We present a new universal source code for distributions of unlabeled binary and ordinal trees that achieves optimal compression to within lower order terms for all tree sources covered by existing universal codes. At the same time, it supports answering many navigational queries on the compressed representation in constant time on the word-RAM; this is not known to be possible for any existing tree compression method. The resulting data structures, “hypersuccinct trees”, hence combine the compression achieved by the best known universal codes with the operation support of the best succinct tree data structures. We apply hypersuccinct trees to obtain a universal compressed data structure for range-minimum queries. It has constant query time and the optimal worst-case space usage of 2n + o(n) bits, but the space drops to 1.736n + o(n) bits on average for random permutations of n elements, and 2 lg ( n r ) + o(n) for arrays with r increasing runs, respectively. Both results are optimal; the former answers an open problem of Davoodi et al. (2014) and Golin et al. (2016). Compared to prior work on succinct data structures, we do not have to tailor our data structure to specific applications; hypersuccinct trees automatically adapt to the trees at hand. We show that they simultaneously achieve the optimal space usage to within lower order terms for a wide range of distributions over tree shapes, including: binary search trees (BSTs) generated by insertions in random order / Cartesian trees of random arrays, random fringe-balanced BSTs, binary trees with a given number of binary/unary/leaf nodes, random binary tries generated from memoryless sources, full binary trees, unary paths, as well as uniformly chosen weight-balanced BSTs, AVL trees, and left-leaning red-black trees. 2012 ACM Subject Classification Theory of computation → Data compression; Theory of computation → Data structures design and analysis

Motivated by a network fault detection problem, we study how recall can be boosted in a decision tree classifier, without sacrificing too much precision. This problem is relevant and novel in the context of transfer learning(TL), in which few target domain training samples are available. We define a geometric optimization problem for boosting the recall of a decision tree classifier, and show it is NP-hard. To solve it efficiently, we propose several near-linear time heuristics, and experimentally validate these heuristics in the context of TL. Our evaluation includes 7 public datasets, as well as 6 network fault datasets, and we compare our heuristics with several existing TL algorithms, as well as exact mixed integer linear programming(MILP) solutions to our optimization problem. We find that our heuristics boost recall in a manner similar to optimal MILP solutions, yet require several orders of magnitude less compute time. In many cases the F1 score of our approach is competitive, and often better, than other TL algorithms. Moreover, our approach can be used as a building block to apply transfer learning to more powerful ensemble methods, such as random forests.

The energy consumption of mobile networks is already substantial nowadays, and only expected to further increase with the roll-out of 5G. Base stations are the key elements in this context: reducing their energy consumption is of paramount importance for network operators, not only to lower operating costs, but also to meet sustainable development goals. Today's base stations are typically over-provisioned, i.e., they comprise multiple cells to meet the peak load in a region. Therefore, substantial energy savings are possible by switching off cells that are under-utilized. This article proposes a data-driven approach to determine the time periods when a cell can be switched off. Forecasting is used to accurately predict network utilization and automatically find the time intervals to reliably switch off a cell. We carefully analyze the requirements of the system as a whole, from data collection to forecasting methods, to enable effective energy savings in practice. Considering several real-world traces from LTE networks, we show that an average of 10.24% energy savings is possible. We explore the trade-offs between energy savings and overhead in switching off cells, and provide insights into the choice of methods accordingly. In particular, we show that the accuracy of forecasting is not the most important factor in achieving energy savings; instead, the prediction (uncertainty) interval plays a key role in being able to achieve energy savings with less impact on end-users. Finally, we propose a model to generate utilization traces that match the distribution of real-world traces obtained from cellular networks.

