Search

Language models (LMs) exhibit impressive performance and generalization capabilities. However, LMs struggle with the persistent challenge of catastrophic forgetting, which undermines their long-term sustainability in continual learning (CL). Existing approaches usually address the issue by incorporating old task data or task-wise inductive bias into LMs. However, old data and accurate task information are often unavailable or costly to collect, hindering the availability of current CL approaches for LMs. To address this limitation, we introduce $\textbf{MIGU}$ ($\textbf{M}$agn$\textbf{I}$tude-based $\textbf{G}$radient $\textbf{U}$pdating for continual learning), a rehearsal-free and task-label-free method that only updates the model parameters with large magnitudes of output in LMs' linear layers. MIGU is based on our observation that the L1-normalized magnitude distribution of the output in LMs' linear layers is different when the LM models deal with different task data. By imposing this simple constraint on the gradient update process, we can leverage the inherent behaviors of LMs, thereby unlocking their innate CL abilities. Our experiments demonstrate that MIGU is universally applicable to all three LM architectures (T5, RoBERTa, and Llama2), delivering state-of-the-art or on-par performance across continual finetuning and continual pre-training settings on four CL benchmarks. For example, MIGU brings a 15.2% average accuracy improvement over conventional parameter-efficient finetuning baselines in a 15-task CL benchmark. MIGU can also seamlessly integrate with all three existing CL types to further enhance performance. Code is available at \href{https://github.com/wenyudu/MIGU}{this https URL}., Comment: preprint, 19 pages

LLMs are computationally expensive to pre-train due to their large scale. Model growth emerges as a promising approach by leveraging smaller models to accelerate the training of larger ones. However, the viability of these model growth methods in efficient LLM pre-training remains underexplored. This work identifies three critical $\underline{\textit{O}}$bstacles: ($\textit{O}$1) lack of comprehensive evaluation, ($\textit{O}$2) untested viability for scaling, and ($\textit{O}$3) lack of empirical guidelines. To tackle $\textit{O}$1, we summarize existing approaches into four atomic growth operators and systematically evaluate them in a standardized LLM pre-training setting. Our findings reveal that a depthwise stacking operator, called $G_{\text{stack}}$, exhibits remarkable acceleration in training, leading to decreased loss and improved overall performance on eight standard NLP benchmarks compared to strong baselines. Motivated by these promising results, we conduct extensive experiments to delve deeper into $G_{\text{stack}}$ to address $\textit{O}$2 and $\textit{O}$3. For $\textit{O}$2 (untested scalability), our study shows that $G_{\text{stack}}$ is scalable and consistently performs well, with experiments up to 7B LLMs after growth and pre-training LLMs with 750B tokens. For example, compared to a conventionally trained 7B model using 300B tokens, our $G_{\text{stack}}$ model converges to the same loss with 194B tokens, resulting in a 54.6\% speedup. We further address $\textit{O}$3 (lack of empirical guidelines) by formalizing guidelines to determine growth timing and growth factor for $G_{\text{stack}}$, making it practical in general LLM pre-training. We also provide in-depth discussions and comprehensive ablation studies of $G_{\text{stack}}$. Our code and pre-trained model are available at $\href{https://llm-stacking.github.io/}{https://llm-stacking.github.io/}$., Comment: Preprint; The project link: $\href{https://llm-stacking.github.io/}{https://llm-stacking.github.io/}$

Large Language Models (LLMs) driven by In-Context Learning (ICL) have significantly improved the performance of text-to-SQL. Previous methods generally employ a two-stage reasoning framework, namely 1) schema linking and 2) logical synthesis, making the framework not only effective but also interpretable. Despite these advancements, the inherent bad nature of the generalization of LLMs often results in hallucinations, which limits the full potential of LLMs. In this work, we first identify and categorize the common types of hallucinations at each stage in text-to-SQL. We then introduce a novel strategy, Task Alignment (TA), designed to mitigate hallucinations at each stage. TA encourages LLMs to take advantage of experiences from similar tasks rather than starting the tasks from scratch. This can help LLMs reduce the burden of generalization, thereby mitigating hallucinations effectively. We further propose TA-SQL, a text-to-SQL framework based on this strategy. The experimental results and comprehensive analysis demonstrate the effectiveness and robustness of our framework. Specifically, it enhances the performance of the GPT-4 baseline by 21.23% relatively on BIRD dev and it yields significant improvements across six models and four mainstream, complex text-to-SQL benchmarks., Comment: Accepted to ACL Findings 2024

