Stochastic simulations such as large-scale, spatiotemporal, age-structured epidemic models are computationally expensive at fine-grained resolution. While deep surrogate models can speed up the simulations, doing so for stochastic simulations and with active learning approaches is an underexplored area. We propose Interactive Neural Process (INP), a deep Bayesian active learning framework for learning deep surrogate models to accelerate stochastic simulations. INP consists of two components, a spatiotemporal surrogate model built upon Neural Process (NP) family and an acquisition function for active learning. For surrogate modeling, we develop Spatiotemporal Neural Process (STNP) to mimic the simulator dynamics. For active learning, we propose a novel acquisition function, Latent Information Gain (LIG), calculated in the latent space of NP based models. We perform a theoretical analysis and demonstrate that LIG reduces sample complexity compared with random sampling in high dimensions. We also conduct empirical studies on three complex spatiotemporal simulators for reaction diffusion, heat flow, and infectious disease. The results demonstrate that STNP outperforms the baselines in the offline learning setting and LIG achieves the state-of-the-art for Bayesian active learning.

Science and engineering fields use computer simulation extensively. These simulations are often run at multiple levels of sophistication to balance accuracy and efficiency. Multi-fidelity surrogate modeling reduces the computational cost by fusing different simulation outputs. Cheap data generated from low-fidelity simulators can be combined with limited high-quality data generated by an expensive high-fidelity simulator. Existing methods based on Gaussian processes rely on strong assumptions of the kernel functions and can hardly scale to high-dimensional settings. We propose Multi-fidelity Hierarchical Neural Processes (MF-HNP), a unified neural latent variable model for multi-fidelity surrogate modeling. MF-HNP inherits the flexibility and scalability of Neural Processes. The latent variables transform the correlations among different fidelity levels from observations to latent space. The predictions across fidelities are conditionally independent given the latent states. It helps alleviate the error propagation issue in existing methods. MF-HNP is flexible enough to handle non-nested high dimensional data at different fidelity levels with varying input and output dimensions. We evaluate MF-HNP on epidemiology and climate modeling tasks, achieving competitive performance in terms of accuracy and uncertainty estimation. In contrast to deep Gaussian Processes with only low-dimensional (< 10) tasks, our method shows great promise for speeding up high-dimensional complex simulations (over 7000 for epidemiology modeling and 45000 for climate modeling).

Deep learning is gaining increasing popularity for spatiotemporal forecasting. However, prior works have mostly focused on point estimates without quantifying the uncertainty of the predictions. In high stakes domains, being able to generate probabilistic forecasts with confidence intervals is critical to risk assessment and decision making. Hence, a systematic study of uncertainty quantification (UQ) methods for spatiotemporal forecasting is missing in the community. In this paper, we describe two types of spatiotemporal forecasting problems: regular grid-based and graph-based. Then we analyze UQ methods from both the Bayesian and the frequentist point of view, casting in a unified framework via statistical decision theory. Through extensive experiments on real-world road network traffic, epidemics, and air quality forecasting tasks, we reveal the statistical and computational trade-offs for different UQ methods: Bayesian methods are typically more robust in mean prediction, while confidence levels obtained from frequentist methods provide more extensive coverage over data variations. Computationally, quantile regression type methods are cheaper for a single confidence interval but require re-training for different intervals. Sampling based methods generate samples that can form multiple confidence intervals, albeit at a higher computational cost.

Alessandro Vespignani research activity is focused on the study of "techno-social" systems, where infrastructures composed of different technological layers are interoperating within the social component that drives their use and development. In this context we aim at understanding how the very same elements assembled in large number can give rise - according to the various forces and elements at play - to different macroscopic and dynamical behaviors, opening the path to quantitative computational approaches and forecasting power. The main research lines pursued at the moment are: Develop analytical and computational models for the co-evolution and interdependence of large-scale social, technological and biological networks. Modeling contagion processes in structured populations. Developing predictive computational tools for the analysis of the spatial spread of emerging diseases. Analyze the dynamics and evolution of information and social networks. Model the adaptive behavior of social systems. Prof. Vespignani is a joint appointment between the College of Science, the College of Computer and Information Science, and the Bouvé College of Health Sciences.