Operating large scale and feature-rich applications is becoming increasingly complex as engineers need to deploy highly configurable software releases on distributed cloud stacks while managing ever-shorter production cycles. Although recent proposals attempt to streamline cloud resources orchestration, there is still a significant challenge in making such solutions generalize to unseen cloud stacks. In other words, the behavior of application-specific Key Performance Indicators (KPIs) and resource configurations, crafted for specific stacks, may differ on heterogeneous deployments, requiring time-consuming policy adjustments. We introduce IMITA, a system that leverages imitation learning to create models by imitating an expert behavior that can be generalized seamlessly to new cloud stacks. To make a generalized model, IMITA maps expert actions taken based on the application KPI space to the space of resource utilization metrics that are universally available in cloud platforms. This mapping enables the model to trigger actions, mimicking expert behavior, upon the occurrence of similar re-source utilization footprints across deployments. We demonstrate IMITA by learning to scale-out Cassandra deployments with diverse configurations and workloads. Our results show IMITA can replicate expert actions across deployments and extrapolate to unseen environments by achieving 50 - 94% fewer false positives actions than traditional threshold-based policies while still adhering to Service-Level Objectives (SLO) and avoiding under-provisioning of resources. Moreover, since collecting data in clouds is costly, IMITA gathers data only for representative configurations to train the imitator model. This approach reduces the size of the collected data to 50%.

Approximate Bayesian inference for neural networks is considered a robust alternative to standard training, often providing good performance on out-of-distribution data. However, Bayesian neural networks (BNNs) with high-fidelity approximate inference via full-batch Hamiltonian Monte Carlo achieve poor generalization under covariate shift, even underperforming classical estimation. We explain this surprising result, showing how a Bayesian model average can in fact be problematic under covariate shift, particularly in cases where linear dependencies in the input features cause a lack of posterior contraction. We additionally show why the same issue does not affect many approximate inference procedures, or classical maximum a-posteriori (MAP) training. Finally, we propose novel priors that improve the robustness of BNNs to many sources of covariate shift.

The virtual resources of 5G networks are expected to scale and support migration to other locations within the substrate. In this context, a configuration for 5G network slices details the instantaneous mapping of the virtual resources across all slices on the substrate, and a feasible configuration satisfies the Service-Level Objectives (SLOs) without overloading the substrate. Reconfiguring a network from a given source configuration to the desired target configuration involves identifying an ordered sequence of feasible configurations from the source to the target. The proposed solutions for finding such a sequence are optimized for data centers and cannot be used as-is for reconfiguring 5G network slices. We present Matryoshka, our divide-and-conquer approach for finding a sequence of feasible configurations that can be used to reconfigure 5G network slices. Unlike previous approaches, Matryoshka also considers the bandwidth and latency constraints between the network functions of network slices. Evaluating Matryoshka required a dataset of pairs of source and target configurations. Because such a dataset is currently unavailable, we analyze proof of concept roll-outs, trends in standardization bodies, and research sources to compile an input dataset. On using Matryoshka on our dataset, we observe that it yields close-to-optimal reconfiguration sequences 10X faster than existing approaches.

The adoption of fast-paced practices for developing virtual network functions (VNFs) allows for continuous software delivery and creates a market advantage for network operators. This adoption, however, is problematic for testing engineers that need to assure, in shorter development cycles, certain quality of highly-configurable product releases running on heterogeneous clouds. Machine learning (ML) can accelerate testing workflows by detecting performance issues in new software builds. However, the overhead of maintaining several models for all combinations of build types, network configurations, and other stack parameters, can quickly become prohibitive and make the application of ML infeasible. We propose Env2Vec, a deep learning architecture that combines contextual features with historical resource usage, and characterizes the various stack parameters that influence the test execution within an embedding space, which allows it to generalize model predictions to previously unseen environments. We integrate a single ML model in the testing workflow to automatically debug errors and pinpoint performance bottlenecks. Results obtained with real testing data show an accuracy between 86.2%-100%, while reducing the false alarm rate by 20.9%-38.1% when reporting performance issues compared to state-of-the-art approaches.

Automated construction of knowledge hierarchies from huge data corpora is gaining increasing attention in recent years, in order to tackle the infeasibility of manually extracting and semantically linking millions of concepts. As a knowledge hierarchy evolves with these automated techniques, there is a need for measures to assess its temporal evolution, quantifying the similarities between different versions and identifying the relative growth of different subgraphs in the knowledge hierarchy. In this paper, we focus on measures that leverage structural properties of the knowledge hierarchy graph to assess the temporal changes. We propose a principled and scalable similarity measure, based on Katz similarity between concept nodes, for comparing different versions of a knowledge hierarchy, modeled as a generic directed acyclic graph. We present theoretical analysis to depict that the proposed measure accurately captures the salient properties of taxonomic hierarchies, assesses changes in the ordering of nodes, along with the logical subsumption of relationships among concepts. We also present a linear time variant of the measure, and show that our measures, unlike previous approaches, are tunable to cater to diverse application needs. We further show that our measure provides interpretability, thereby identifying the key structural and logical difference in the hierarchies. Experiments on a real DBpedia and biological knowledge hierarchy showcase that our measures accurately capture structural similarity, while providing enhanced scalability and tunability. Also, we demonstrate that the temporal evolution of different subgraphs in this knowledge hierarchy, as captured purely by our structural measure, corresponds well with the known disruptions in the related subject areas.