Current graph clustering methods emphasize individual node and edge con nections, while ignoring higher-order organization at the level of motif. Re cently, higher-order graph clustering approaches have been designed by motif based hypergraphs. However, these approaches often suffer from hypergraph fragmentation issue seriously, which degrades the clustering performance greatly. Moreover, real-world graphs usually contain diverse motifs, with nodes participating in multiple motifs. A key challenge is how to achieve precise clustering results by integrating information from multiple motifs at the node level. In this paper, we propose a multi-order graph clustering model (MOGC) to integrate multiple higher-order structures and edge connections at node level. MOGC employs an adaptive weight learning mechanism to au tomatically adjust the contributions of different motifs for each node. This not only tackles hypergraph fragmentation issue but enhances clustering accuracy. MOGC is efficiently solved by an alternating minimization algo rithm. Experiments on seven real-world datasets illustrate the effectiveness of MOGC.

Hypothesis testing is a statistical method used to draw conclusions about populations from sample data, typically represented in tables. With the prevalence of graph representations in real-life applications, hypothesis testing in graphs is gaining importance. In this work, we formalize node, edge, and path hypotheses in attributed graphs. We develop a sampling-based hypothesis testing framework, which can accommodate existing hypothesis-agnostic graph sampling methods. To achieve accurate and efficient sampling, we then propose a Path-Hypothesis-Aware SamplEr, PHASE, an m- dimensional random walk that accounts for the paths specified in a hypothesis. We further optimize its time efficiency and propose PHASEopt. Experiments on real datasets demonstrate the ability of our framework to leverage common graph sampling methods for hypothesis testing, and the superiority of hypothesis-aware sampling in terms of accuracy and time efficiency.

Interactive Data Analysis, the collaboration between humans and LLM agents, enables real-time data exploration for informed decision-making. The challenges and costs of collecting realistic interactive logs for data analysis hinder the quantitative evaluation of Large Language Model (LLM) agents in this task. To mitigate this issue, we introduce Tapilot-Crossing, a new benchmark to evaluate LLM agents on interactive data analysis. Tapilot-Crossing contains 1024 interactions, covering 4 practical scenarios: Normal, Action, Private, and Private Action. Notably, Tapilot-Crossing is constructed by an economical multi-agent environment, Decision Company, with few human efforts. We evaluate popular and advanced LLM agents in Tapilot-Crossing, which underscores the challenges of interactive data analysis. Furthermore, we propose Adaptive Interaction Reflection (AIR), a self-generated reflection strategy that guides LLM agents to learn from successful history. Experiments demonstrate that Air can evolve LLMs into effective interactive data analysis agents, achieving a relative performance improvement of up to 44.5%., Comment: 30 pages, 7 figures

In the era of Large Language Models (LLMs), Knowledge Distillation (KD) emerges as a pivotal methodology for transferring advanced capabilities from leading proprietary LLMs, such as GPT-4, to their open-source counterparts like LLaMA and Mistral. Additionally, as open-source LLMs flourish, KD plays a crucial role in both compressing these models, and facilitating their self-improvement by employing themselves as teachers. This paper presents a comprehensive survey of KD's role within the realm of LLM, highlighting its critical function in imparting advanced knowledge to smaller models and its utility in model compression and self-improvement. Our survey is meticulously structured around three foundational pillars: \textit{algorithm}, \textit{skill}, and \textit{verticalization} -- providing a comprehensive examination of KD mechanisms, the enhancement of specific cognitive abilities, and their practical implications across diverse fields. Crucially, the survey navigates the intricate interplay between data augmentation (DA) and KD, illustrating how DA emerges as a powerful paradigm within the KD framework to bolster LLMs' performance. By leveraging DA to generate context-rich, skill-specific training data, KD transcends traditional boundaries, enabling open-source models to approximate the contextual adeptness, ethical alignment, and deep semantic insights characteristic of their proprietary counterparts. This work aims to provide an insightful guide for researchers and practitioners, offering a detailed overview of current methodologies in KD and proposing future research directions. Importantly, we firmly advocate for compliance with the legal terms that regulate the use of LLMs, ensuring ethical and lawful application of KD of LLMs. An associated Github repository is available at https://github.com/Tebmer/Awesome-Knowledge-Distillation-of-LLMs., Comment: 44 pages

