Program
October 22, 2025
Room: Nery, 1st floor
| Introductions + Opening: Chair: J. Nelson Amaral — University of Alberta | ||
| 8:00-8:10 | Welcoming and Opening Remarks | |
| 8:10-8:30 | Who's in the Room?: Getting-Connected Activity | |
| J. Nelson Amaral — University of Alberta | ||
| Session 1: Computers and Software Systems Chair: Alfredo Goldman — University of São Paulo | ||
| 08:30 - 09:00 | Powering Intelligent Development in the LLM Era: Innovation Across Languages, Compilers, and IDEs | |
| Yaoqing Gao — Huawei Technologies Canada | ||
| 09:00 - 09:30 | There and back again - Learnings from software that guides hardware design that models the software | |
| Rodolfo Jardim de Azevedo — University of Campinas | ||
| 09:30 - 10:00 | Navigating Software Stack Complexity in the Age of AI Acceleration: From Retargeting to Continuous Practices | |
| Sandro Rigo — University of Campinas | 10:00 - 10:30 | Working Break |
| Session 2: Compilers and Artificial Intelligence Chair: Esteban Mocskos — University of Buenos Aires | ||
| 10:30 - 11:00 | Empowering Opportunities with AI: Lessons Learned | |
| Anderson Rocha — University of Campinas | ||
| 11:00 - 11:30 | Towards an End-to-End Tensor Compiler | |
| Fernando Quintão Pereira — Federal University of Minas Gerais | ||
| 11:30 - 12:30 | Panel: The Next Ten Years of Computing Systems: The Role of Research & Development | |
| J. Nelson Amaral — University of Alberta (chair) | ||
| Yaoqing Gao — Huawei Technologies Canada | ||
| Edson Borin — University of Campinas | ||
| Rodolfo Jardim de Azevedo — University of Campinas | ||
| Fernando Quintão Pereira — Federal University of Minas Gerais | ||
| 12:30 - 14:00 | Lunch | Session 3: Artificial Intelligence and Applications I Chair: Marco Eduardo Valle — University of Campinas |
| 14:00 - 14:30 | The Era of DC SuperPods: Tech Innovation and Challenges | |
| Yu Baifeng — Huawei Technologies Canada | ||
| 14:30 - 15:00 | Real-Time Adaptive Multi-Objective Learning with Optimality Guarantees | |
| Bruno da Silva — University of Massachusetts | ||
| 15:00 - 15:30 | Towards Efficient Vision, Language and Multimodal Models | |
| Artur Jordão — University of São Paulo | ||
| 15:30 - 16:00 | Working Break | Session 4: Artificial Intelligence and Applications II Chair: Jun Luo — Huawei Technologies Canada | 16:00 - 16:30 | LatinX in AI: Bridging Academic Research and Industry Innovation for the Latin American Community |
| Abraham Ramos — LXAI | ||
| 16:30 - 17:30 | Panel: Reasoning and Decision Making for Next-Generation AI Systems | |
| Esther Colombini — University of Campinas (chair) | ||
| Jun Luo — Huawei Technologies Canada | ||
| Vicky Wang — Huawei Technologies Canada | ||
| Marco Eduardo Valle — University of Campinas | ||
| Bruno Castro da Silva — University of Massachusetts | ||
| 17:30 - 19:30 | Personal Break | |
| 19:30 | Bus departs to dinner | |
| 20:00 - 22:00 | Dinner at Fogo de Chão Jardins, Rua Augusta, 2077 - Cerqueira César Jardins, São Paulo |
October 23, 2025
Room: Nery, 1st floor
| Session 5: Database and Information Systems I Chair: Mario Nascimento — Northeastern University | ||
| 08:30 - 09:00 | GaussDB: A Cutting-Edge Cloud-Native for the AI Era | |
| Tianqing Wang — Huawei Technologies Canada | ||
| 09:00 - 09:30 | Large Language Models in Data Engineering: Opportunities and Challenges | |
| Altigran S. Silva — Federal University of Amazonas | ||
| 09:30 - 10:00 | More with Less - Sustainable AI Approaches for Natural Language Processing leveraging Data Engineering | |
| Marco Gonçalves — Federal University of Minas Gerais | ||
| 10:00 - 10:30 | Working Break | |
| Session 6: Database and Information Systems II Chair: Essam Mansour — Concordia University | ||
| 10:30 - 11:00 | Data and Model Management at Gypscie | |
| Fábio Porto — National Laboratory of Scientific Computing | ||
| 11:00 - 11:30 | Effective and Responsible AI for Health: Diagnosing Cardiac Conditions from ECGs | |
| Wagner Meira — Federal University of Minas Gerais | ||
| 11:30 - 12:30 | Panel: Data Management and AI | |
| Mario Nascimento — Northeastern University (chair) | ||
| Essam Mansour — Concordia University | ||
| Anderson Chaves Carniel — Huawei Technologies Research & Development (UK) Limited | ||
| Fábio Porto — National Laboratory of Scientific Computing | ||
| Marco Gonçalves — Federal University of Minas Gerais | ||
| 12:30 - 13:50 | Lunch | |
| Session 7: Trustworthy Software and User Behavior Chair: Wagner Meira — Federal University of Minas Gerais | ||
| 13:50 - 14:20 | Drivers and Dangers of Online Behavior: From Information Spread in Social Media to User Exposure in Mobility Data | |
| Jussara Almeida — Federal University of Minas Gerais | ||
| 14:20 - 14:50 | Beyond Machine Learning and Foundation Models: A Research Roadmap for Multi-Paradigm AI Engineering | |
| Marcos Kalinowski — PUC Rio de Janeiro | ||