As Multi-access Edge Computing (MEC) and 5G technologies evolve, new applications are emerging with unprecedented capacity and real-time requirements. At the core of such applications there is a need for machine learning (ML) to create value from the data at the edge. Current ML systems transfer data from geo-distributed streams to a central datacenter for modeling. The model is then moved to the edge and used for inference or classification. These systems can be ineffective because they introduce significant demand for data movement and model transfer in the critical path of learning. Furthermore, a full model may not be needed at each edge location. An alternative is to train and update the models online at each edge with local data, in isolation from other edges. Still, this approach can worsen the accuracy of models due to reduced data availability, especially in the presence of local data shifts. In this paper we propose Cartel, a system for collaborative learning in edge clouds, that creates a model-sharing environment in which tailored models at each edge can quickly adapt to changes, and can be as robust and accurate as centralized models. Results show that Cartel adapts to workload changes 4 to 8x faster than isolated learning, and reduces model size, training time and total data transfer by 3x, 5.7x and ~1500x, respectively, when compared to centralized learning.

Today, software operation engineers rely on application key performance indicators (KPIs) for sizing and orchestrating cloud resources dynamically. KPIs are monitored to assess the achievable performance and to configure various cloud-specific parameters such as flavors of instances and autoscaling rules, among others. Usually, keeping KPIs within acceptable levels requires application expertise which is expensive and can slow down the continuous delivery of software. Expertise is required because KPIs are normally based on application-specific quality-of-service metrics, like service response time and processing rate, instead of generic platform metrics, like those typical across various environments (e.g., CPU and memory utilization, I/O rate, etc.) In this paper, we investigate the feasibility of outsourcing the management of application performance from developers to cloud operators. In the same way that the serverless paradigm allows the execution environment to be fully managed by a third party, we discuss a monitorless model to streamline application deployment by delegating performance management. We show that training a machine learning model with platform-level data, collected from the execution of representative containerized services, allows inferring application KPI degradation. This is an opportunity to simplify operations as engineers can rely solely on platform metrics -- while still fulfilling application KPIs -- to configure portable and application agnostic rules and other cloud-specific parameters to automatically trigger actions such as autoscaling, instance migration, network slicing, etc. Results show that monitorless infers KPI degradation with an accuracy of 97% and, notably, it performs similarly to typical autoscaling solutions, even when autoscaling rules are optimally tuned with knowledge of the expected workload.

Many problems in areas as diverse as recommendation systems, social network analysis, semantic search, and distributed root cause analysis can be modeled as pattern search on labeled graphs (also called "heterogeneous information networks" or HINs). Given a large graph and a query pattern with node and edge label constraints, a fundamental challenge is to find the top-k matches according to a ranking function over edge and node weights. For users, it is difficult to select value k. We therefore propose the novel notion of an any-k ranking algorithm: for a given time budget, return as many of the top-ranked results as possible. Then, given additional time, produce the next lower-ranked results quickly as well. It can be stopped anytime, but may have to continue until all results are returned. This paper focuses on acyclic patterns over arbitrary labeled graphs. We are interested in practical algorithms that effectively exploit (1) properties of heterogeneous networks, in particular selective constraints on labels, and (2) that the users often explore only a fraction of the top-ranked results. Our solution, KARPET, carefully integrates aggressive pruning that leverages the acyclic nature of the query, and incremental guided search. It enables us to prove strong non-trivial time and space guarantees, which is generally considered very hard for this type of graph search problem. Through experimental studies we show that KARPET achieves running times in the order of milliseconds for tree patterns on large networks with millions of nodes and edges.

