arXiv subjects

Zilin Du

Publications and source records attributed to Zilin Du.

12 recordsLinked to original sources

arXiv2026

On the Difficulty of Learning a Meta-network for Training Data Selection

Synthetic data are increasingly used to train neural networks, yet distributional mismatch with real data limits their effectiveness when used indiscriminately. A common strategy is to learn data weights via bi-level optimization, which we refer to as Meta-learning for Training-data Selection (MTS). Interestingly, in practice, MTS often performs below expectation. We identify two obstacles in properly training MTS: a poor gradient signal-to-noise ratio (GSNR), which causes optimization difficulties, and lack of informative features that correlates with data quality. We present a mathematical analysis of MTS, which reveals the dynamics of normalized data weights and the relation between disparate data quality and poor GSNR. The analysis suggests a a simple yet effective solution: increasing the batch size. Further, we propose a set of informative features that capture the positions of training data in their distributions and training dynamics. Experiments across four benchmarks show consistent improvements, achieving average gains of 5.49% over training without selection and 2.89% over the strongest baseline.

cs.LG↗

arXiv2024

Paint Outside the Box: Synthesizing and Selecting Training Data for Visual Grounding

Visual grounding aims to localize the image regions based on a textual query. Given the difficulty of large-scale data curation, we investigate how to effectively learn visual grounding under data-scarce settings in this paper. To address the data scarcity, we propose a novel framework, POBF (Paint Outside the Box and Filter). POBF synthesizes images by inpainting outside the box, tackling a label misalignment issue encountered in previous works. Furthermore, POBF leverages an innovative filtering scheme to select the most effective training data. This scheme combines a hardness score and an overfitting score, balanced by a penalty term. Extensive experiments across four benchmark datasets demonstrate that POBF consistently improves performance, achieving an average gain of 5.83\% over the real-data-only method and outperforming leading baselines by 2.29\%-3.85\% in accuracy. Additionally, we validate the robustness and generalizability of POBF across various generative models, training data sizes, and model architectures.

cs.CV↗

arXiv2024

Diversify, Rationalize, and Combine: Ensembling Multiple QA Strategies for Zero-shot Knowledge-based VQA

Knowledge-based Visual Question-answering (K-VQA) often requires the use of background knowledge beyond the image. However, we discover that a single knowledge generation strategy is often insufficient for all K-VQA questions. To this end, we propose Diversification, Evidence Truncation, and Combination for Knowledge-based Elucidation (DietCoke), which utilizes a bundle of complementary question-answering tactics and aggregates their answers using textual rationales. DietCoke comprises of three stages: diversification, rationalization, and ensemble. The diversification stage generates three distinctive decision contexts, each leading to its own answer candidate. The rationalization stage generates two rationales, the automatic rationale and the mechanistic rationale, for each answer candidate using decorrelated techniques. Finally, in the ensemble stage, an LLM informed by the rationales selects one answer from the three candidates. Experiments show that DietCoke significantly outperforms state-of-the-art LLM-based baselines by 2.8% on OK-VOA and 4.7% on A-OKVOA and that the strategies in the ensembles are highly complementary. Code is available at: https://github.com/limiaoyu/DietCoke

cs.CL↗

arXiv2024

Generating Synthetic Datasets for Few-shot Prompt Tuning

A major limitation of prompt tuning is its dependence on large labeled training datasets. Under few-shot learning settings, prompt tuning lags far behind full-model fine-tuning, limiting its scope of application. In this paper, we leverage the powerful LLMs to synthesize task-specific labeled data for training the soft prompts. We first introduce a distribution-aligned weighted generator tuning (DawGen) method to encourage generating in-distribution data that aligns with the few-shot real data. Then, we train soft prompts on both synthetic and real datasets using a gradient surgery approach, which eliminates the conflicting gradients from different data sources. Experiments on seven sentence-pair classification datasets demonstrate the effectiveness of our proposed method for boosting prompt tuning in few-shot learning settings. Results on QQP, MRPC, and SICK datasets are even comparable to the performance of transfer learning from large real-world datasets, showing the promise of synthetic data as an alternative for enhancing soft prompt tuning.

