The Role of Adaptivity in Source Identification with Time Queries

Understanding epidemic propagation in large networks is an important but challenging task, especially since we usually lack information, and the information that we have is often counter-intuitive. An illustrative example is the dependence of the final size of the epidemic on the location of the initial infected agents (sources): common sense dictates that the most dangerous location for the sources is the largest city, but the second chapter of the thesis shows that this holds true only if the epidemic is just above the infection threshold.Identifying the initial infected agents can help us better understand the epidemic. The focus of this thesis is on identifying the very first infected agent, also called the source or patient zero. According to the standard assumptions, a few agents reveal their symptom onset time, and then it is our goal to identify the source based on this information, together with full knowledge of the underlying network. Unfortunately, even if we can choose the set of agents that are queried about their symptom onset time, the number of queries required for reliable source identification is too large for practical applications. In this thesis, we carefully assess if this issue can be mitigated by introducing adaptivity to the standard assumptions. Our main goal is to study the reduction in the query complexity if the queries can be chosen adaptively to previous answers, but we also investigate whether adaptively querying the edges can relax the full knowledge assumption on the network.Providing rigorous proofs about source identification with time queries is difficult. A notable exception is when the infection is passed with a known, deterministic delay from each agent to all of its neighbors, in which case the number of required non-adaptive and adaptive queries are equivalent to well-known notions in combinatorics; the metric dimension (MD) and the sequential metric dimension (SMD), respectively. We extend previous results in the field by computing the MD of a large class of random trees, where adaptivity can significantly reduce the query complexity, and the SMD of Erdos-Rényi random networks, where the reduction is found to be small, at most a constant factor. We address the case of non-deterministic diffusion processes for the first time in the mathematical literature: on the path graph, we observe a striking, double logarithmic decrease in adaptive query complexity compared to the non-adaptive case.Our analysis on the robustness of the MD to adding a single edge to specially constructed and d-dimensional grid networks suggests that even small changes in the network could easily derail source identification algorithms. This is concerning since it is difficult to obtain a perfect dataset about the entire contact network in practice. Inspired by recent implementations of contact tracing, we propose new source identification assumptions, where not only the symptom onset times, but also the edges of the network are queried by the algorithm, resulting in less, but potentially higher quality information. We propose two local search algorithms that outperform state of the art identification algorithms tailored to the new assumptions, and we analytically approximate their success probabilities on realistic random graph models. The adaptive assumptions enable us to evaluate our algorithms on a COVID-19 epidemic simulator: the first time that source identification algorithms are tested on such a complex dataset.

Graph Representation Learning with Optimal Transport: Analysis and Applications

Effrosyni Simou

In several machine learning settings, the data of interest are well described by graphs. Examples include data pertaining to transportation networks or social networks. Further, biological data, such as proteins or molecules, lend themselves well to graph-structured descriptions. In such settings, it is desirable to design representation learning algorithms that can take into account the information captured by the underlying structure. This thesis focuses on providing new methods to ingrain geometrical information in end-to-end deep learning architectures for graphs through the use of elements from Optimal Transport theory.First, we consider the setting where the data can be described by a fixed graph. In this setting, each datapoint corresponds to a node of the graph and the edges among nodes signify their pairwise relations. We propose an autoencoder architecture that consists of a linear layer in the encoder and a novel Wasserstein barycentric layer at the decoder. Our proposed barycentric layer takes into account the underlying geometry through the diffusion distance of the graph and provides a way to obtain structure-aware, non-linear interpolations, which lead to directly interpretable node embeddings that are stable to perturbations of the graph structure. Further, the embeddings obtained with our proposed autoencoder achieve competitive or superior results compared to state-of-the-art methods in node classification tasks.Second, we consider the setting where the data consist of multiple graphs. In that setting, the graphs can be of varying size and, therefore, a global pooling operation is needed in order to obtain representations of fixed size and enable end-to-end learning with deep networks. We propose a global pooling operation that optimally preserves the statistical properties of the representations. In order to do so, we take into account the geometry of the representation space and introduce a global pooling layer that minimizes the Wasserstein distance between a graph representation and its pooled counterpart, of fixed size. This is achieved by performing a Wasserstein gradient flow with respect to the pooled representation. Our proposed method demonstrates promising results in the task of graph classification.Overall, in this thesis we provide new methods for incorporating geometrical information in end-to-end deep learning architectures for graph structured data. We believe that our proposals will contribute to the development of algorithms that learn meaningful representations and that take fully into account the geometry of the data under consideration.