An increasing number of applications are modeled and analyzed in network form, where nodes represent entities of interest and edges represent interactions or relationships between entities. Commonly, such relationship analysis tools assume homogeneity in both node type and edge type. Recent research has sought to redress the assumption of homogeneity and focused on mining heterogeneous information networks (HINs) where both nodes and edges can be of different types. Building on such efforts, in this work, we articulate a novel approach for mining relationships across entities in such networks while accounting for user preference over relationship type and interestingness metric. We formalize the problem as a top-k lightest paths problem, contextualized in a real-world communication network, and seek to find the k most interesting path instances matching the preferred relationship type. Our solution, PROphetic HEuristic Algorithm for Path Searching (PRO-HEAPS), leverages a combination of novel graph preprocessing techniques, well-designed heuristics and the venerable A* search algorithm. We run our algorithm on real-world large-scale graphs and show that our algorithm significantly outperforms a wide variety of baseline approaches with speedups as large as 100X. To widen the range of applications, we also extend PRO-HEAPS to (i) support relationship analysis between two groups of entities and (ii) allow pattern path in the query to contain logical statements with operators AND, OR, NOT, and wild-card “.”. We run experiments using this generalized version of PRO-HEAPS and demonstrate that the advantage of PRO-HEAPS becomes even more pronounced for these general cases. Furthermore, we conduct a comprehensive analysis to study how the performance of PRO-HEAPS varies with respect to various attributes of the input HIN. We finally conduct a case study to demonstrate valuable applications of our algorithm.

In the bitprobe model the primary concern is the number of bits that must be probed in a data structure in order to answer queries, and how constraining the number probes affects the storage requirements. In this note we revisit an explicit membership scheme of Radhakrishnan, Raman, and Rao [ESA 2001], which stores an n=2 element subset of a universe [1,m], occupies O(m2/3) bits, and supports membership queries in t=2 bit probes. We show that, with a small modification to their storage scheme, their query scheme also works for n=3 elements.

The Second Workshop on Knowledge Graphs and Semantics for Text Retrieval, Analysis, and Understanding (KG4IR), Michigan, USA, 12 July 2018

The rising need to harvest domain specific knowledge in several applications is largely limited by the ability to dynamically grow structured knowledge representations, due to the increasing emergence of new concepts and their semantic relationships with existing ones. Such enrichment of existing hierarchical knowledge sources with new information to better model the "changing world" presents two-fold challenges: (1) Detection of previously unknown entities or concepts, and (2) Insertion of the new concepts into the knowledge structure, respecting the semantic integrity of the created relationships. To this end we propose a novel framework, ETF, to enrich large-scale, generic taxonomies with new concepts from resources such as news and research publications. Our approach learns a high-dimensional embedding for the existing concepts of the taxonomy, as well as for the new concepts. During the insertion of a new concept, this embedding is used to identify semantically similar neighborhoods within the existing taxonomy. The potential parent-child relationships linking the new concepts to the existing ones are then predicted using a set of semantic and graph features. Extensive evaluation of ETF on large, real-world taxonomies of Wikipedia and WordNet showcase more than 5% F1-score improvements compared to state-of-the-art baselines. We further demonstrate that ETF can accurately categorize newly emerging concepts and question-answer pairs across different domains.

Current practices for provisioning and autoscaling applications in cloud are limited and require expert knowledge to deliver acceptable quality of service (QoS). The complexity of the cloud stack requires an iterative process to tune several parameters with robustness to varying workloads, software or hardware upgrades, failures, resource interference, etc. Usually, this process is performed by humans, which limits the agility to adapt to stack changes. The need for human expertise relates to the fact that different applications can have different and highly specific key performance indicators (KPI’s). Experts can configure provisioning and autoscaling actions based on application KPIs but that process needs to be repeated for the various combinations of applications, services, and platforms. That is, the KPI-based rules for one platform do not necessarily apply to other environments given the multiplicity of variables that can affect such rules. Recently, several proposals have included machine learning to remove some of the complexity and human intervention. Still, these techniques require significant amounts of data (e.g., in case of techniques based on deep learning), or a considerable number of iterations to converge to the appropriate rules (e.g., in case of techniques based on reinforcement learning). These limitations can be prohibitive in cloud because 1) the amount of data required can be difficult to obtain, particularly in multi-vendor environments with management policies and specific infrastructure conditions that limit data access, and 2) the number of parameters (or dimensions) to explore can be be unfeasible, even for simple applications. To overcome these problems, we propose a solution that differentiate from others in two major aspects. First, our proposal is generic and portable to different environments, and robust to changes in the stack. We achieve this by training a model that (indirectly) requires application specific KPIs one time: during the training phase. However, when deployed, the model only requires platform utilization metrics that are common to all infrastructures (e.g., CPU, Memory, IO, etc.). Second, we reduce the data requirements and exploration space by using apprenticeship learning (also called learning from demonstration) which makes use of an expert