| 14:50 - 15:20 | From Requirements to Code: Assessing the Impact of Foundation Models on Software Development | |
| Eduardo Figueiredo — Federal University of Minas Gerais | ||
| 15:20 - 15:40 | Working Break | |
| Session 8: Software Development, AI, and Energy Chair: Filipe Roseiro Cogo — Huawei Technologies Canada | ||
| 15:40 - 16:00 | Understanding and Improving the Use of Large Language Models for Software Engineering | |
| Paulo Borba — Federal University of Pernambuco | ||
| 16:00 - 16:20 | How AI Can Benefit from and Contribute to Energy Resilience | |
| Xun Gong — Huawei Technologies Canada | ||
| 16:20 - 16:50 | Rethinking Deep Learning: From Data-Follows-Model to Model-Follows-Data Intelligence | |
| Yasser Khalil — Huawei Technologies Canada | ||
| 16:50 - 17:05 | Language Models for Network Configuration | |
| Nelson Fonseca — University of Campinas | ||
| 17:05 - 18:05 | Panel: Shaping the Future of Trustworthy AI in Practice: Challenges and Pathways | |
| Ahmed E. Hassan — Queen's University (Chair) | ||
| Sihao Li — Huawei Technologies Canada | ||
| Eduardo Figueiredo — Federal University of Minas Gerais | ||
| Marcos Kalinowski — PUC Rio de Janeiro | ||
| Alfredo Goldman — University of São Paulo | ||
| 18:05 - 19:30 | Personal Break | |
| 19:30 | Bus departs to dinner | |
| 20:00 - 22:00 | Dinner at BRAZ Tratoria Paulista Alameda Ministro Rocha Azevedo, 72, Bela Vista, São Paulo |
Powering Intelligent Development in the LLM Era: Innovation Across Languages, Compilers, and IDEs Yaoqing Gao — Huawei Canada
Abstract: The era of large language models is redefining application scenarios, reshaping development paradigms, revolutionizing hardware architectures, and reframing value propositions. In response, we are pioneering intelligent innovation through the unified evolution of programming languages, compilers, and integrated development environments (IDEs). Our approach centers on three foundational pillars: (1) Cangjie Language: A New Programming Language for All-Scenario Intelligence; (2) BiSheng Compiler: A Unified High-Performance Intelligent Compiler for All-Scenario on Multi-Architecture; (3) BitFun IDE: A Next-Generation AI-Native IDE for All-Scenario Intelligent Development.
Biography: Dr. Yaoqing Gao is Chief Expert and Scientist of Huawei’s Compilers and Programming Languages Lab. He leads Huawei’s compiler business strategy and technology planning, driving innovation in compiler technology and product R&D for Huawei’s self-developed chips, including Kunpeng, Ascend, and Kirin. His leadership has been instrumental in supporting the sustained success and competitiveness of Huawei’s ICT and device businesses. For his contributions, Dr. Gao has received Huawei’s highest staff honors—the Individual Gold Medal Award and the Team Gold Medal Award—on multiple occasions. Before joining Huawei, Dr. Gao conducted research and teaching on programming models, programming languages and compiler technology, parallel and distributed processing, computer architectures, and artificial intelligence at the National University of Defense Technology, Tsinghua University, the National University of Singapore, the University of Tokyo, and the University of Alberta. He later moved to industry as Chief Architect of IBM’s compiler team, where he led compiler innovation and global product delivery for IBM Power servers, Blue Gene supercomputers, CELL processors, System z mainframes, and other large-scale projects. At IBM, he received multiple Outstanding Technical Achievement Awards and Master Inventor Awards. Dr. Gao has published more than 50 papers in leading international journals and conferences and is the author of over 40 international patents.
There and back again - Learnings from software that guides hardware design that models the software
Rodolfo Jardim de Azevedo — University of Campinas
Abstract: In this talk, we will go through our studies on hardware design for accelerators exploring mostly the results in ISA specialization, external accelerators, going up to processing in memory for specific workloads. We will also discuss some of the challenges and opportunities in this space and how we can improve the hardware-software co-design for better performance and energy efficiency. We will also show some of our recent studies where the opposite path is necessary, moving some hardware features back to software to reduce hardware costs while keeping the performance requirements. We will also discuss the importance of benchmarks and how they can help in the design of hardware and software systems. Specifically, we will look at one of our recent works to create a processor collection with more than 100 open-source implementations for ample evaluation of software and hardware systems. Finally, we will discuss some of the future directions in this field and how we can continue to improve the hardware-software co-design for better performance and energy efficiency.