Global popularity (GP) bias is the phenomenon that popular items are recommended much more frequently than they should be, which goes against the goal of providing personalized recommendations and harms user experience and recommendation accuracy. Many methods have been proposed to reduce GP bias but they fail to notice the fundamental problem of GP, i.e., it considers popularity from a \textit{global} perspective of \textit{all users} and uses a single set of popular items, and thus cannot capture the interests of individual users. As such, we propose a user-aware version of item popularity named \textit{personal popularity} (PP), which identifies different popular items for each user by considering the users that share similar interests. As PP models the preferences of individual users, it naturally helps to produce personalized recommendations and mitigate GP bias. To integrate PP into recommendation, we design a general \textit{personal popularity aware counterfactual} (PPAC) framework, which adapts easily to existing recommendation models. In particular, PPAC recognizes that PP and GP have both direct and indirect effects on recommendations and controls direct effects with counterfactual inference techniques for unbiased recommendations. All codes and datasets are available at \url{https://github.com/Stevenn9981/PPAC}., Comment: Accepted by WWW'24 as a research full paper

Transportation companies and organizations routinely collect huge volumes of passenger transportation data. By aggregating these data (e.g., counting the number of passengers going from a place to another in every 30 minute interval), it becomes possible to analyze the movement behavior of passengers in a metropolitan area. In this paper, we study the problem of finding important trends in passenger movements at varying granularities, which is useful in a wide range of applications such as target marketing, scheduling, and travel intent prediction. Specifically, we study the extraction of movement patterns between regions that have significant flow. The huge number of possible patterns render their detection computationally hard. We propose algorithms that greatly reduce the search space and the computational cost of pattern detection. We study variants of patterns that could be useful to different problem instances, such as constrained patterns and top-k ranked patterns.

Multi-domain recommendation (MDR) aims to provide recommendations for different domains (e.g., types of products) with overlapping users/items and is common for platforms such as Amazon, Facebook, and LinkedIn that host multiple services. Existing MDR models face two challenges: First, it is difficult to disentangle knowledge that generalizes across domains (e.g., a user likes cheap items) and knowledge specific to a single domain (e.g., a user likes blue clothing but not blue cars). Second, they have limited ability to transfer knowledge across domains with small overlaps. We propose a new MDR method named EDDA with two key components, i.e., embedding disentangling recommender and domain alignment, to tackle the two challenges respectively. In particular, the embedding disentangling recommender separates both the model and embedding for the inter-domain part and the intra-domain part, while most existing MDR methods only focus on model-level disentangling. The domain alignment leverages random walks from graph processing to identify similar user/item pairs from different domains and encourages similar user/item pairs to have similar embeddings, enhancing knowledge transfer. We compare EDDA with 12 state-of-the-art baselines on 3 real datasets. The results show that EDDA consistently outperforms the baselines on all datasets and domains. All datasets and codes are available at https://github.com/Stevenn9981/EDDA., Comment: Accepted by CIKM'23 as a Long paper

Mobility data captures the locations of moving objects such as humans, animals, and cars. With the availability of GPS-equipped mobile devices and other inexpensive location-tracking technologies, mobility data is collected ubiquitously. In recent years, the use of mobility data has demonstrated significant impact in various domains including traffic management, urban planning, and health sciences. In this paper, we present the emerging domain of mobility data science. Towards a unified approach to mobility data science, we envision a pipeline having the following components: mobility data collection, cleaning, analysis, management, and privacy. For each of these components, we explain how mobility data science differs from general data science, we survey the current state of the art and describe open challenges for the research community in the coming years., Comment: Updated to reflect the major revision for ACM Transactions on Spatial Algorithms and Systems (TSAS). This version reflects the final version accepted by ACM TSAS

