Query Arrives Research Articles

Adwords with Unknown Budgets and Beyond

In the classic Adwords problem introduced by [Mehta A, Saberi A, Vazirani U, Vazirani V (2007) Adwords and generalized online matching. J. ACM 54(5):22-es.], we have a bipartite graph between advertisers and queries. Each advertiser has a maximum budget that is known a priori. Queries are unknown a priori and arrive sequentially. When a query arrives, advertisers make bids, and we (immediately and irrevocably) decide which (if any) Ad to display based on the bids and advertiser budgets. The winning advertiser for each query pays their bid up to their remaining budget. Our goal is to maximize total budget used without any foreknowledge of the arrival sequence (which could be adversarial). We consider the setting where the online algorithm does not know the advertisers’ budgets a priori and the budget of an advertiser is revealed to the algorithm only when it is exceeded. A naïve greedy algorithm is 0.5 competitive for this setting, and finding an algorithm with better performance remained an open problem. We show that no deterministic algorithm has competitive ratio better than 0.5 and give the first (randomized) algorithm with strictly better performance guarantee. We show that the competitive ratio of our algorithm is at least 0.522 but also strictly less than [Formula: see text]. We present novel applications of budget oblivious algorithms in search ads and beyond. In particular, we show that our algorithm achieves the best possible performance guarantee for deterministic online matching in the presence of multichannel traffic [Manshadi V, Rodilitz S, Saban D, Suresh A (2022) Online algorithms for matching platforms with multi-channel traffic. Proc. 23rd ACMConf. Econom. Comput. (ACM, NewYork), 986–987.]. This paper was accepted by Omar Besbes, revenue management and market analytics. Funding: NSF Division of Civil, Mechanical, and Manufacturing Innovation [Grant 2340306], and Google Research Scholar Program. Supplemental Material: The online appendices are available at https://doi.org/10.1287/mnsc.2021.03243 .

Fast Online Hashing with Multi-Label Projection

Hashing has been widely researched to solve the large-scale approximate nearest neighbor search problem owing to its time and storage superiority. In recent years, a number of online hashing methods have emerged, which can update the hash functions to adapt to the new stream data and realize dynamic retrieval. However, existing online hashing methods are required to update the whole database with the latest hash functions when a query arrives, which leads to low retrieval efficiency with the continuous increase of the stream data. On the other hand, these methods ignore the supervision relationship among the examples, especially in the multi-label case. In this paper, we propose a novel Fast Online Hashing (FOH) method which only updates the binary codes of a small part of the database. To be specific, we first build a query pool in which the nearest neighbors of each central point are recorded. When a new query arrives, only the binary codes of the corresponding potential neighbors are updated. In addition, we create a similarity matrix which takes the multi-label supervision information into account and bring in the multi-label projection loss to further preserve the similarity among the multi-label data. The experimental results on two common benchmarks show that the proposed FOH can achieve dramatic superiority on query time up to 6.28 seconds less than state-of-the-art baselines with competitive retrieval accuracy.

Near-Duplicate Sequence Search at Scale for Large Language Model Memorization Evaluation

Recent studies show that large language models (LLM) unintendedly memorize part of the training data, which brings serious privacy risks. For example, it has been shown that over 1% of tokens generated unprompted by an LLM are part of sequences in the training data. However, current studies mainly focus on the exact memorization behaviors. In this paper, we propose to evaluate how many generated texts have near-duplicates (e.g., only differ by a couple of tokens out of 100) in the training corpus. A major challenge of conducting this evaluation is the huge computation cost incurred by near-duplicate sequence searches. This is because modern LLMs are trained on larger and larger corpora with up to 1 trillion tokens. What's worse is that the number of sequences in a text is quadratic to the text length. To address this issue, we develop an efficient and scalable near-duplicate sequence search algorithm in this paper. It can find (almost) all the near-duplicate sequences of the query sequence in a large corpus with guarantees. Specifically, the algorithm generates and groups the min-hash values of all the sequences with at least t tokens (as very short near-duplicates are often irrelevant noise) in the corpus in linear time to the corpus size. We formally prove that only 2 n+1/t+1 -1 min-hash values are generated for a text with n tokens in expectation. Thus the index time and size are reasonable. When a query arrives, we find all the sequences sharing enough min-hash values with the query using inverted indexes and prefix filtering. Extensive experiments on a few large real-world LLM training corpora show that our near-duplicate sequence search algorithm is efficient and scalable.