Biography: Rodolfo Azevedo is Full Professor at the the University of Campinas (UNICAMP). He received his Ph.D. in Computer Science from the University of Campinas (UNICAMP) in 2002 and is a member of the Computer Science graduate program where he advises master and Ph.D. students. He got several prizes in his career including six best papers in conferences, the Zeferino Vaz Academic Award and Teaching Award from Unicamp. He has CNPq Research Fellowship and was honored 8 times in the Computer Science and Computer Engineering graduations. He coordinated the graduate program in Computer Science, was Director of the Institute of Computing and President of the São Paulo Virtual University (UNIVESP). His main research interests include Computer Architecture, Hardware Software Integration, and Processor Specialization
Navigating Software Stack Complexity in the Age of AI Acceleration: From Retargeting to Continuous Practices
Sandro Rigo — University of Campinas
Abstract: As the demand for AI grows across domains—from embedded and edge devices to large-scale HPC and datacenter deployments—the diversity of hardware platforms is rapidly increasing. We have seen new CPUs, GPUs, TPUs, custom ASICs, and the rise of RISC-V as an open and flexible architectural alternative for meeting different performance, power, and cost goals. This growing hardware fragmentation creates significant challenges for building performant and maintainable system-level software, especially in a diverse and fast-evolving field like AI.
In this talk, we explore these challenges through the lens of a real-world experience: retargeting code generation and low-level libraries, such as OpenBLAS, to a custom RISC-V matrix accelerator. We highlight how subtle ISA design decisions ripple through the software stack, requiring careful adaptation, testing, and validation. We then shift focus to the role of robust continuous integration and validation practices in addressing these challenges. Such practices enable rapid prototyping, ensure correctness and compliance, and support performance regression tracking across evolving hardware/software configurations. We conclude by outlining our perspective on how a Continuous Practice framework can unify hardware and software co-development to support the sustained evolution and customization of low-level AI software stacks.
Biography: Sandro Rigo received his Ph.D. in computer science from the University of Campinas (UNICAMP), Brazil (2004), for his work on the ArchC Architecture Description Language. He works as an Associate Professor at the Institute of Computing at UNICAMP. Prof Rigo was the chair of the BSC program in Computer science from 2009 to 2011. He served as the Information Technology and Communication Coordinator for the University of Campinas (2017-2019) and Director of the Computing Centre from 2017 to 2021. He co-received the best paper awards in the Symposium on Computer Architecture and High-Performance Computing (SBAC-PAD 2004), Workshop on Computer Architecture and High-Performance Systems (WSCAD 2012), and the Brazilian Symposium for Information Security (SBSEG 2016). His current research interests are on several aspects of computer systems design, including Parallelization and acceleration of Machine Learning Models in Heterogeneous Distributed Systems, Parallel programming models in HPC clusters, energy efficiency in computer systems, code generation and optimization, and exploring new architectures and accelerators for AI and HPC.
Empowering Opportunities with AI: Lessons Learned
Anderson Rocha — University of Campinas
Abstract: Artificial Intelligence is transforming industries and societies, but its real impact goes far beyond the hype. In this talk, we will explore what truly works—and what doesn’t—in building and deploying AI systems, from data governance and policy challenges to successful real-world applications. Drawing on examples from impactful projects at Recod.ai and in industry, we will highlight promising areas of innovation, discuss pitfalls to avoid, and offer a forward-looking perspective on emerging trends shaping the next wave of AI opportunities.
Biography: Anderson Rocha is an IEEE Fellow, IEEE Distinguished Lecturer, and IEEE Biometrics Distinguished Lecturer. He is a Full Professor of Artificial Intelligence and Digital Forensics at the Institute of Computing, University of Campinas (Unicamp), Brazil. He is the head of the Artificial Intelligence Lab, Recod.ai, at Unicamp and was the Director of the Institute from 2019 to 2023. He is an elected affiliate of the Brazilian Academy of Sciences (ABC) and the Brazilian Academy of Forensic Sciences (ABC). He is a three-term elected member of the IEEE Information Forensics and Security Technical Committee (IFS-TC) and a two-term former chair of that committee. He is a Microsoft Research and Google Research Faculty Fellow. In addition, in 2016 he was awarded the Tan Chin Tuan (TCT) Fellowship and the Asia Pacific Artificial Intelligence Association Fellowship. He has been the principal investigator of several research projects in partnership with public funding agencies in Brazil and abroad, as well as with national and multinational companies, having already filed and licensed several patents. He is a Brazilian CNPq research scholar (PQ1C). He is ranked among the top 2% of research scientists worldwide, according to PLOS One/Stanford and Research.com studies. Finally, he is now a LinkedIn Top Voice in Artificial Intelligence for continuously raising awareness of AI and its potential impacts on society at large.