Text-to-SQL parsing, which aims at converting natural language instructions into executable SQLs, has gained increasing attention in recent years. In particular, Codex and ChatGPT have shown impressive results in this task. However, most of the prevalent benchmarks, i.e., Spider, and WikiSQL, focus on database schema with few rows of database contents leaving the gap between academic study and real-world applications. To mitigate this gap, we present Bird, a big benchmark for large-scale database grounded in text-to-SQL tasks, containing 12,751 pairs of text-to-SQL data and 95 databases with a total size of 33.4 GB, spanning 37 professional domains. Our emphasis on database values highlights the new challenges of dirty database contents, external knowledge between NL questions and database contents, and SQL efficiency, particularly in the context of massive databases. To solve these problems, text-to-SQL models must feature database value comprehension in addition to semantic parsing. The experimental results demonstrate the significance of database values in generating accurate text-to-SQLs for big databases. Furthermore, even the most effective text-to-SQL models, i.e. ChatGPT, only achieves 40.08% in execution accuracy, which is still far from the human result of 92.96%, proving that challenges still stand. Besides, we also provide an efficiency analysis to offer insights into generating text-to-efficient-SQLs that are beneficial to industries. We believe that BIRD will contribute to advancing real-world applications of text-to-SQL research. The leaderboard and source code are available: https://bird-bench.github.io/., Comment: NeurIPS 2023

The task of text-to-SQL parsing, which aims at converting natural language questions into executable SQL queries, has garnered increasing attention in recent years, as it can assist end users in efficiently extracting vital information from databases without the need for technical background. One of the major challenges in text-to-SQL parsing is domain generalization, i.e., how to generalize well to unseen databases. Recently, the pre-trained text-to-text transformer model, namely T5, though not specialized for text-to-SQL parsing, has achieved state-of-the-art performance on standard benchmarks targeting domain generalization. In this work, we explore ways to further augment the pre-trained T5 model with specialized components for text-to-SQL parsing. Such components are expected to introduce structural inductive bias into text-to-SQL parsers thus improving model's capacity on (potentially multi-hop) reasoning, which is critical for generating structure-rich SQLs. To this end, we propose a new architecture GRAPHIX-T5, a mixed model with the standard pre-trained transformer model augmented by some specially-designed graph-aware layers. Extensive experiments and analysis demonstrate the effectiveness of GRAPHIX-T5 across four text-to-SQL benchmarks: SPIDER, SYN, REALISTIC and DK. GRAPHIX-T5 surpass all other T5-based parsers with a significant margin, achieving new state-of-the-art performance. Notably, GRAPHIX-T5-large reach performance superior to the original T5-large by 5.7% on exact match (EM) accuracy and 6.6% on execution accuracy (EX). This even outperforms the T5-3B by 1.2% on EM and 1.5% on EX., Comment: Accepted to AAAI 2023 main conference (oral)

A subgraph is constructed by using a subset of vertices and edges of a given graph. There exist many graph properties that are hereditary for subgraphs. Hence, researchers from different communities have paid a great deal of attention in studying numerous subgraph problems, on top of the ordinary graph problems. Many algorithms are proposed in studying subgraph problems, where one common approach is by extracting the patterns and structures of a given graph. Due to the complex structures of certain types of graphs and to improve overall performances of the existing frameworks, machine learning techniques have recently been employed in dealing with various subgraph problems. In this article, we present a comprehensive review on five well known subgraph problems that have been tackled by using machine learning methods. They are subgraph isomorphism (both counting and matching), maximum common subgraph, community detection and community search problems. We provide an outline of each proposed method, and examine its designs and performances. We also explore non-learning-based algorithms for each problem and a brief discussion is given. We then suggest some promising research directions in this area, hoping that relevant subgraph problems can be tackled by using a similar strategy. Since there is a huge growth in employing machine learning techniques in recent years, we believe that this survey will serve as a good reference point to relevant research communities., Comment: v1: 41 pages; v2: 40 pages ; v3: This version focuses on just subgraph problems (discussions on other classic graph problems can be found in the earlier versions)