DFS is Unsparsable and Lookahead Can Help in Maximal Matching

In this paper we study two problems in the context of fully dynamic graph algorithms that is, when we have to handle updates (insertions and removals of edges), and answer queries regarding the current graph, preferably with a better time bound than that when running a classical algorithm from scratch each time a query arrives. In the first part we show that there are dense (directed) graphs having no nontrivial strong certificates for maintaining a depth-first search tree, hence the so-called sparsification technique cannot be applied effectively to this problem. In the second part, we show that a maximal matching can be maintained in an (undirected) graph with a deterministic amortized update cost of O(log m) (where m is the all-time maximum number of the edges), provided that a lookahead of length m is available, i.e. we can “take a peek” at the next m update operations in advance.

Query-aware locality-sensitive hashing scheme for $$l_p$$ norm

The problem of c-Approximate Nearest Neighbor (c-ANN) search in high-dimensional space is fundamentally important in many applications, such as image database and data mining. Locality-Sensitive Hashing (LSH) and its variants are the well-known indexing schemes to tackle the c-ANN search problem. Traditionally, LSH functions are constructed in a query-oblivious manner, in the sense that buckets are partitioned before any query arrives. However, objects closer to a query may be partitioned into different buckets, which is undesirable. Due to the use of query-oblivious bucket partition, the state-of-the-art LSH schemes for external memory, namely C2LSH and LSB-Forest, only work with approximation ratio of integer $$c \ge 2$$cź2. In this paper, we introduce a novel concept of query-aware bucket partition which uses a given query as the "anchor" for bucket partition. Accordingly, a query-aware LSH function under a specific $$l_p$$lp norm with $$p \in (0,2]$$pź(0,2] is a random projection coupled with query-aware bucket partition, which removes random shift required by traditional query-oblivious LSH functions. The query-aware bucket partitioning strategy can be easily implemented so that query performance is guaranteed. For each $$l_p$$lp norm $$(p \in (0,2])$$(pź(0,2]), based on the corresponding p-stable distribution, we propose a novel LSH scheme named query-aware LSH (QALSH) for c-ANN search over external memory. Our theoretical studies show that QALSH enjoys a guarantee on query quality. The use of query-aware LSH function enables QALSH to work with any approximation ratio $$c > 1$$c>1. In addition, we propose a heuristic variant named QALSH$$^+$$+ to improve the scalability of QALSH. Extensive experiments show that QALSH and QALSH$$^+$$+ outperform the state-of-the-art schemes, especially in high-dimensional space. Specifically, by using a ratio $$c < 2$$c<2, QALSH can achieve much better query quality.

Query-aware locality-sensitive hashing for approximate nearest neighbor search

Locality-Sensitive Hashing (LSH) and its variants are the well-known indexing schemes for the c -Approximate Nearest Neighbor ( c -ANN) search problem in high-dimensional Euclidean space. Traditionally, LSH functions are constructed in a query-oblivious manner in the sense that buckets are partitioned before any query arrives. However, objects closer to a query may be partitioned into different buckets, which is undesirable. Due to the use of query-oblivious bucket partition, the state-of-the-art LSH schemes for external memory, namely C2LSH and LSB-Forest, only work with approximation ratio of integer c ≥ 2. In this paper, we introduce a novel concept of query-aware bucket partition which uses a given query as the "anchor" for bucket partition. Accordingly, a query-aware LSH function is a random projection coupled with query-aware bucket partition, which removes random shift required by traditional query-oblivious LSH functions. Notably, query-aware bucket partition can be easily implemented so that query performance is guaranteed. We propose a novel query-aware LSH scheme named QALSH for c -ANN search over external memory. Our theoretical studies show that QALSH enjoys a guarantee on query quality. The use of query-aware LSH function enables QALSH to work with any approximation ratio c &gt; 1. Extensive experiments show that QALSH outperforms C2LSH and LSB-Forest, especially in high-dimensional space. Specifically, by using a ratio c &lt; 2, QALSH can achieve much better query quality.