Towards an End-to-End Tensor Compiler
Fernando Quintão Pereira — Federal University of Minas Gerais
Abstract: A tensor compiler translates high-level representations of neural networks into efficient code that runs on diverse architectures such as CPUs, GPUs, or TPUs. With the growing popularity of artificial intelligence applications, interest in the design and implementation of such compilers has surged. Yet, much remains to be done before we have tools fully capable of exploiting advances in hardware technology and emerging software trends.
This talk will present analyses and optimizations developed at UFMG’s Compilers Lab to make tensor compilers more efficient. In particular, we will discuss a new kernel autotuning algorithm recently incorporated into the Apache TVM compiler and a kernel fusion optimization deployed in the Xtensa Neural Network Compiler from Cadence in 2024. These contributions represent not only practical improvements to widely used tools but also theoretical innovations in a rapidly evolving research field.
Biography: Fernando Pereira received his Ph.D. from the University of California, Los Angeles in 2008. Since November 2009, he has been an Associate Professor in the Department of Computer Science at the Federal University of Minas Gerais, where he leads the Compilers Lab. His research develops principles and techniques that enable programmers to use modern hardware resources more efficiently through static analyses and compiler optimizations. His contributions include the puzzle-based register allocator (Patent US 2009/0083721 A1), the Divergence Analysis currently adopted in Mesa3D, the Droplet Search algorithm used in the Apache TVM compiler, and the MST Kernel Fusion algorithm employed in the Xtensa Neural Network Compiler.
The Era of DC SuperPods: Tech Innovation and Challenges
Baifeng Yu — Huawei Canada
Abstract: As large-scale AI models advance into the era of trillions of parameters, workloads based on agents are increasingly prevalent. Data center architectures are evolving from traditional clusters to supercluster-scale computing structures. These DC superclusters integrate heterogeneous computing domains (CPU, GPU, NPU, DPU) to achieve ultra-low latency, high bandwidth, high reliability, and resilient distributed collaboration, providing a physical foundation for unified MFU optimization in training-inference. This topic addresses the challenges of availability and high performance in memory-centric supercluster systems, including memory semantics, QoS for packet processing of varying sizes, and system resilience capabilities.
Biography: Baifeng Yu, Technical Team Lead at the Toronto Computing Data Applied and Acceleration Lab(DAAL). He is primarily responsible for the research and development (R&D) planning of the computing product line at CARI. His work involves exploring emerging technology directions and planning technical partnerships across areas such as AI and infrastructure compute acceleration, graphics and physical AI, and big data&AI.
Real-Time Adaptive Multi-Objective Learning with Optimality Guarantees
Bruno da Silva — University of Massachusetts
Abstract: In this talk, I will highlight recent work from our group at the University of Massachusetts, focusing on reinforcement learning (RL) and multi-objective decision-making. RL techniques enable AI systems to learn how to make decisions in unknown environments and to solve complex tasks without human intervention. RL has recently been applied in a wide range of high-impact real-world settings. Examples include controlling stratospheric balloon fleets to restore communications in disaster zones, autonomously optimizing cooling in large-scale data centers, training agile robots to inspect factories, flying U-2 spy planes as AI copilots, and even controlling fusion plasma in nuclear reactors.
I will focus on our recent work in multi-objective RL. Here, agents solving a task or optimizing a system must trade off between (possibly conflicting) objectives such as minimizing cost, improving decision accuracy, increasing speed, reducing energy use, and minimizing risks. A key challenge for real-world deployment is enabling RL agents to adapt their strategies when priorities change, without the need to stop and retrain. I will introduce a novel method that identifies a small set of reusable decision-making strategies that can be composed in real time to produce solutions guaranteed to be optimal under new priorities.
Our technique addresses a long-standing open problem in the literature: how to ensure optimality when priorities change in real time, without requiring further training or data collection, while scaling to high-dimensional, real-world problems. I will conclude by outlining promising future directions that combine optimal decision-making, hierarchical control, and machine learning methods with high-confidence guarantees of safety and reliability.
Biography: Bruno C. da Silva is an Assistant Professor in the College of Computer Sciences at the University of Massachusetts. He was previously an Associate Professor at the Federal University of Rio Grande do Sul (UFRGS) in Brazil and a postdoctoral associate in the Aerospace Controls Laboratory at MIT. He received his Ph.D. in Computer Science from the University of Massachusetts in 2014, advised by Andrew Barto, recipient of the 2024 Turing Award and widely regarded as a founding father of reinforcement learning. Bruno was also a visiting researcher at the Laboratory of Computational Neuroscience in Rome, where he developed novel control algorithms for humanoid robots, and at Adobe Research in California, where he worked on large-scale machine learning methods for digital marketing optimization. His research lies at the intersection of machine learning, reinforcement learning, robotics, and the design of ML systems with high-confidence safety guarantees.
