1,721,024 research outputs found
Master programme in data science and scientific computing – A joint effort of the Universities of Trieste, Udine and SISSA
Full text linkRetrieval-Augmented Mining of Temporal Logic Specifications from Data
No full textThe integration of cyber-physical systems (CPS) into everyday life raises the critical necessity of ensuring their safety and reliability. An important step in this direction is requirement mining, i.e. inferring formally specified system properties from observed behaviors, in order to discover knowledge about the system. Signal Temporal Logic (STL) offers a concise yet expressive language for specifying requirements, particularly suited for CPS, where behaviors are typically represented as time series data. This work addresses the task of learning STL requirements from observed behaviors in a data-driven manner, focusing on binary classification, i.e. on inferring properties of the system which are able to discriminate between regular and anomalous behaviour, and that can be used both as classifiers and as monitors of the compliance of the CPS to desirable specifications. We present a novel framework that combines Bayesian Optimization (BO) and Information Retrieval (IR) techniques to simultaneously learn both the structure and the parameters of STL formulae, without restrictions on the STL grammar. Specifically, we propose a framework that leverages a dense vector database containing semantic-preserving continuous representations of millions of formulae, queried for facilitating the mining of requirements inside a BO loop. We demonstrate the effectiveness of our approach in several signal classification applications, showing its ability to extract interpretable insights from system executions and advance the state-of-the-art in requirement mining for CPS
Hybrid systems and biology. continuous and discrete modeling for systems biology
No full textHybrid Systems are dynamical systems presenting both discrete and continuous evolution. Hybrid Automata are a formal model for hybrid systems, originally proposed to study embedded systems, where a discrete control acts on a continuously changing environment. The presence of both discrete and continuous dynamics makes this formalism appealing also for modeling biological systems. However, the situation in this case is subtler, as there is no natural separation into discrete and continuous components. No surprise, then, that hybrid automata have been used in systems biology in rather different ways. Some approaches, like the description of biological switches, concentrate on the use of model-checking routines. Other applications, like the switching between continuous and discrete/stochastic simulation, focus on the exploitation of the interplay between discreteness and continuity in order to reduce the computational burden of numerical simulation, yet maintaining an acceptable precision. We will survey the use of hybrid automata in systems biology, through a series of cases studies that we deem interesting and paradigmatic
Refining Mean-field Approximations by Dynamic State Truncation
No full textMean-field models are an established method to analyze large stochastic systems with N interacting objects by means of simple deterministic equations that are asymptotically correct when N tends to infinity. For finite N, mean-field equations provide an approximation whose accuracy is model- and parameter-dependent. Recent research has focused on refining the approximation by computing suitable quantities associated with expansions of order and to the mean-field equation. In this paper we present a new method for refining mean-field approximations. It couples the master equation governing the evolution of the probability distribution of a truncation of the original state space with a mean-field approximation of a time-inhomogeneous population process that dynamically shifts the truncation across the whole state space. We provide a result of asymptotic correctness in the limit when the truncation covers the state space; for finite truncations, the equations give a correction of the mean-field approximation. We apply our method to examples from the literature to show that, even with modest truncations, it is effective in models that cannot be refined using existing techniques due to non-differentiable drifts, and that it can outperform the state of the art in challenging models that cause instability due orbit cycles in their mean-field equations
Fluid approximation of broadcasting systems
Full text linkNature-inspired paradigms have been proposed to design and forecast behaviour of open distributed systems, such as sensor networks and the internet of things. In these paradigms system behaviour emerges from (complex) interactions among a large number of agents. Modelling these interactions in terms of classical point-to-point communication is often not practical. This is due to the large scale and the open nature of the systems, which means that partners for point-to-point communication may not be available at any given time. Nevertheless the need for efficient formal verification of qualitative and quantitative properties of these systems is of utmost importance, especially given their proposed pervasive and transparent nature. CARMA is a recently proposed formal modelling language for open distributed systems, which is equipped with a broadcast communication in order to meet the communication challenges of such systems. The inclusion of quantitative information about the timing and probability of actions gives rise to models suitable for analysing questions such as the probability that information will achieve total coverage within a system, or the expected market share that might be gained by competing service providers relying on viral advertising. The ability to express models is not the only challenge, because the scale of the systems we are interested in often defies discrete state-based analysis techniques such as stochastic simulation. This is the problem that we address in this paper as we consider how to provide an efficient fluid approximation, supporting efficient and accurate quantitative analysis of large scale systems, for a language that incorporates broadcast communication