The availability of novel digital data streams that can be used as proxy for monitoring infectious disease incidence is ushering in a new era for real-time forecast approaches to disease spreading. Here, we propose the first seasonal influenza forecast framework based on a stochastic, spatially structured mechanistic model (individual level microsimulation) initialized with geo-localized microblogging data. The framework provides for more than 600 census areas in the United States, Italy and Spain, the initial conditions for a stochastic epidemic computational model that generates an ensemble of forecasts for the main indicators of the epidemic season: peak time and intensity. We evaluate the forecasts accuracy and reliability by comparing the results with the data from the official influenza surveillance systems in the US, Italy and Spain in the seasons 2014/15 and 2015/16. In all countries studied, the proposed framework provides reliable results with leads of up to 6 weeks that became more stable and accurate with progression of the season. The results for the United States have been generated in real-time in the context of the Centers for Disease Control and Prevention ``Forecasting the Influenza Season Challenge''. A characteristic feature of the mechanistic modeling approach is in the explicit estimate of key epidemiological parameters relevant for public health decision-making that cannot be achieved with statistical models that do not consider the disease dynamic. Furthermore, the presented framework allows the fusion of multiple data streams in the initialization stage and can be enriched with census, weather and socioeconomic data.

We analyze the traffic-weighted Web host graph obtained from a large sample of real Web users over about seven months. A number of interesting structural properties are revealed by this complex dynamic network, some in line with the well-studied boolean link host graph and others pointing to important differences. We find that while search is directly involved in a surprisingly small fraction of user clicks, it leads to a much larger fraction of all sites visited. The temporal traffic patterns display strong regularities, with a large portion of future requests being statistically predictable by past ones. Given the importance of topological measures such as PageRank in modeling user navigation, as well as their role in ranking sites for Web search, we use the traffic data to validate the PageRank random surfing model. The ranking obtained by the actual frequency with which a site is visited by users differs significantly from that approximated by the uniform surfing/teleportation behavior modeled by PageRank, especially for the most important sites. To interpret this finding, we consider each of the fundamental assumptions underlying PageRank and show how each is violated by actual user behavior

The understanding of the immense and intricate topological structure of the World Wide Web (WWW) is a major scientific and technological challenge. This has been recently tackled by characterizing the properties of its representative graphs, in which vertices and directed edges are identified with Web pages and hyperlinks, respectively. Data gathered in large-scale crawls have been analyzed by several groups resulting in a general picture of the WWW that encompasses many of the complex properties typical of rapidly evolving networks. In this article, we report a detailed statistical analysis of the topological properties of four different WWW graphs obtained with different crawlers. We find that, despite the very large size of the samples, the statistical measures characterizing these graphs differ quantitatively, and in some cases qualitatively, depending on the domain analyzed and the crawl used for gathering the data. This spurs the issue of the presence of sampling biases and structural differences of Web crawls that might induce properties not representative of the actual global underlying graph. In short, the stability of the widely accepted statistical description of the Web is called into question. In order to provide a more accurate characterization of the Web graph, we study statistical measures beyond the degree distribution, such as degree-degree correlation functions or the statistics of reciprocal connections. The latter appears to enclose the relevant correlations of the WWW graph and carry most of the topological information of the Web. The analysis of this quantity is also of major interest in relation to the navigability and searchability of the Web.

Automatic extraction of semantic information from text and links in Web pages is key to improving the quality of search results. However, the assessment of automatic semantic measures is limited by the coverage of user studies, which do not scale with the size, heterogeneity, and growth of the Web. Here we propose to leverage human-generated metadata --- namely topical directories --- to measure semantic relationships among massive numbers of pairs of Web pages or topics. The Open Directory Project classifies millions of URLs in a topical ontology, providing a rich source from which semantic relationships between Web pages can be derived. While semantic similarity measures based on taxonomies (trees) are well studied, the design of well-founded similarity measures for objects stored in the nodes of arbitrary ontologies (graphs) is an open problem. This paper defines an information-theoretic measure of semantic similarity that exploits both the hierarchical and non-hierarchical structure of an ontology. An experimental study shows that this measure improves significantly on the traditional taxonomy-based approach. This novel measure allows us to address the general question of how text and link analyses can be combined to derive measures of relevance that are in good agreement with semantic similarity. Surprisingly, the traditional use of text similarity turns out to be ineffective for relevance ranking.

We offer the first large-scale analysis of Web traffic based on network flow data. Using data collected on the Internet2 network, we constructed a weighted bipartite clientserver host graph containing more than 18 × 10^6 vertices and 68 × 10^6 edges valued by relative traffic flows. When considered as a traffic map of the World-Wide Web, the generated graph provides valuable information on the statistical patterns that characterize the global information flow on the Web. Statistical analysis shows that client-server connections and traffic flows exhibit heavy-tailed probability distributions lacking any typical scale. In particular, the absence of an intrinsic average in some of the distributions implies the absence of a prototypical scale appropriate for server design, Web-centric network design, or traffic modeling. The inspection of the amount of traffic handled by clients and servers and their number of connections highlights non-trivial correlations between information flow and patterns of connectivity as well as the presence of anomalous statistical patterns related to the behavior of users on the Web. The results presented here may impact considerably the modeling, scalability analysis, and behavioral study of Web applications.