Towards Efficient Vision, Language and Multimodal Models
Artur Jordão — University of São Paulo
Abstract: Deep learning models, especially large ones, continue to push the boundaries in solving diverse cognitive tasks. Unfortunately, developing these models demands substantial computational resources and energy consumption. Hence, state-of-the-art models are computationally demanding, and their development draws the attention of the scientific community to the environmental impact of AI (GreenAI). Pruning emerges as an effective mechanism to address the capacity-computational cost dilemma by eliminating structures (neurons or layers) from large models. This talk introduces recent research within this promising, active and exciting field. It delves into pruning techniques as a pillar of GreenAI and a foundation for the next wave of efficient large vision, language, and multimodal models.
Biography: Artur Jordao is an Assistant Professor at the University of São Paulo (USP), in the Department of Computer Engineering and Digital Systems. He earned his Master’s and Ph.D. degrees in Computer Science from the Department of Computer Science at the Federal University of Minas Gerais (UFMG), focusing on machine learning and deep learning. He was a post-doctoral researcher at the the University of Campinas (UNICAMP). His research focuses on improving the efficiency of large-scale models.
LatinX in AI: Bridging Academic Research and Industry Innovation for the Latin American Community
Abraham Ramos — LXAI
Abstract: LatinX in AI (LXAI) was founded to address the representation gap and build a thriving global community for Latinx individuals in Artificial Intelligence. This talk will provide a comprehensive overview of our organization's mission, history , and core programs, including our official workshops at top-tier conferences , our peer-reviewed journal , and our mentorship and supercomputer access initiatives. With a special focus on the workshop's theme, this presentation will delve into LXAI's crucial role as a bridge between academic excellence and industry innovation. We will explore how our platforms create a direct talent pipeline for students, translate academic research into industrial impact, and foster vital cross-sector collaboration. Attendees will leave with a clear understanding of the opportunities available for their students, their institutions, and their own collaborative research.
Biography: Abraham Ramos is an AI-Business Researcher and Finance Workshop Lead for the LatinX in AI Organization. With over four years of experience in finance and accounting, he provides critical financial stewardship for more than 15 international workshops hosted at premier AI conferences across America, Europe, and Asia. As a researcher, Mr. Ramos focuses on the artificial intelligence landscape in Latin America, translating market trends and policy analysis into actionable strategies. He is a published author in economics and financial journals and has presented his research at international conferences. His work is dedicated to analyzing and unlocking the potential of AI to foster growth and innovation throughout Latin America.
GaussDB: A Cutting-Edge Cloud-Native for the AI Era
Tianqing Wang — Huawei Canada
Abstract: As AI continues to reshape industries, the role of databases as the foundation of intelligent applications has never been more critical. In this talk, we introduce GaussDB and GaussVector—Huawei’s unified, cloud-native intelligent database platform designed for the Data + AI era. GaussDB, an enterprise-class distributed relational database, has demonstrated outstanding performance and reliability across key sectors including finance, government, and manufacturing. We will explore its cutting-edge innovations in performance optimization, high availability, and intelligent automation using AI. Additionally, we will discuss the evolution of GaussVector and its role in bridging structured data management with AI workloads. Finally, the talk will highlight the technical roadmap ahead and the key challenges in advancing database technologies for next-generation intelligent systems.
Biography: Tianqing Wang is a Senior Technical Expert in database systems with core R&D contributions to Huawei’s GaussDB. As a key member, he: 1). Architected the GaussDB analytical engine, leading innovations in resource management, WLM, multi-tenancy, and scheduling 2). Spearheaded GaussDB’s autonomous database development from inception to production 3). Drove GaussDB’s AI+DB integration to industry leadership, establishing DB4AI/AI4DB ecosystems 4). Pioneered cloud-native database solutions and now explores Data+AI convergence 5). Author of 10+ top-tier papers and patents, his work has shaped multiple generations of GaussDB’s technical evolution.
Large Language Models in Data Engineering: Opportunities and Challenges
Altigran S. Silva — Federal University of Amazonas
Abstract: Large Language Models (LLMs) are reshaping the way we approach data engineering by enabling new forms of query processing, metadata management, and integration across heterogeneous repositories. In this talk, I will highlight recent advances from our research on unifying polystore and datalake architectures with the support of LLMs, focusing on query translation, ranking, and bias mitigation. I will also discuss open challenges, such as scalability, fairness, and evaluation, and suggest research directions that could foster stronger collaboration between academia and industry.
