Differences
This shows you the differences between two versions of the page.
| Both sides previous revisionPrevious revisionNext revision | Previous revision | ||
| aira:start [2026/10/15 00:12] – [Schedule Autumn 2027] mzk | aira:start [2026/11/12 09:06] (current) – [2025-11-12] mzk | ||
|---|---|---|---|
| Line 21: | Line 21: | ||
| ===== Schedule Autumn 2027 ===== | ===== Schedule Autumn 2027 ===== | ||
| + | * **[PHD TRACK] 2026.11.12**: | ||
| + | * Meeting link: | ||
| + | * Recording: | ||
| + | * Presentation slides: | ||
| + | |||
| + | * **[RESEARCH TRACK] 2026.10.29**: | ||
| + | * Meeting link: | ||
| + | * Recording: | ||
| + | * Presentation slides: | ||
| + | |||
| + | * **[PHD TRACK] 2026.10.22**: | ||
| + | * Meeting link: | ||
| + | * Recording: | ||
| + | * Presentation slides: | ||
| + | |||
| * **[PHD TRACK] 2026.10.30**: | * **[PHD TRACK] 2026.10.30**: | ||
| * Meeting link: | * Meeting link: | ||
| Line 27: | Line 42: | ||
| - | | + | * **[INTRO] 2026.10.08**: |
| - | * Meeting link: TDA | + | * Meeting link: TDA |
| - | * Recording: | + | * Recording: |
| - | * Presentation slides: | + | * Presentation slides: |
| ===== Schedule Spring 2026 ===== | ===== Schedule Spring 2026 ===== | ||
| Line 593: | Line 608: | ||
| + | |||
| + | ==== 2026-11-12 ==== | ||
| + | <WRAP column 15%> | ||
| + | {{ : | ||
| + | </ | ||
| + | |||
| + | <WRAP column 75%> | ||
| + | |||
| + | **Speaker**: | ||
| + | |||
| + | **Title**: Automated Ensemble Design in the Wild. | ||
| + | |||
| + | **Abstract**: | ||
| + | Combining machine learning models can improve predictive performance, | ||
| + | This talk presents OKAPI, a multi-objective genetic programming method that jointly searches for ensemble membership and prediction fusion procedures. Ensembles are represented as stateful trees that accumulate and transform model predictions through an evolved sequence of operations. Predictive performance and structural complexity are treated as separate objectives, allowing the search to discover compact ensembles with inspectable fusion procedures. | ||
| + | Experiments spanning medical image classification, | ||
| + | |||
| + | **Biogram**: | ||
| + | Damian Kucharski received his BEng (2021) and MSc (2022) degrees in Computer Science from the Silesian University of Technology, Poland, where he recently submitted his doctoral thesis on automated ensemble design. His research focuses on evolutionary computation and the automated design and optimisation of machine learning ensembles, with applications in biomedical data analysis and Earth observation. Alongside his academic work, he has several years of industrial experience in machine learning and deep learning, including medical imaging, processing large datasets, and model optimisation. His interests also include GPU computing and efficient model training and inference. | ||
| + | </ | ||
| + | <WRAP clear></ | ||
| + | |||
| + | ==== 2026-10-29 ==== | ||
| + | <WRAP column 15%> | ||
| + | {{ : | ||
| + | </ | ||
| + | |||
| + | <WRAP column 75%> | ||
| + | |||
| + | **Speaker**: | ||
| + | |||
| + | **Title**: From raw data to deployment: feature extraction and (rapid) prototyping for production-ready ML models | ||
| + | |||
| + | **Abstract**: | ||
| + | Deep learning and LLMs have led many to assume that hand-crafted features are no longer needed. This seminar argues that careful feature extraction is still one of the most cost-effective ways to move from raw data to a practical model, especially when data are limited, computing resources are small, or decisions must be explained. Using difficult dermatoscopic images of skin lesions as an example, I show how a systematic search over dozens of wavelet bases, tested with most learning paradigms and several datasets, identifies a small group of features that perform best across all of them. Compared with CNNs, they perform at least as well while requiring far fewer computing resources. This approach can become a repeatable workflow for rapid prototyping in other areas. | ||