Resilience of Bayesian Layer-Wise Explanations under Adversarial Attacks
Full text linkWe consider the problem of the stability of saliency-based explanations of Neural Network predictions under adversarial attacks in a classification task. Saliency interpretations of deterministic Neural Networks are remarkably brittle even when the attacks fail, i.e. for attacks that do not change the classification label. We empirically show that interpretations provided by Bayesian Neural Networks are considerably more stable under adversarial perturbations of the inputs and even under direct attacks to the explanations. By leveraging recent results, we also provide a theoretical explanation of this result in terms of the geometry of the data manifold. Additionally, we discuss the stability of the interpretations of high level representations of the inputs in the internal layers of a Network. Our results demonstrate that Bayesian methods, in addition to being more robust to adversarial attacks, have the potential to provide more stable and interpretable assessments of Neural Network predictions
Neural Predictive Monitoring for Collective Adaptive Systems
Full text linkReliable bike-sharing systems can lead to numerous environmental, economic and social benefits and therefore play a central role in the effective development of smart cities. Bike-sharing models deal with spatially distributed stations and interact with an unpredictable environment, the users. Monitoring the trustworthiness of such a collective system is of paramount importance to ensure a good quality of the delivered service, but this task can become computationally demanding due to the complexity of the model under study. Neural Predictive Monitoring (NPM) [5], a neural-network learning-based approach to predictive monitoring (PM) with statistical guarantees, can be employed to preemptively detect violations of a specific requirement – e.g. a station has no more bikes available or a station is full. The computational efficiency of NPM makes PM applicable at runtime even on embedded devices with limited computational power. The goal of this paper is to demonstrate the applicability of NPM on collective adaptive systems such as bike-sharing systems. In particular, we first analyze the performance of NPM over a collective system evolving deterministically. Then, following [7], we tackle a more realistic scenario, where sensors allow only for partial observability and where the system evolves in a stochastic fashion. We evaluate the approach on multiple bike-sharing network topologies, obtaining highly accurate predictions and effective error detection rules
Neural Predictive Monitoring Under Partial Observability
Full text linkWe consider the problem of predictive monitoring (PM), i.e., predicting at runtime future violations of a system from the current state. We work under the most realistic settings where only partial and noisy observations of the state are available at runtime. Such settings directly affect the accuracy and reliability of the reachability predictions, jeopardizing the safety of the system. In this work, we present a learning-based method for PM that produces accurate and reliable reachability predictions despite partial observability (PO). We build on Neural Predictive Monitoring (NPM), a PM method that uses deep neural networks for approximating hybrid systems reachability, and extend it to the PO case. We propose and compare two solutions, an end-to-end approach, which directly operates on the rough observations, and a two-step approach, which introduces an intermediate state estimation step. Both solutions rely on conformal prediction to provide 1) probabilistic guarantees in the form of prediction regions and 2) sound estimates of predictive uncertainty. We use the latter to identify unreliable (and likely erroneous) predictions and to retrain and improve the monitors on these uncertain inputs (i.e., active learning). Our method results in highly accurate reachability predictions and error detection, as well as tight prediction regions with guaranteed coverage
Timeseria: An object-oriented time series processing library
No full textTimeseria is an object-oriented time series processing library implemented in Python, which aims at making it easier to manipulate time series data and to build statistical and machine learning models on top of it. Unlike common data analysis frameworks, it builds up from well defined and reusable logical units (objects), which can be easily combined together in order to ensure a high level of consistency. Thanks to this approach, Timeseria can address by design several non-trivial issues which are often underestimated, such as handling data losses, non-uniform sampling rates, differences between aggregated data and punctual observations, time zones, daylight saving times, and more. Timeseria comes with a comprehensive set of base data structures, data transformations for resampling and aggregation, common data manipulation operations, and extensible models for data reconstruction, forecasting and anomaly detection. It also integrates a fully featured, interactive plotting engine capable of handling even millions of data points
- …