Biography: Altigran Soares da Silva is a Professor of Computer Science at the Federal University of Amazonas (UFAM), where he leads research in data management, information retrieval, and artificial intelligence. He has extensive experience in creating technology-based startups, with successful exits to companies such as Google, Linx, and Jusbrasil. Beyond his entrepreneurial activities, he has served in leadership roles in graduate education and research policy in Brazil and is currently engaged in national and international collaborations on data engineering and AI. His work has spanned information retrieval and extraction, and the application of machine learning techniques to large-scale data ecosystems.
More with Less - Sustainable AI Approaches for Natural Language Processing leveraging Data Engineering
Marco Gonçalves — Federal University of Minas Gerais
Abstract: This talk critically examines the sustainability of current AI models, especially large language models (LLMs), highlighting the high computational, financial, and environmental costs of the “more data, more hardware, more energy” paradigm. It introduces alternative strategies such as instance selection to reduce resource consumption while preserving effectiveness, and emphasizes the importance of responsible and inclusive AI, with initiatives like INCT TILDIAR driving ethical and sustainable research in Brazil
Biography: Marcos Andre Gonçalves is a full professor of Computer Science at Universidade Federal de Minas Gerais. He has more than 400 published peer-reviewed papers, h-index of 61, and numerous awards including advisor of the best Ph.D. dissertation in Brazil (Premio Capes 2024). He is currently a member of the Advisory Committee for Computer Science of the Brazilian National Research Council (CNPq).
Data and Model Management at Gypscie
Fábio Porto — National Laboratory of Scientific Computing
Abstract: Gypscie is an ML/AI lifecycle artifacts management system. It inherits and expands the approach taken to manage large datasets to the complete set of artifacts produced and used in ML/AI pipelines. In particular, we will discuss aspects of its dataflow scheduling in heterogeneous environments, its knowledge-graph rule-based integration view approach, and the use of SAVIME to query in-memory raw datasets.
Biography: Fabio Porto is a Senior Researcher at the National Laboratory of Scientific Computing, where he is the head of the Data Extreme Lab. He holds an International Chair position at INRIA in France (2024–2028) and was a Visiting Professor Researcher at the National University of Singapore (NUS), Computer Science Department, in 2020. He currently serves as an Area Reviewer for ICDE 2026 and for several journals. His current research interests include designing efficient systems for running complex workloads, such as ML training dataflows, and data integration processes targeting knowledge graph GAV construction.
Effective and Responsible AI for Health: Diagnosing Cardiac Conditions from ECGs
Wagner Meira — Federal University of Minas Gerais
Abstract: AI became a pervasive technology in our lives, and its use in health tasks is often considered one of the most promising and challenging. More recently, the demand for such technology evolved towards the so-called responsible AI, which aims to develop and deploy AI in a safe, transparent, fair, trustworthy, and ethical way. In this talk, we present our work on AI models for health, in particular for diagnosing cardiovascular conditions from electrocardiograms, and how we improved them not only regarding their performance and effectiveness, but also towards a more responsible and applicable solution. We conclude by outlining some trends and directions of research and development in responsible AI.
Biography: Wagner Meira Jr. (PhD, CS, University of Rochester, 1997) is a full professor of Computer Science at Universidade Federal de Minas Gerais, Brazil. He has published more than 300 papers in top venues and is a co-author of the books Data Mining and Analysis - Fundamental Concepts and Algorithms and Data Mining and Machine Learning - Fundamental Concepts and Algorithms, both published by Cambridge University Press in 2014 and 2020, respectively. His research focuses on scalability, efficiency, and characterization of large-scale parallel and distributed systems, from massively parallel to Internet-based platforms, and on data mining and machine learning algorithms, as well as their application to areas such as health, information retrieval, bioinformatics, e-governance, and cybersecurity.
Drivers and Dangers of Online Behavior: From Information Spread in Social Media to User Exposure in Mobility Data
Jussara Almeida — Federal University of Minas Gerais
Abstract: Human behavior is inherently heterogeneous, multifaceted, and dynamic, with aspects that are often subjective and difficult to extract from digital traces. In this talk, I will present a brief overview of our research on modeling user behavior in two domains: information spread in social media and individual mobility. In the social media context, I will present recent results showing how individual and collective behaviors shape the diffusion of information. In the mobility context, I will discuss the privacy risks associated with the growing use of mobility data to power digital services, underscoring the need for new methods to quantify user exposure. I will conclude by highlighting recent contributions from our group toward developing such methods to better inform privacy protection strategies.
Biography: Jussara M Almeida is a Full Professor at the Universidade Federal de Minas Gerais, Brazil, where she currently leads the Social Computing Laboratory (LOCUS) at the Department of Computer Science. She holds a Ph.D. from the University of Wisconsin-Madison (US) and is a former affiliate member of the Brazilian Academy of Sciences. Her main research interests include social computing, user behavior analysis, as well as performance analysis and modeling of large-scale distributed systems.
