Themis Palpanas

There is an increasingly pressing need, by several applications in diverse domains, for developing techniques able to manage and analyze very large collections of sequences, or data series. Examples of such applications come from various monitoring applications, including in power utility companies, where we need to apply machine learning techniques for knowledge extraction. It is not unusual for these applications to involve numbers of data series in the order of hundreds of millions to billions, which are often times not analyzed in their full detail due to their sheer size. However, no existing data management solution can offer native support for sequences and the corresponding operators necessary for complex analytics.

Foundation models enable the reuse of pretrained knowledge across time series classification (TSC) tasks, but transfer to a new task still requires target information to establish the task specific decision rule. Existing surveys mainly organize time series foundation models by architecture, pretraining objective, training data, or source modality, while the target classification procedure receives less attention. We instead organize foundation models for TSC into three transfer mechanisms: representation transfer, fine-tuning and adaptation, and in-context learning. Alongside the survey, we develop and release the TSC-FM Benchmark, a unified evaluation infrastructure covering 198 TSC datasets and multiple target settings. We use the benchmark to study how downstream classifiers, input settings, and task characteristics affect the assessment of pretrained models. The results show that the same pretrained representation can receive substantially different assessments under different downstream classifiers, that the best reported low shot model differs between univariate and multivariate settings, and that method groups lose accuracy at different rates as the number of classes increases. These findings suggest that transfer capability depends on the match between pretrained knowledge, target task structure, and the target procedure, rather than on the pretrained backbone alone. Future research should both expand the effective range of transfer and reliably identify when a target task falls outside that range.

Approximate Nearest Neighbor Search (ANNS) in high-dimensional Euclidean spaces is a fundamental problem with broad applications. Subspace Collision is a newly proposed ANNS framework that provides a novel paradigm for similarity search and achieves superior indexing and query performance. However, the subspace collision framework remains data-agnostic and query-oblivious, resulting in imbalanced index construction and wasted query overhead. In this paper, we address these limitations from two aspects: first, we design a subspace-oriented data transformation mechanism by averaging the entropies computed over each subspace of the transformed data, which ensures balanced subspace partitioning (in an information theoretical sense) and enables data-adaptive subspace collision; second, we present query-aware and scalable query strategies that dynamically allocate overhead for each query and accelerate collision probing within subspaces. Building on these ideas, we propose a novel data-adap t ive and query- a ware subspa c e c o llision method, abbreviated as TaCo, which achieves efficient and accurate ANN search while maintaining an excellent balance between indexing and query performance. Extensive experiments on real-world datasets demonstrate that, when compared to state-of-the-art subspace collision methods, TaCo achieves up to 8× speedup in indexing and reduces to 0.6× memory footprint, while achieving over 1.5× query throughput. Moreover, TaCo achieves state-of-the-art indexing performance and provides an effective balance between indexing and query efficiency, even when compared with advanced methods beyond the subspace-collision paradigm.

Visualizing multiple time series presents fundamental tradeoffs between scalability and visual clarity. Time series capture the behavior of many large-scale real-world processes, from stock market trends to urban activities. Users often gain insights by visualizing them as line charts, juxtaposing or superposing multiple time series to compare them and identify trends and patterns. However, existing representations struggle with scalability: when covering long time spans, leading to visual clutter from too many small multiples or overlapping lines. We propose TiVy, a new algorithm that summarizes time series using sequential patterns. It transforms the series into a set of symbolic sequences based on subsequence visual similarity using Dynamic Time Warping (DTW), then constructs a disjoint grouping of similar subsequences based on the frequent sequential patterns. The grouping result, a visual summary of time series, provides uncluttered superposition with fewer small multiples. Unlike common clustering techniques, TiVy extracts similar subsequences (of varying lengths) aligned in time. We also present an interactive time series visualization that renders large-scale time series in real-time. Our experimental evaluation shows that our algorithm (1) extracts clear and accurate patterns when visualizing time series data, (2) achieves a significant speed-up (1000ttttttttttttttttttttttttttt) compared to a straightforward DTW clustering. We also demonstrate the efficiency of our approach to explore hidden structures in massive time series data in two usage scenarios.

Approximate Nearest Neighbor Search (ANNS) presents an inherent tradeoff between performance and recall (i.e., result quality). Each ANNS algorithm provides its own algorithm-dependent parameters to allow applications to influence the recall/performance tradeoff of their searches. This situation is doubly problematic. First, the application developers have to experiment with these algorithm-dependent parameters to fine-tune the parameters that produce the desired recall for each use case. This process usually takes a lot of effort. Even worse, the chosen parameters may produce good recall for some queries, but bad recall for hard queries. To solve these problems, we present DARTH, a method that uses target declarative recall. DARTH uses a novel method for providing target declarative recall on top of an ANNS index by employing an adaptive early termination strategy integrated into the search algorithm. Through a wide range of experiments, we demonstrate that DARTH effectively meets user-defined recall targets while achieving significant speedups, up to 14.6x (average: 6.8x; median: 5.7x) faster than the search without early termination for HNSW and up to 41.8x (average: 13.6x; median: 8.1x) for IVF.