arXiv subjects

Florent Masseglia

Publications and source records attributed to Florent Masseglia.

5 recordsLinked to original sources

arXiv2026

Disjoint or Overlapping? Inference Windowing for Reconstruction-Based Time Series Anomaly Detection

Reconstruction-based methods are widely used for time series anomaly detection, where models are trained to reconstruct subsequences, and anomalies are identified through reconstruction errors. However, reported results are often hard to compare due to heterogeneous evaluation practices and underspecified inference procedures. In this paper, we revisit reconstruction-based anomaly detection in the univariate offline setting and study the role of the inference stride, which controls whether subsequences are processed as disjoint windows or with overlap. We propose a unified training, tuning, and multi-seed evaluation protocol on the curated TSB-AD benchmark, and study how overlapping inference affects anomaly detection performance for a range of reconstruction models, including PCA-based baselines, DLinear, an AutoEncoder, TimesNet, and Transformer variants. The results show that across all models, overlapping windows yield consistent improvements, with average relative gain up to +28%, and can alter method rankings. We further analyze variability across datasets, random seeds, and hyperparameter configurations. Finally, we complement the benchmark study with an evaluation on the full UCR archive using localization criteria aligned with sliding-window reconstruction. Overall, our results highlight that reconstruction-based anomaly detection performance depends not only on model architecture and training, but also on inference choices, motivating a clear and reproducible protocol. Our results show that reconstructionbased baselines achieve strong performance on both TSB-AD and UCR benchmarks, supporting them as competitive and practical approaches for univariate time series anomaly detection.

cs.LG↗

arXiv2024

SoftED: Metrics for Soft Evaluation of Time Series Event Detection

Time series event detection methods are evaluated mainly by standard classification metrics that focus solely on detection accuracy. However, inaccuracy in detecting an event can often result from its preceding or delayed effects reflected in neighboring detections. These detections are valuable to trigger necessary actions or help mitigate unwelcome consequences. In this context, current metrics are insufficient and inadequate for the context of event detection. There is a demand for metrics that incorporate both the concept of time and temporal tolerance for neighboring detections. This paper introduces SoftED metrics, a new set of metrics designed for soft evaluating event detection methods. They enable the evaluation of both detection accuracy and the degree to which their detections represent events. They improved event detection evaluation by associating events and their representative detections, incorporating temporal tolerance in over 36\% of experiments compared to the usual classification metrics. SoftED metrics were validated by domain specialists that indicated their contribution to detection evaluation and method selection.

cs.LG↗

arXiv2020

Feedback Clustering for Online Travel Agencies Searches: a Case Study

Understanding choices performed by online customers is a growing need in the travel industry. In many practical situations, the only available information is the flight search query performed by the customer with no additional profile knowledge. In general, customer flight bookings are driven by prices, duration, number of connections, and so on. However, not all customers might assign the same importance to each of those criteria. Here comes the need of grouping together all flight searches performed by the same kind of customer, that is having the same booking criteria. The effectiveness of some set of recommendations, for a single cluster, can be measured in terms of the number of bookings historically performed. This effectiveness measure plays the role of a feedback, that is an external knowledge which can be recombined to iteratively obtain a final segmentation. In this paper, we describe our Online Travel Agencies (OTA) flight search use case and highlight its specific features. We address the flight search segmentation problem motivated above by proposing a novel algorithm called Split-or-Merge (S/M). This algorithm is a variation of the Split-Merge-Evolve (SME) method. The SME method has already been introduced in the community as an iterative process updating a clustering given by the K-means algorithm by splitting and merging clusters subject to feedback independent evaluations. No previous application of the SME method to the real-word data is reported in literature to the best of our knowledge. Here, we provide experimental evaluations over real-world data to the SME and the S/M methods. The impact on our domain-specific metrics obtained under the SME and the S/M methods suggests that feedback clustering techniques can be very promising in the handling of the domain of OTA flight searches.

cs.IR↗

arXiv2019

Apprentissage de la pensée informatique : de la formation des enseignant$\cdot$e$\cdot$s à la formation de tou$\cdot$te$\cdot$s les citoyen$\cdot$ne$\cdot$s

In recent years, in France, computer learning (under the term of code) has entered the school curriculum, in primary and high school. This learning is also aimed at developing computer thinking to enable students, girls and boys, to start master all aspects of the digital world (science, technology, industry, culture). However, neither teachers, nor parents are trained to teach or educate on these topics. Furthermore, if the educational system progresses progressively towards these objectives, in everyday life and in professional context there is also a need for lifelong training in computer thinking. Large-scale projects on coding initiation are now quite successful in supporting the training of professionals in education on these topics. However, they require an infrastructure of people and important resources to maintain their level of efficiency. In order to further develop the objectives ofhelping people to demystify IT thinking, we aim to question here the way by which it is possible to conceive a concrete and operational initiative that addresses this issue. A huge challenge: Let's share a proposal here and discuss it.

cs.CY↗

arXiv2018

A Distributed Collaborative Filtering Algorithm Using Multiple Data Sources

Collaborative Filtering (CF) is one of the most commonly used recommendation methods. CF consists in predicting whether, or how much, a user will like (or dislike) an item by leveraging the knowledge of the user's preferences as well as that of other users. In practice, users interact and express their opinion on only a small subset of items, which makes the corresponding user-item rating matrix very sparse. Such data sparsity yields two main problems for recommender systems: (1) the lack of data to effectively model users' preferences, and (2) the lack of data to effectively model item characteristics. However, there are often many other data sources that are available to a recommender system provider, which can describe user interests and item characteristics (e.g., users' social network, tags associated to items, etc.). These valuable data sources may supply useful information to enhance a recommendation system in modeling users' preferences and item characteristics more accurately and thus, hopefully, to make recommenders more precise. For various reasons, these data sources may be managed by clusters of different data centers, thus requiring the development of distributed solutions. In this paper, we propose a new distributed collaborative filtering algorithm, which exploits and combines multiple and diverse data sources to improve recommendation quality. Our experimental evaluation using real datasets shows the effectiveness of our algorithm compared to state-of-the-art recommendation algorithms.

cs.IR↗