Beyond Machine Learning and Foundation Models: A Research Roadmap for Multi-Paradigm AI Engineering
Marcos Kalinowski — PUC Rio de Janeiro
Abstract: Rapid advancements in AI paradigms—Machine Learning, Reinforcement Learning, Symbolic AI, Foundation Models, and Multi-Agent Systems—have expanded AI’s capabilities while introducing new software engineering challenges. Each paradigm has inherent limitations: Machine Learning excels at pattern recognition but lacks explicit reasoning; Reinforcement Learning optimizes rewards but is unpredictable; Symbolic AI provides structure but struggles with flexibility; Foundation Models generalize knowledge but raise interpretability concerns; and Multi-Agent Systems enable distributed intelligence but introduce emergent behaviors. Just as the human brain integrates perception, learning, reasoning, and decision-making, AI systems can combine multiple paradigms to achieve adaptable intelligence. While AI Engineering has made progress in building robust Machine Learning systems, our understanding of Intelligence Engineering—the integration of multiple AI paradigms—remains limited. Furthermore, software engineering aspects such as requirements and architecture are still poorly understood across different AI paradigms and their combinations. Without deeper insights and systematic methodologies, we still lack the necessary means to build trustworthy, scalable, and maintainable intelligent systems that effectively integrate these paradigms. This talk presents a research roadmap for AI Engineering, highlighting the urgent need for evidence-based intelligence engineering and the development of systematic specification approaches, architectural patterns, and tactics for multi-paradigm AI integration.
Biography: Marcos Kalinowski is a Professor of Software Engineering at PUC-Rio, Brazil. His research focuses on AI Engineering, Empirical Software Engineering, and Human Aspects of Software Engineering. Before transitioning to academia, he accumulated over a decade of industry experience. He leads the ExACTa PUC-Rio lab, where he supervises PhD research and leads R&D projects with industry partners to develop AI engineering solutions and real-world AI systems. His lab's work has resulted in award-winning AI deployments and multiple US patents. He has authored several books and over 200 research publications, earning 20+ distinguished paper awards, including ACM Distinguished Paper Awards at ICSE (the flagship conference in Software Engineering) and CAIN (the leading conference on AI Engineering). He holds a research productivity distinction from the Brazilian Research Council (CNPq) and an honorary 'Scientist of the State' distinction from Rio de Janeiro's state research agency (FAPERJ). His research is also supported by the Kunumi Institute and the Stone Institute. He serves the community as associate editor of the Empirical Software Engineering journal, IEEE Software, and the Journal of Systems and Software. He also serves as part of the organizing and program committees of several international conferences, with a highlight being serving as General Chair of ICSE 2026, bringing the main scientific Software Engineering conference to Brazil for the first time. He is an active member of ACM, IEEE, ISERN, and the Brazilian Computer Society.
From Requirements to Code: Assessing the Impact of Foundation Models on Software Development
Eduardo Figueiredo — Federal University of Minas Gerais
Abstract: The increasing capabilities of foundation models offer a promising avenue for enhancing human-intensive software development activities from requirements to coding. In this direction, vibe coding and agents have emerged as a transformative paradigm in modern software development by proposing human-in-the-loop interaction through prompt-based conversational workflows to support end-to-end software generation. This talk aims to share our current research on assessing the effectiveness of foundation models for creating requirements and maintaining code quality. We also aim to foster discussions about the future of software development, for instance, with respect to automated code generation. Based on data from empirical studies, we argue that experienced developers are better prepared to critically evaluate the outputs of foundation models in requirements. We also observed in our studies that these models are only effective in easy code quality related assessments. We conclude that software professionals should be careful when using foundation models in software development since their critical failures are related to non-trivial tasks.
Biography: Eduardo Figueiredo is an associate professor and head of the Software Engineering Laboratory (LabSoft) at Federal University of Minas Gerais (UFMG) since 2010. He is also a Research Fellow level B in CNPq, the Brazilian Funding Agency for Scientific and Technological Development. Eduardo received his PhD degree in Software Engineering from Lancaster University (UK) in 2009 and was a visiting researcher at Carnegie Mellon University (CMU) in 2017. Since January 2025, Eduardo has also been a visiting researcher at the University of Ottawa (Canada). Eduardo has published more than one hundred peer-reviewed papers in software engineering journals and conferences. His research interests include artificial intelligence for software engineering, source code analysis, software quality, and empirical software engineering. Website: http://www.dcc.ufmg.br/~figueiredo
Understanding and Improving the Use of Large Language Models for Software Engineering Paulo Borba — Federal University of Pernambuco
Abstract: This talk explores current challenges in understanding and improving the use of Large Language Models (LLMs) for Software Engineering activities. We are particularly interested in identifying factors associated with better LLM performance, so that developers can understand for which tasks LLMs are most useful. In particular, we discuss our results on using LLMs to generate automated tests aimed at detecting semantic code integration conflicts (interference). Finally, we briefly describe how this research is conducted in the context of a long-term industry–academia collaboration.