With the advent of Big Data paradigm, filtering, retrieval, and linking of unstructured multi-modal data has become a necessity. Assigning topic labels to contents, that accurately capture the meaning and contextual information, is a fundamental problem in organizing unstructured data. The usage of manually-assigned tags for this purpose introduces inconsistencies because of different "surface forms". On the other hand, existing automated approaches either use hierarchical multi-label classification, or are unsupervised and rely on (undirected) graph measures leveraging taxonomies. While the former requires large training data set to learn the characteristics of each topic class, the latter lacks the flexibility to learn broad range of related topics and are less accurate.We propose a novel framework, ANNOTATE based on a small set of features and directed traversal of taxonomies to learn a broad spectrum of related topics using limited training data. We also show that our approach provides accurate labels for several domains without the need for re-training. For instance, the framework, trained on a small set of BBC news articles, exhibits close matches to user-generated tags for Quora documents. Experimental results, on the same model, for news classification and identifying aspects of Amazon product reviews, based on Amazon Mechanical Turk evaluation show our approach to be significantly better than state-of-the-art.We further present real-life case studies of our proposed framework for automatically tagging Quora posts, and topically segmenting, indexing and linking related YouTube videos (using our publicly available Chrome browser extension).

Many problems in areas as diverse as recommendation systems, social network analysis, semantic search and distributed root cause analysis can be modeled as graph query patterns on heterogeneous information networks. A fundamental challenge is, given a large network and a graph query pattern with node and edge label constraints, to find the top-k matches according to a monotone ranking function over edge and node weights. We focus on sum-of-weights for acyclic queries over arbitrary graphs. Furthermore, we consider the more challenging scenario where k is not given in advance, but the user interactively requests matching patterns in order of ranking. Sub-graph isomorphism is hard in general, even for path queries. We are therefore interested in practical algorithms that effectively exploit (1) properties of heterogeneous networks, in particular highly selective constraints on nodes and edges, and (2) that the user will often explore only a fraction of the top-ranked results. Our solution, {ouralgorithmname}, carefully integrates aggressive pruning that leverages the acyclic nature of the query, and incremental guided search. It enables us to prove strong non-trivial time and space guarantees, which is generally considered very hard for this type of non-trivial graph search problem. Through experimental studies we show that {ouralgorithmname} achieves running times in the order of milliseconds for tree patterns on large networks with millions of nodes and edges.

Set intersection counting appears as a subroutine in many techniques used in natural language processing, in which similarity is often measured as a function of document cooccurence counts between pairs of noun phrases or entities. Such techniques include clustering of text phrases and named entities, topic labeling, entity disambiguation, sentiment analysis, and search for synonyms.These techniques can have real-time constraints that require very fast computation of thousands of set intersection counting queries with little space overhead and minimal error. On one hand, while sketching techniques for approximate intersection counting exist and have very fast query time, many have issues with accuracy, especially for pairs of lists that have low Jaccard similarity. On the other hand, space-efficient computation of exact intersection sizes is particularly challenging in real-time.In this paper, we show how an efficient space-time trade-off can be achieved for exact set intersection counting, by combining state-of-the-art algorithms with precomputation and judicious use of compression. In addition, we show that the performance can be further improved by combining the best aspects of these algorithms. We present experimental evidence that realtime computation of exact intersection sizes is feasible with low memory overhead: we improve the mean query time of baseline approaches by over a factor of 100 using a data structure that takes merely twice the size of an inverted index. Overall, in our experiments, we achieve running times within the same order of magnitude as well-known approximation techniques.