT and infrastructure analysis.
To improve the security of software-defined networks, one of the most widely used approaches has been the development of intrusion detection systems (IDS) based on machine learning (ML) and deep learning (DL) models that are deployed at the level of control. However, given the way this approach operates, which is reactive and depends on sending information from data plane devices to the control plane, there are limitations in the scalability of the solution.
Despite the advantages offered by data plane programmability for anomaly detection using ML/DL models, the implementation of this approach is not trivial. Programmable switches face challenges due to their memory and processing and storage capacity constraints. For this reason, models implemented in the data plane must meet the condition of having low computational resource consumption, while maintaining high accuracy in anomaly detection. As an approach to address the challenges represented by these limitations, the use of so-called Tiny Machine Learning – TinyML has recently been proposed. TinyML is a paradigm that facilitates the use of ML on devices with limited processing and low memory capacity.
This project proposes the implementation of compressed ML/DL techniques developed using TinyML for the detection of DoS and DDoS attacks in programmable switches in a software-defined network. The objective is to validate the operation of these ML/DL techniques in the data plane, comparing metrics such as the accuracy in the detection of anomalies, and the consumption of resources in the switches.
Distributed denial of service (DDoS) is a frequent threat to computer networks due to its disruption of the services they offer. This disruption results in network instability and/or inoperability. There are different types of DDoS attacks, each with a different mode of operation, so that their detection has become a difficult task for network monitoring and control systems.
The joint stacking of Machine Learning (ML) models consists of establishing a two-layer architecture, where layer 0 (base models) consists of two or more different learning algorithms that are trained with the same data set, and layer 1 (metamodel), which is trained in the best way from the predictions of the base models to establish the final prediction.
This work is based on the exploration and selection of a data set that represents DDoS attack events and carrying out its treatment in a pre-processing phase, resulting in the training, validation and test data sets. Subsequently, in layer 0, a set of ML models will be instantiated to be trained. Once trained, predictions are made on the validation set. Finally, the base models make predictions on the test data set. These predictions feed the metamodel to make the final predictions and, consequently, obtain the metrics of the joint stacked model, such as accuracy, precision, recall, F1 score, confusion matrix, and ROC-AUC, among others. Throughout the training of the models, different configurations will be tested and the hyper-parameters that present a better result based on the proposed metrics will be chosen.
InteliGente is an innovative initiative dedicated to integrating AI into education, fostering equity, and personalizing learning experiences for diverse student needs. By empowering students to develop AI solutions for social good, we are advancing education, democratizing AI, and addressing disparities in access to technology, particularly in regions such as the Global South. The project’s main innovation lies in the integration of active learning methodologies with a focus on AI.
Furthermore, the project applies a problem-based learning approach, where students are challenged to develop AI solutions for real problems, promoting hands-on learning and applying theoretical knowledge. This method improves students’ understanding of AI and develops crucial skills such as critical thinking, problem-solving, and teamwork. Key competencies developed in the project include argumentation, communication, knowledge, empathy, planning, organization, teamwork, responsibility, citizenship, and scientific, critical, and creative thinking. Additionally, InteliGente aims to encourage participants to develop solutions that align with the Sustainable Development Goals (SDGs), such as (SDG 3) Good health and well-being; (SDG 4) Quality education; (SDG 5) Gender equality; (SDG 10) Reduced inequalities; and (SDG 13) Climate action.
PSYDEH works with rural and indigenous women in the Sierra Otomí-Tepehua-Nahua region of Hidalgo, where they face severe socio economic challenges. PSYDEH currently implements two interconnected field programs. The Sierra Madre Network is a cooperative incubation program and a community leadership school. Tech for All is a digital inclusion program that addresses socioeconomic disparities in access to electricity, technology, and online tools in the isolated communities where we work.
In this context, PSYDEH proposes a project that builds on our core programming that links women’s economic empowerment with digital inclusion to increase the capacity of female partners to use Internet tools to develop, market, and sell their textiles and design a sustainable tourism route that brings economic and social benefits to the region. Key project activities include building two additional digital resource centers in the region, conducting basic computer skills workshops and a two-month intensive digital entrepreneurship course, supporting cooperatives in developing branding and marketing strategies, and developing a regional online sales platform where women can sell their products. They also involve promoting a pilot sustainable tourism route online and expanding that route through the use of online payment systems, digital design software and social media marketing.