Improve the diagnosis of atrial hypertrophy with the local discriminative support vector machine.

Computer-aided diagnosis (CAD) approaches succeed in detecting a number of diseases, however, they are not good at addressing atrial hypertrophy disease due to the lack of training data. Support Vector Machine (SVM) is very popular in few CAD solutions to atrial hypertrophy. Yet the performance of SVM is moderate in atrial hypertrophy detection compared to its success in other classification problems. In this paper we propose a novel CAD algorithm, Local Discriminative SVM (LDSVM), to overcome the above-mentioned difficulty. LDSVM consists of a global SVM that is trained on the training data, and a local kNN that is trained based on the information of SVM and query. When a query arrives, SVM gives the initial decision. If the initial decision is less confident, local kNN begins to modify the initial decision. LDSVM improves the accuracy of SVM through some contributions: the risk tube, the discriminant information derived from SVM hyperplane, the new metric and the self-tuning size of neighborhood. Empirical evidence on real medical datasets show high performance of LDSVM over the peers in addressing atrial hypertrophy problems.

Develop Local Fuzzy Classifier to Modify Low-Confidence Output of Global Classifier

This paper proposes an assembling classifier consisting of a global classifier and a local classifier, named as GCLC. To this end, we present a weighted Support Vector Machine (wSVM) that serves as the global classifier, and a fuzzy k-nearest neighbor (fkNN) that serves as the local one. When a query arrives, wSVM labels it firstly. If the global decision is below some threshold, the local fkNN works to provide an improved decision. Extensive experiments on real datasets demonstrate the performance of GCLC compared with the state of the art.

Online Advertisement, Optimization and Stochastic Networks

In this paper, we propose a stochastic model to describe how search service providers charge client companies based on users' queries for the keywords related to these companies' ads by using certain advertisement assignment strategies. We formulate an optimization problem to maximize the long-term average revenue for the service provider under each client's long-term average budget constraint and design an online algorithm which captures the stochastic properties of users' queries and click-through behaviors. We solve the optimization problem by making connections to scheduling problems in wireless networks, queueing theory and stochastic networks. Unlike prior models, we do not assume that the number of query arrivals is known. Due to the stochastic nature of the arrival process considered here, either temporary “free” service, i.e., service above the specified budget (which we call “overdraft”) or under-utilization of the budget (which we call “underdraft”) is unavoidable. We prove that our online algorithm can achieve a revenue that is within <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">O</i> (ε) of the optimal revenue while ensuring that the overdraft or underdraft is <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">O</i> (1/ε) , where ε can be arbitrarily small. With a view towards practice, we can show that one can always operate strictly under the budget. In addition, we extend our results to a click-through rate maximization model and also show how our algorithm can be modified to handle non-stationary query arrival processes and clients with short-term contracts. Our algorithm also allows us to quantify the effect of errors in click-through rate estimation on the achieved revenue. We show that we lose at most Δ/1+Δ fraction of the revenue if Δ is the relative error in click-through rate estimation. We also show that in the long run, an expected overdraft level of Ω(log(1/ε)) is unavoidable (a universal lower bound) under any stationary ad assignment algorithm which achieves a long-term average revenue within <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">O</i> (ε) of the offline optimum.

Scheduling Periodic Continuous Queries in Real-Time Data Broadcast Environments

On-demand broadcast is a promising data dissemination approach in mobile computing environments thanks to its adaptability and scalability for large-scale and dynamic workload. An important class of emerging data broadcast applications needs to monitor multiple time-varying data items continuously to be kept aware of the up-to-date information. This paper investigates the broadcast schedule problem for disseminating timely data to periodic continuous queries, and a systematic and highly efficient solution for applications of this type is provided. In particular, we propose a novel measure, called Bandwidth Utilization, to quantify the minimum bandwidth demand of a periodic continuous query set. The timing predictability can be ensured if a set of periodic continuous queries passes a bandwidth utilization based schedulability test. The schedulability test techniques are also extended to deal with dynamic query arrival and departure. An efficient online scheduling algorithm, called RM-UO, is developed, which can fulfill the timing constraints combined with the proposed query release and deletion policies. To demonstrate the effectiveness of theoretical results, an illustrative algorithm implementation is presented along with comprehensive performance analysis. Simulation results show that our solution offers nice timing predictability whereas other comparable best effort scheduling algorithms such as SIN-Q and DTIU experience different deadline miss ratios at different query workloads.

