The widespread adoption of digital services, along with the scale and complexity at which they operate, has made incidents in IT operations increasingly more likely, diverse, and impactful. This has led to the rapid development of a central aspect of "Artificial Intelligence for IT Operations" (AIOps), focusing on detecting anomalies in vast amounts of multivariate time series data generated by service entities. In this paper, we begin by introducing a unifying framework for benchmarking unsupervised anomaly detection (AD) methods, and highlight the problem of shifts in normal behaviors that can occur in practical AIOps scenarios. To tackle anomaly detection under domain shift, we then cast the problem in the framework of domain generalization and propose a novel approach, Domain-Invariant VAE for Anomaly Detection (DIVAD), to learn domain-invariant representations for unsupervised anomaly detection. Our evaluation results using the Exathlon benchmark show that the two main DIVAD variants significantly outperform the best unsupervised AD method in maximum performance, with 20% and 15% improvements in maximum peak F1-scores, respectively. Evaluation using the Application Server Dataset further demonstrates the broader applicability of our domain generalization methods.

Access to high-quality data repositories and benchmarks have been instrumental in advancing the state of the art in many experimental research domains. While advanced analytics tasks over time series data have been gaining lots of attention, lack of such community resources severely limits scientific progress. In this paper, we present Exathlon, the first comprehensive public benchmark for explainable anomaly detection over high-dimensional time series data. Exathlon has been systematically constructed based on real data traces from repeated executions of large-scale stream processing jobs on an Apache Spark cluster. Some of these executions were intentionally disturbed by introducing instances of six different types of anomalous events (e.g., misbehaving inputs, resource contention, process failures). For each of the anomaly instances, ground truth labels for the root cause interval as well as those for the extended effect interval are provided, supporting the development and evaluation of a wide range of anomaly detection (AD) and explanation discovery (ED) tasks. We demonstrate the practical utility of Exathlon's dataset, evaluation methodology, and end-to-end data science pipeline design through an experimental study with three state-of-the-art AD and ED techniques.

In this demo, we introduce Exathlon - a new benchmarking platform for explainable anomaly detection over high-dimensional time series. We designed Exathlon to support data scientists and researchers in developing and evaluating learned models and algorithms for detecting anomalous patterns as well as discovering their explanations. This demo will showcase Exathlon's curated anomaly dataset, novel benchmarking methodology, and end-to-end data science pipeline in action via example usage scenarios.

The efficient coding hypothesis predicts that sensory neurons adjust their coding resources to optimally represent the stimulus statistics of their environment. To test this prediction in the moth olfactory system, we have developed a stimulation protocol that mimics the natural temporal structure within a turbulent pheromone plume. We report that responses of antennal olfactory receptor neurons to pheromone encounters follow the temporal fluctuations in such a way that the most frequent stimulus timescales are encoded with maximum accuracy. We also observe that the average coding precision of the neurons adjusted to the stimulus-timescale statistics at a given distance from the pheromone source is higher than if the same encoding model is applied at a shorter, non-matching, distance. Finally, the coding accuracy profile and the stimulus-timescale distribution are related in the manner predicted by the information theory for the many-to-one convergence scenario of the moth peripheral sensory system.

Long-distance olfactory search behaviors depend on odor detection dynamics. Due to turbulence, olfactory signals travel as bursts of variable concentration and spacing and are characterized by long-tail distributions of odor/no-odor events, challenging the computing capacities of olfactory systems. How animals encode complex olfactory scenes to track the plume far from the source remains unclear. Here we focus on the coding of the plume temporal dynamics in moths. We compare responses of olfactory receptor neurons (ORNs) and antennal lobe projection neurons (PNs) to sequences of pheromone stimuli either with white-noise patterns or with realistic turbulent temporal structures simulating a large range of distances (8 to 64 m) from the odor source. For the first time, we analyze what information is extracted by the olfactory system at large distances from the source. Neuronal responses are analyzed using linear-nonlinear models fitted with white-noise stimuli and used for predicting responses to turbulent stimuli. We found that neuronal firing rate is less correlated with the dynamic odor time course when distance to the source increases because of improper coding during long odor and no-odor events that characterize large distances. Rapid adaptation during long puffs does not preclude however the detection of puff transitions in PNs. Individual PNs but not individual ORNs encode the onset and offset of odor puffs for any temporal structure of stimuli. A higher spontaneous firing rate coupled to an inhibition phase at the end of PN responses contributes to this coding property. This allows PNs to decode the temporal structure of the odor plume at any distance to the source, an essential piece of information moths can use in their tracking behavior.