Biography: Paulo Borba is Professor of Software Engineering at the Federal University of Pernambuco (UFPE), where he leads the Software Productivity Group, and holds a Research Productivity fellowship from the Brazilian Research Council (CNPq). He investigates and develops tools and techniques for improving software development quality and productivity levels, especially by reducing unnecessary effort and frustration in software developers work activities. He pursues a mix of academic excellence with societal relevance. His main research interests are in the following topics and their integration: LLMs for Software Engineering, advanced (semi-structured, structured, semantic) code merging tools, code integration conflicts, continuous integration and deployment, software modularity, software product lines, and refactoring. Paulo has supervised more than 45 academic theses and dissertations, and published more than 200 research papers, including publications at top tier conferences (such as ICSE, ASE, PLDI, and OOPSLA) and journals (such as IEEE TSE, Empirical Software Engineering, Journal of Systems and Software, Information and Software Technology, and Theoretical Computer Science). Paulo is Associate Editor of IEEE Transactions on Software Engineering. Paulo has received a prestigious award from the Brazilian Computer Science (Software Engineering committee) for his career achievements. He has been involved, both as a member and leader, in a number of R&D&I collaborative projects, involving companies and universities, in Brazil, Europe, and the United States. He was Director of the Informatics Center and co-founder of Qualiti Software Processes, a spin-off of the Informatics Center specialized on software development tools and processes. He holds a D.Phil. in Computing from Oxford University.
How AI Can Benefit from and Contribute to Energy Resilience
Xun Gong — Huawei Technologies Canada
Abstract: In the era of artificial intelligence (AI), powering increasing AI loads and data centers in a reliable and resilient fashion becomes essential for the sustainable development of AI technologies. From the “Energy for AI” perspective, renewables are replacing fuel-type generators, and recent extreme weather events have underscored the importance of resilience in grid operations to withstand and recover from disruptions. To secure uninterrupted energy supply, grid resilience must reach certain levels through proper design and coordination of digitalized energy resources. From the “AI for Energy” perspective, AI can help handle uncertainty and nonlinearity in complex event-based power grid operation, and thus can help enhance energy resilience in turn. This presentation will introduce energy resilience and its relationship with AI from these two perspectives.
Biography: Xun Gong is a Staff Researcher at Huawei Technologies Canada. He was a Postdoctoral Researcher at McGill University from 2021 to 2022. His research interests include data-driven modeling, forecasting, control and optimization, and AI applications for microgrid and smart grid systems.
Rethinking Deep Learning: From Data-Follows-Model to Model-Follows-Data Intelligence
Yasser Khalil — Huawei Technologies Canada
Abstract: Contemporary deep learning largely follows a data-follows-model paradigm, where massive datasets are centralized for model training. Motivated by the 6G vision of enabling AI-driven value creation across intelligent networks, this talk introduces a shift toward a model-follows-data paradigm, where models adapt and move across decentralized nodes instead of aggregating data centrally. We present a decentralized learning framework that supports model and data heterogeneity, enabling collaboration without sacrificing data privacy. We also discuss Machine Unlearning (MU) techniques that enable selective forgetting without retraining from scratch, highlighting our recent advances in developing efficient unlearning algorithms. Together, these approaches pave the way toward privacy-preserving, adaptive, and distributed AI systems for the next generation of network intelligence.
Biography: Dr. Yasser Khalil is a Senior Researcher at Huawei’s Noah’s Ark Lab. He researches federated learning, knowledge transfer, machine unlearning, and Agentic AI. He has received multiple Huawei honors for his contributions, including the 2012 Star Award, Q1 CARI President’s Spot Award, Director’s Spot Award, Future Star Individual Award, Optical BU President’s Award, and the Commercial Contribution Team Award. Dr. Khalil has published in leading international journals and conferences and holds nine patents. He earned his Ph.D. in Electrical and Computer Engineering from the University of Ottawa in 2022.
Vicky Wang — Huawei Technologies Canada
Biography: Dr. Vicky Wang is a Technical Planning Researcher and Technical Partnership Manager at the Toronto Computing Data Applied and Acceleration Lab (DAAL). Her work focuses on identifying emerging technology directions and building collaborations across diverse areas, including reinforcement learning (RL) and infrastructure compute acceleration, embodied AI, and technical partnerships. Dr. Wang received her Ph.D. in electronic and Computer Engineering from Purdue University in 2023. She also received an Master's degree in Electrical Engineering from Purdue University in 2018.
Essam Mansour — Concordia University
Biography: Essam Mansour is an Associate Professor at Concordia. He is a research leader and system builder with over 15 years of experience in scalable data infrastructure, generative AI for specialized domains, and multi-agent systems. He is the Head of the CoDS Lab at Concordia University, where he leads collaborations with Google, IBM, and the National Research Council Canada. His work bridges research and engineering through open-source systems for knowledge graphs, automated data science, LLM-powered platforms, and conversational AI. Essam is the General Co-Chair of ICDE 2026.