DISA

Centre for Data Intensive Sciences and Applications

Welcome to our Higher Research Seminar in February

2026-02-02

Where? Onsite: D2272 and via zoom
Registration: Please sign up for the seminar via this link https://forms.gle/7LK5jZfjVwvYAf4L8 by February 18. This is especially important if you plan to attend onsite so we can make sure there is fika for everyone.

Abstracts
Accelerate ML: Overlap of computation and collective communication in multi-GPU systems – Minyu Cui 
The rapid growth of large-scale machine learning (ML) has made distributed training across multiple GPUs a fundamental building block of modern ML systems. As model sizes continue to increase and computational throughput improves, communication overhead has emerged as one of the dominant performance bottlenecks in multi-GPU computing paradigms. Conventional training pipelines in multi-GPU systems perform computation and communication sequentially, which leads to idle compute resources, limited scalability, and inefficient hardware utilization. 

In my research plan, I aim to accelerate multi-GPU ML by overlapping the two dominant operations: computation (such as GEMM) and collective communication. I will explore two complementary and efficient directions. First, my research will explore overlapping computation and communication kernels to hide communication latency. Second, it will further investigate fusing computation and communication into a single GPU kernel to enable efficient fine-grained overlap. These efforts will initially focus on improving operator-level performance and will subsequently be extended to enhance end-to-end training performance. 

I used to love Python… – Morgan Ericsson  
Some 15 years ago, when I did a lot of NLP, I learned Python 2, because it was the language that made the most sense (that was not Perl). I found it to be a beautiful language that made it fast and easy to translate thoughts into code. The rich ecosystem often turned coding into “gluing” things together, and since the things were often written in, e.g., C, it was fast enough. These days, I find it a lot more frustrating. The things were always silos, but a few years ago, I never found it to be a problem. These days, you are stuck with things that might play well together, if the authors were aware and took the time to integrate. If you are lucky, the things will support the platform (cpu, cuda, mps, …) that you are running on, but if not, well, then it’s not so much fun. You are also hitting all kinds of performance issues and bugs in the various things and gaps between them. So, for some work, I don’t like Python very much these days. My talk will rant about the problem but also try to find a way forward, looking at helpful tools for today and ideal solutions for tomorrow… 

Welcome to our PhD-seminar in February

Where? Onsite: D2272 and via zoom
Registration: Please sign up for the seminar via this link https://forms.gle/QB5oiSWMpVofjBHY6 by February 11. This is especially important if you plan to attend onsite so we can make sure there is fika for everyone.

Abstracts
Novelty Detection Using Time-Series Transient Data from the Lead-Acid Forklift Batteries – Zijie Feng  
In industrial applications, monitoring the battery health of electrically powered machinery and detecting abnormal operating conditions is a persistent yet critical challenge. Traditionally, anomaly detection systems are developed reactively: abnormalities are identified only after they occur, often leading to operational disruptions and economic losses. Novelty detection offers an alternative by learning normal behavior and detecting previously unseen abnormalities.  

In this work, we compare a diverse set of data-driven novelty detection methods using simulated time-series transient data derived from real lead-acid forklift battery measurements, aiming to identify suitable solutions for different types of anomalies. 

Potential of Graph Neural Networks for Software Architecture Recovery– Rakshanda Jabeen  
Software architecture recovery (SAR) aims to uncover a system’s modular structure directly from source code, supporting comprehension and maintenance when documentation is missing or outdated. In this work, we investigate the potential of graph neural networks (GNNs) and unsupervised learning for SAR by modeling software systems as heterogeneous, multi-relational graphs. Nodes represent software entities (e.g., files or classes) and typed edges capture structural and functional dependencies such as calls, imports, inheritance, and other code-level relations. To complement dependency structures with meaning, we integrate semantic signals from source code identifiers and related textual artifacts via contextual code embeddings (e.g., Word2Vec), yielding representations that capture both what entities do and how they interact. 

We study heterogeneous GNN encoders that aggregate information across relation types, including heterogeneous graph convolution and heterogeneous attention mechanisms, to analyze the trade-off between fixed normalization and adaptive neighbor weighting in software graphs. On top of these encoders, we explore two unsupervised training paradigms: (i) graph autoencoding, where embeddings are learned by reconstructing observed dependency relations, and (ii) contrastive representation learning inspired by Deep Graph Infomax, which maximizes agreement between embeddings from the original graph and perturbed views. The resulting entity embeddings are clustered to recover candidate architectural modules. Preliminary results across multiple open-source systems indicate that combining semantic cues with structural and functional dependencies produces more meaningful module separation than using structure alone, demonstrating that modern graph representation learning is a promising direction for robust, automated SAR beyond heuristic baselines.