Publications

We propose MicroPPO, a reinforcement learning approach for real-time management of power flows in small-scale energy systems.

Hoeffding Trees (HT) and Extremely Fast Decision Trees (EFDT) are popular for mining data streams, with EFDT offering faster learning but suffering from accuracy drops due to subtree pruning. To address this, we propose PLASTIC, an incremental decision tree that restructures pruned subtrees without impacting predictions, leveraging decision tree plasticity.

We present a UCB-sampling policy for the Budgeted MAB problem that uses asymmetric confidence intervals to overcome issues of existing policies; our policy achieves logarithmic regret and outperforms existing policies in synthetic and real settings.

Our paper presents ABCD, a novel approach for change detection in high-dimensional data streams. ABCD detects changes accurately and provides insights into the specific subspace where changes occur. By leveraging an encoder-decoder model and Bernstein’s inequality, ABCD quantifies the severity of changes and outperforms other methods in our experiments.

We address the problem of distinguishing between different types of outliers in decentralized data. We present a “tandem” solution that combines local and federated outlier detectors to effectively identify those types.

We propose an algorithm, Maximum Mean Discrepancy Adaptive Windowing (MMDAW), which leverages the well-known Maximum Mean Discrepancy (MMD) two-sample test, and facilitates its efficient online computation on windows whose size it flexibly adapts.

We introduce EvoAutoML, an evolution-based online learning framework consisting of heterogeneous and connectable models that supports large and diverse configuration spaces and adapts to the online learning scenario.

We present the vision of Fogsy, a holistic industrial AI management system aiding to support domain experts to manage analytical AI pipelines.