| + | |||
| + | **Biogram**: | ||
| + | Grzegorz Surówka, PhD, DSc, is an assistant professor at the Faculty of Physics, Astronomy and Applied Computer Science (FAIS) of the Jagiellonian University in Kraków, where he also coordinates the Student Computer Laboratory (9 computer labs, infrastructure and service servers). | ||
| + | He began his career as a physicist, specializing in astrophysics and nuclear physics. In 2001 he received his PhD in physical sciences from the Jagiellonian University. His doctoral research was carried out at the GSI Helmholtz Centre for Heavy Ion Research in Darmstadt, Germany. | ||
| + | He later spent many years as an Ericsson telecommunications trainer, teaching courses worldwide. | ||
| + | He then moved into data science, focusing on feature extraction, wavelets and medical image classification. In 2025 he obtained his habilitation (DSc) in information and communication technology. | ||
| + | He currently works on medical imaging, feature engineering and explainable AI. | ||
| + | Outside academia, he works in cybersecurity, | ||
| + | </ | ||
| + | <WRAP clear></ | ||
| + | |||
| + | ==== 2026-10-22 ==== | ||
| + | <WRAP column 15%> | ||
| + | {{ : | ||
| + | </ | ||
| + | |||
| + | <WRAP column 75%> | ||
| + | |||
| + | **Speaker**: | ||
| + | |||
| + | **Title**: Isolation-Based Anomaly Detection for Evolving Data Streams | ||
| + | |||
| + | **Abstract**: | ||
| + | Anomalies are rare, high-impact events, and detecting them in continuous data streams remains a persistent robustness challenge for deployed AI systems. Streaming settings break the assumptions most detectors rely on: each instance is seen once under bounded memory, the underlying distribution drifts over time, and labels are unavailable or expensive. My research develops four isolation-based detectors designed for these conditions. In the unsupervised domain, one preserves a global view of the cumulative distribution via reservoir sampling and incremental subtree regrowing, while the other tracks the recent local distribution using a sliding window and selectively replaces the weakest ensemble members. Building on these, two semi-supervised extensions use a small labelling budget, with label information guiding scoring, ensemble maintenance, | ||
| + | |||
| + | **Biogram**: | ||
| + | Jia (Justin) Liu is a PhD candidate at the University of Waikato, New Zealand, supervised by Prof. Albert Bifet. His research focuses on streaming machine learning and anomaly detection, with particular emphasis on isolation forest methods for evolving data streams. His work addresses core challenges in stream learning, including concept drift adaptation, semi-supervised learning under label scarcity, and evaluation methodologies for imbalanced streaming data. His recent contributions include the Adaptive Isolation Forest (Discovery Science 2025) and the Streaming Isolation Forest (PAKDD 2025), alongside ongoing work on semi-supervised extensions of isolation-based detectors. He has submitted his doctoral thesis and expects to complete his degree in 2026. | ||
| + | </ | ||
| + | <WRAP clear></ | ||
| + | |||
| + | ==== 2026-10-15 ==== | ||
| + | <WRAP column 15%> | ||
| + | {{ : | ||
| + | </ | ||
| + | |||
| + | <WRAP column 75%> | ||
| + | |||
| + | **Speaker**: | ||
| + | |||
| + | **Title**: Group counterfactual explanations: | ||
| + | |||
| + | **Abstract**: | ||
| + | Existing methods for group counterfactual explanations often fail to generalize, rely on strict assumptions, | ||
| + | |||
| + | **Biogram**: | ||
| + | Enrique Valero Leal is a final-year ELLIS PhD student at both the Technical University of Madrid and Ludwig-Maximilians University of Munich. His research focuses on interpretable machine learning, | ||
| + | </ | ||
| + | <WRAP clear></ | ||
| ==== 2026-06-18 ==== | ==== 2026-06-18 ==== | ||