Keynote Speakers
Prof. Peter Rossmanith
RWTH Aachen University, Germany
Prof. Rossmanith is a professor of Theoretical Computer Science at RWTH Aachen University. His research focuses on efficient algorithms, online algorithms with advice, and theoretical foundations of computer science.
Online Algorithms with Predictions
The integration of methods from artificial intelligence, particularly machine learning, has the potential to profoundly impact both science and society. Notably, machine learning can enhance classical algorithms and, perhaps more surprisingly, improve the performance of online algorithms that must make decisions based on current data, without knowledge of future developments.
A prime example is finding the fastest route in a road network, where predicting traffic congestion and roadblocks that may arise during the journey is crucial. Machine learning can provide predictions about future events, although their accuracy is not guaranteed. This raises the question: can we design algorithms that effectively incorporate these predictions while maintaining performance comparable to that of the best algorithms that do not rely on them? In other words, how can we harness the benefits of machine learning predictions while mitigating the risks associated with their potential inaccuracy?
Prof. Mhd. Reza M. I. Pulungan
Universitas Gadjah Mada, Indonesia
Prof. Reza Pulungan is a professor of Computer Science with expertise in stochastic processes and systems modeling. He has contributed significantly to energy-efficient high-performance computing.
A Deep Reinforcement Learning-Based Power Management System for High-Performance Computing Clusters
High energy consumption remains a significant challenge in high-performance computing (HPC) systems, which often feature hundreds or thousands of nodes that draw substantial power even in idle or standby modes. Although powering down unused nodes can improve energy efficiency, choosing the wrong time to do so can degrade the quality of service by delaying job execution. Machine learning, particularly reinforcement learning (RL), has shown promise in determining the optimal times to switch nodes on or off. In this study, we enhance the performance of a deep reinforcement learning (DRL) agent for HPC power management by integrating curriculum learning (CL), a training approach that introduces tasks with gradually increasing difficulty. Using the Batsim-py simulation framework, we compare the proposed CL-based agent to both a baseline DRL method (without CL) and the conventional fixed-time timeout strategy. Experimental results confirm that an easy-to-hard curriculum outperforms other training orders in terms of reducing wasted energy usage. The best agent achieves a 3.73% energy reduction over the baseline DRL method and a 4.66% improvement compared to the best timeout configuration (shutdown every 15 minutes of idle time). Additionally, it reduces the average job waiting time by 9.24% and maintains a higher job-filling rate, indicating more effective resource utilization. Sensitivity tests across various switch-on durations, power levels, and cluster sizes further reveal the agent’s adaptability to changing system parameters without retraining. These findings demonstrate that curriculum learning can significantly enhance DRL-based power management in HPC, striking a balance between energy savings, quality of service, and robustness to diverse configurations.
Prof. Eko Sediyono
Universitas Kristen Satya Wacana, Indonesia
Prof. Eko Sediyono serves as Vice Rector for Research and Innovation. His research bridges AI with development goals and focuses on technology-based education and national advancement.
Research Challenges in Indonesia: Unlocking the Potential of Artificial Intelligence for Sustainable Development
This presentation explores the potential of Artificial Intelligence (AI) to support Indonesia’s progress toward achieving the Sustainable Development Goals (SDGs). It highlights how AI applications—such as smart irrigation, disease detection, adaptive education tools, and climate modeling—can address national development challenges. The presentation also outlines the current AI research landscape in Indonesia, which has shown significant growth, and features a case study on AI-based tuberculosis detection.
Despite the potential, several research challenges remain, including limited data quality, infrastructure gaps, talent shortages, and policy barriers. To overcome these, the presentation proposes strengthening the AI research ecosystem through national data policy reform, cross-sector and international collaboration, and capacity building. It concludes with a call to align AI initiatives with inclusive, ethical, and sustainable development goals.
Prof. Derwin Suhartono
Bina Nusantara University, Indonesia
Prof. Derwin Suhartono is the Dean of the School of Computer Science at BINUS. He specializes in NLP and AI systems aligned with human language and behavior.
Seriously Not Serious: What Sarcasm Tells Us About Language and Intelligence
Sarcasm remains a persistent challenge in natural language understanding due to its implicit meaning, cultural dependency, and contextual ambiguity. In low-resource language settings such as Indonesian, the scarcity of annotated data further complicates this task. Recent work on sarcasm detection has introduced the first benchmark dataset for Indonesian and evaluated multiple modeling strategies, including classical machine learning, transformer-based models, and data augmentation using word embeddings and generative approaches. Results demonstrate that context-aware and culturally aligned methods significantly improve performance. These findings underscore the broader implications of sarcasm as a test case for building more human-aligned, socially aware AI systems.