An Autonomous Agent Approach to Query Optimization in Stream Grids

Stream grids are wide-area grid computing environments that are fed by a set of stream data sources, and Queries arrive at the grid from users and applications external to the system. The kind of queries considered in this work is long-running continuous (LRC) queries, which are neither short-lived nor infinitely long lived. The queries are “open” from the grid perspective as the grid cannot control or predict the arrival of a query with time, location, required data and query revocations. Query optimization in such an environment has two major challenges, i.e., optimizing in a multi-query environment and continuous optimization, due to new query arrivals and revocations. As generating a globally optimal query plan is an intractable problem, this work explores the idea of emergent optimization where globally optimal query plans emerge as a result of local autonomous decisions taken by the grid nodes. Drawing concepts from evolutionary game theory, grid nodes are modeled as autonomous agents that seek to maximize a self-interest function using one of a set of different strategies. Grid nodes change strategies in response to variations in query arrival and revocation patterns, which is also autonomously decided by each grid node.

Queueing system with a phase process of query arrival in session

A multi-server quenching system with sessional query arrival is considered. The query arrival in each session is described by the considerably Phase Type (PH) arrival flow. The session acceptance in the system is limited with the help of tokens. Sessions that come into the system at the moment when free tokens are lacking are rejected. An algorithm for finding the joint distribution of the number of sessions and the number of queries in the system is provided. Several problems of the system’s parameters optimization are considered.

Method of adaptive controllable parallel execution of operations in object database management systems

The design of parallel database management systems (DBMSs) normally implies using special-purpose multiprocessor computing systems. Most often, a DBMS is supposed to work in an exclusive mode of operation. However, in the class of x86-based multiprocessor computing systems designed for mass usage, the exclusive mode of DBMS operation with respect to other software is often not secured. In addition, the legacy software for this class of computing systems is often not designed for mass parallel usage. When the exclusive mode requirement is ignored and the resources of the computing system are not used in low-load DBMS regimes, the efficiency of using resources of the computing system as a whole reduces. This paper considers a method of program organization of controlled parallelism at the level of internal DBMS operations, allowing for their controlled execution based on the state of the entire computing system. This method made it possible to significantly reduce the time of response for low densities of query arrival in the ODB-Jupiter commercial object DBMS developed in the Inteltec Plus scientific manufacturing center. The method of controlled parallel execution can be used in a wide class of program systems.

Online Data Gathering for Maximizing Network Lifetime in Sensor Networks

Energy-constrained sensor networks have been deployed widely for monitoring and surveillance purposes. Data gathering in such networks is often a prevalent operation. Since sensors have significant power constraints (battery life), energy efficient methods must be employed for data gathering to prolong network lifetime. We consider an online data gathering problem in sensor networks, which is stated as follows: assume that there is a sequence of data gathering queries, which arrive one by one. To respond to each query as it arrives, the system builds a routing tree for it. Within the tree, the volume of the data transmitted by each internal node depends on not only the volume of sensed data by the node itself, but also the volume of data received from its children. The objective is to maximize the network lifetime without any knowledge of future query arrivals and generation rates. In other words, the objective is to maximize the number of data gathering queries answered until the first node in the network fails. For the problem of concern, in this paper, we first present a generic cost model of energy consumption for data gathering queries if a routing tree is used for the query evaluation. We then show the problem to be NP-complete and propose several heuristic algorithms for it. We finally conduct experiments by simulation to evaluate the performance of the proposed algorithms in terms of network lifetime delivered. The experimental results show that, among the proposed algorithms, one algorithm that takes into account both the residual energy and the volume of data at each sensor node significantly outperforms the others