Heterogeneous Information Networks (HINs) are labeled graphs that depict relationships among different types of entities (e.g., users, movies and directors). For HINs, meta-path-based recommenders (MPRs) utilize meta-paths (i.e., abstract paths consisting of node and link types) to predict user preference, and have attracted a lot of attention due to their explainability and performance. We observe that the performance of MPRs is highly sensitive to the meta-paths they use, but existing works manually select the meta-paths from many possible ones. Thus, to discover effective meta-paths automatically, we propose the Reinforcement learning-based Meta-path Selection (RMS) framework. Specifically, we define a vector encoding for meta-paths and design a policy network to extend meta-paths. The policy network is trained based on the results of downstream recommendation tasks and an early stopping approximation strategy is proposed to speed up training. RMS is a general model, and it can work with all existing MPRs. We also propose a new MPR called RMS-HRec, which uses an attention mechanism to aggregate information from the meta-paths. We conduct extensive experiments on real datasets. Compared with the manually selected meta-paths, the meta-paths identified by RMS consistently improve recommendation quality. Moreover, RMS-HRec outperforms state-of-the-art recommender systems by an average of 7% in hit ratio. The codes and datasets are available on https://github.com/Stevenn9981/RMS-HRec., Comment: This paper is accepted as a full research paper by CIKM 2022

Crowdsourcing can be used to determine a total order for an object set (e.g., the top-10 NBA players) based on crowd opinions. This ranking problem is often decomposed into a set of microtasks (e.g., pairwise comparisons). These microtasks are passed to a large number of workers and their answers are aggregated to infer the ranking. The number of microtasks depends on the budget allocated for the problem. Intuitively, the higher the number of microtask answers, the more accurate the ranking becomes. However, it is often hard to decide the budget required for an accurate ranking. We study how a ranking process can be terminated early, and yet achieve a high-quality ranking and great savings in the budget. We use statistical tools to estimate the quality of the ranking result at any stage of the crowdsourcing process and terminate the process as soon as the desired quality is achieved. Our proposed early-stopping module can be seamlessly integrated with most existing inference algorithms and task assignment methods. We conduct extensive experiments and show that our early-stopping module is better than other existing general stopping criteria. We also implement a prototype system to demonstrate the usability and effectiveness of our approach in practice.

In this paper, we propose a Deep Reinforcement Learning (RL) framework for task arrangement, which is a critical problem for the success of crowdsourcing platforms. Previous works conduct the personalized recommendation of tasks to workers via supervised learning methods. However, the majority of them only consider the benefit of either workers or requesters independently. In addition, they cannot handle the dynamic environment and may produce sub-optimal results. To address these issues, we utilize Deep Q-Network (DQN), an RL-based method combined with a neural network to estimate the expected long-term return of recommending a task. DQN inherently considers the immediate and future reward simultaneously and can be updated in real-time to deal with evolving data and dynamic changes. Furthermore, we design two DQNs that capture the benefit of both workers and requesters and maximize the profit of the platform. To learn value functions in DQN effectively, we also propose novel state representations, carefully design the computation of Q values, and predict transition probabilities and future states. Experiments on synthetic and real datasets demonstrate the superior performance of our framework.

Densest subgraph discovery (DSD) is a fundamental problem in graph mining. It has been studied for decades, and is widely used in various areas, including network science, biological analysis, and graph databases. Given a graph G, DSD aims to find a subgraph D of G with the highest density (e.g., the number of edges over the number of vertices in D). Because DSD is difficult to solve, we propose a new solution paradigm in this paper. Our main observation is that a densest subgraph can be accurately found through a k-core (a kind of dense subgraph of G), with theoretical guarantees. Based on this intuition, we develop efficient exact and approximation solutions for DSD. Moreover, our solutions are able to find the densest subgraphs for a wide range of graph density definitions, including clique-based and general pattern-based density. We have performed extensive experimental evaluation on eleven real datasets. Our results show that our algorithms are up to four orders of magnitude faster than existing approaches.