EPFL2022

Statistical Physics Methods for Community Detection

Chun Lam Chan

This thesis is devoted to information-theoretic aspects of community detection. The importance of community detection is due to the massive amount of scientific data today that describes relationships between items from a network, e.g., a social network. Items from this network can be inherently partitioned into a known number of communities, but the partition can only be inferred from the data. To estimate the underlying partition, data scientists can apply any type of advanced statistical techniques; but the data could be very noisy, or the number of data is inadequate. A fundamental question here is about the possibility of weak recovery: does the data contain a sufficient amount of information that enables us to produce a non-trivial estimate of the partition? For the purpose of mathematical analysis, the above problem can be formulated as Bayesian inference on generative models. These models, including the stochastic block model (SBM) and censored block model (CBM), consider a random graph generated based on a hidden partition that divides the nodes in the graph into labelled groups. In the SBM, nodes are connected with a probability depending on the labels of the endpoints. Whereas, in the CBM, hidden variables are measured through a noisy channel, and the measurement outcomes form a weighted graph. In both models, inference is the task of recovering the hidden partition from the observed graph. The criteria for weak recovery can be studied via an information-theoretic quantity called mutual information. Once the asymptotic mutual information is computed, phase transitions for the weak recovery can be located. This thesis pertains to rigorous derivations of single-letter variational expressions for the asymptotic mutual information for models in community detection. These variational expressions, known as the replica predictions, come from heuristic methods of statistical physics. We present our development of new rigorous methods for confirming the replica predictions. These methods are based on extending the recently introduced adaptive interpolation method. We prove the replica prediction for the SBM in the dense-graph regime with two groups of asymmetric size. The existing proofs in the literature are indirect, as they involve mapping the model to an external problem whose mutual information is determined by a combination of methods. Here, on the contrary, we provide a self-contained and direct proof. Next, we extend this method to sparse models. Before this thesis, adaptive interpolation was known for providing a conceptually simple proof for replica predictions for dense graphs. Whereas, for a sparse graph, the replica prediction involves a more complicated variational expression, and rigorous confirmations are often lacking or obtained by rather complicated methods. Therefore, we focus on a simple version of CBM on sparse graphs, where hidden variables are measured through a binary erasure channel, for which we fully prove the replica prediction by the adaptive interpolation. The key for extending the adaptive interpolation to a broader class of sparse models is a concentration result for the so-called "multi-overlaps". This concentration forms the basis of the replica "symmetric" prediction. We prove this concentration result for a related sparse model in the context of physics. This provides inspiration for further development of the adaptive interpolation.

EPFL2019

Distributed constraint satisfaction for coordinating and integrating a large-scale, heterogeneous enterprise

Carlos Eisenberg

Market forces are continuously driving public and private organisations towards higher productivity, shorter process and production times, and fewer labour hours. To cope with these changes, organisations are adopting new organisational models of coordination and cooperation that increase their flexibility, consistency, efficiency, productivity and profit margins. In this thesis an organisational model of coordination and cooperation is examined using a real life example; the technical integration of a distributed large-scale project of an international physics collaboration. The distributed resource constraint project scheduling problem is modelled and solved with the methods of distributed constraint satisfaction. A distributed local search method, the distributed breakout algorithm (DisBO), is used as the basis for the coordination scheme. The efficiency of the local search method is improved by extending it with an incremental problem solving scheme with variable ordering. The scheme is implemented as central algorithm, incremental breakout algorithm (IncBO), and as distributed algorithm, distributed incremental breakout algorithm (DisIncBO). In both cases, strong performance gains are observed for solving underconstrained problems. Distributed local search algorithms are incomplete and lack a termination guarantee. When problems contain hard or unsolvable subproblems and are tightly or overconstrained, local search falls into infinite cycles without explanation. A scheme is developed that identifies hard or unsolvable subproblems and orders these to size. This scheme is based on the constraint weight information generated by the breakout algorithm during search. This information, combined with the graph structure, is used to derive a fail first variable order. Empirical results show that the derived variable order is 'perfect'. When it guides simple backtracking, exceptionally hard problems do not occur, and, when problems are unsolvable, the fail depth is always the shortest. Two hybrid algorithms, BOBT and BOBT-SUSP are developed. When the problem is unsolvable, BOBT returns the minimal subproblem within the search scope and BOBT-SUSP returns the smallest unsolvable subproblem using a powerful weight sum constraint. A distributed hybrid algorithm (DisBOBT) is developed that combines DisBO with DisBT. The distributed hybrid algorithm first attempts to solve the problem with DisBO. If no solution is available after a bounded number of breakouts, DisBO is terminated, and DisBT solves the problem. DisBT is guided by a distributed variable order that is derived from the constraint weight information and the graph structure. The variable order is incrementally established, every time the partial solution needs to be extended, the next variable within the order is identified. Empirical results show strong performance gains, especially when problems are overconstrained and contain small unsolvable subproblems.

EPFL2003