With the rapid development of information technologies, various big graphs are prevalent in many real applications (e.g., social media and knowledge bases). An important component of these graphs is the network community. Essentially, a community is a group of vertices which are densely connected internally. Community retrieval can be used in many real applications, such as event organization, friend recommendation, and so on. Consequently, how to efficiently find high-quality communities from big graphs is an important research topic in the era of big data. Recently a large group of research works, called community search, have been proposed. They aim to provide efficient solutions for searching high-quality communities from large networks in real-time. Nevertheless, these works focus on different types of graphs and formulate communities in different manners, and thus it is desirable to have a comprehensive review of these works. In this survey, we conduct a thorough review of existing community search works. Moreover, we analyze and compare the quality of communities under their models, and the performance of different solutions. Furthermore, we point out new research directions. This survey does not only help researchers to have a better understanding of existing community search solutions, but also provides practitioners a better judgment on choosing the proper solutions.

A powerful approach to detecting erroneous data is to check which potentially dirty data records are incompatible with a user's domain knowledge. Previous approaches allow the user to specify domain knowledge in the form of logical constraints (e.g., functional dependency and denial constraints). We extend the constraint-based approach by introducing a novel class of statistical constraints (SCs). An SC treats each column as a random variable, and enforces an independence or dependence relationship between two (or a few) random variables. Statistical constraints are expressive, allowing the user to specify a wide range of domain knowledge, beyond traditional integrity constraints. Furthermore, they work harmoniously with downstream statistical modeling. We develop CODED, an SC-Oriented Data Error Detection system that supports three key tasks: (1) Checking whether an SC is violated or not on a given dataset, (2) Identify the top-k records that contribute the most to the violation of an SC, and (3) Checking whether a set of input SCs have conflicts or not. We present effective solutions for each task. Experiments on synthetic and real-world data illustrate how SCs apply to error detection, and provide evidence that CODED performs better than state-of-the-art approaches.

Given a graph $G$ and a vertex $q\in G$, the community search (CS) problem aims to efficiently find a subgraph of $G$ whose vertices are closely related to $q$. Communities are prevalent in social and biological networks, and can be used in product advertisement and social event recommendation. In this paper, we study profiled community search (PCS), where CS is performed on a profiled graph. This is a graph in which each vertex has labels arranged in a hierarchical manner. Extensive experiments show that PCS can identify communities with themes that are common to their vertices, and is more effective than existing CS approaches. As a naive solution for PCS is highly expensive, we have also developed a tree index, which facilitate efficient and online solutions for PCS.

Crowdsourcing employs human workers to solve computer-hard problems, such as data cleaning, entity resolution, and sentiment analysis. When crowdsourcing tabular data, e.g., the attribute values of an entity set, a worker's answers on the different attributes (e.g., the nationality and age of a celebrity star) are often treated independently. This assumption is not always true and can lead to suboptimal crowdsourcing performance. In this paper, we present the T-Crowd system, which takes into consideration the intricate relationships among tasks, in order to converge faster to their true values. Particularly, T-Crowd integrates each worker's answers on different attributes to effectively learn his/her trustworthiness and the true data values. The attribute relationship information is also used to guide task allocation to workers. Finally, T-Crowd seamlessly supports categorical and continuous attributes, which are the two main datatypes found in typical databases. Our extensive experiments on real and synthetic datasets show that T-Crowd outperforms state-of-the-art methods in terms of truth inference and reducing the cost of crowdsourcing.

Social networks are commonly used for marketing purposes. For example, free samples of a product can be given to a few influential social network users (or "seed nodes"), with the hope that they will convince their friends to buy it. One way to formalize marketers' objective is through influence maximization (or IM), whose goal is to find the best seed nodes to activate under a fixed budget, so that the number of people who get influenced in the end is maximized. Recent solutions to IM rely on the influence probability that a user influences another one. However, this probability information may be unavailable or incomplete. In this paper, we study IM in the absence of complete information on influence probability. We call this problem Online Influence Maximization (OIM) since we learn influence probabilities at the same time we run influence campaigns. To solve OIM, we propose a multiple-trial approach, where (1) some seed nodes are selected based on existing influence information; (2) an influence campaign is started with these seed nodes; and (3) users' feedback is used to update influence information. We adopt the Explore-Exploit strategy, which can select seed nodes using either the current influence probability estimation (exploit), or the confidence bound on the estimation (explore). Any existing IM algorithm can be used in this framework. We also develop an incremental algorithm that can significantly reduce the overhead of handling users' feedback information. Our experiments show that our solution is more effective than traditional IM methods on the partial information., Comment: 13 pages. To appear in KDD 2015. Extended version