cs.CL↗

arXiv2023

Training on Synthetic Data Beats Real Data in Multimodal Relation Extraction

The task of multimodal relation extraction has attracted significant research attention, but progress is constrained by the scarcity of available training data. One natural thought is to extend existing datasets with cross-modal generative models. In this paper, we consider a novel problem setting, where only unimodal data, either text or image, are available during training. We aim to train a multimodal classifier from synthetic data that perform well on real multimodal test data. However, training with synthetic data suffers from two obstacles: lack of data diversity and label information loss. To alleviate the issues, we propose Mutual Information-aware Multimodal Iterated Relational dAta GEneration (MI2RAGE), which applies Chained Cross-modal Generation (CCG) to promote diversity in the generated data and exploits a teacher network to select valuable training samples with high mutual information with the ground-truth labels. Comparing our method to direct training on synthetic data, we observed a significant improvement of 24.06% F1 with synthetic text and 26.42% F1 with synthetic images. Notably, our best model trained on completely synthetic images outperforms prior state-of-the-art models trained on real multimodal data by a margin of 3.76% in F1. Our codebase will be made available upon acceptance.

cs.AI↗

arXiv2023

Training Multimedia Event Extraction With Generated Images and Captions

Contemporary news reporting increasingly features multimedia content, motivating research on multimedia event extraction. However, the task lacks annotated multimodal training data and artificially generated training data suffer from distribution shift from real-world data. In this paper, we propose Cross-modality Augmented Multimedia Event Learning (CAMEL), which successfully utilizes artificially generated multimodal training data and achieves state-of-the-art performance. We start with two labeled unimodal datasets in text and image respectively, and generate the missing modality using off-the-shelf image generators like Stable Diffusion and image captioners like BLIP. After that, we train the network on the resultant multimodal datasets. In order to learn robust features that are effective across domains, we devise an iterative and gradual training strategy. Substantial experiments show that CAMEL surpasses state-of-the-art (SOTA) baselines on the M2E2 benchmark. On multimedia events in particular, we outperform the prior SOTA by 4.2% F1 on event mention identification and by 9.8% F1 on argument identification, which indicates that CAMEL learns synergistic representations from the two modalities. Our work demonstrates a recipe to unleash the power of synthetic training data in structured prediction.

cs.MM↗

arXiv2022

HUSP-SP: Faster Utility Mining on Sequence Data

High-utility sequential pattern mining (HUSPM) has emerged as an important topic due to its wide application and considerable popularity. However, due to the combinatorial explosion of the search space when the HUSPM problem encounters a low utility threshold or large-scale data, it may be time-consuming and memory-costly to address the HUSPM problem. Several algorithms have been proposed for addressing this problem, but they still cost a lot in terms of running time and memory usage. In this paper, to further solve this problem efficiently, we design a compact structure called sequence projection (seqPro) and propose an efficient algorithm, namely discovering high-utility sequential patterns with the seqPro structure (HUSP-SP). HUSP-SP utilizes the compact seq-array to store the necessary information in a sequence database. The seqPro structure is designed to efficiently calculate candidate patterns' utilities and upper bound values. Furthermore, a new upper bound on utility, namely tighter reduced sequence utility (TRSU) and two pruning strategies in search space, are utilized to improve the mining performance of HUSP-SP. Experimental results on both synthetic and real-life datasets show that HUSP-SP can significantly outperform the state-of-the-art algorithms in terms of running time, memory usage, search space pruning efficiency, and scalability.

cs.DB↗

arXiv2021

Utility-driven Mining of Contiguous Sequences

Recently, contiguous sequential pattern mining (CSPM) gained interest as a research topic, due to its varied potential real-world applications, such as web log and biological sequence analysis. To date, studies on the CSPM problem remain in preliminary stages. Existing CSPM algorithms lack the efficiency to satisfy users' needs and can still be improved in terms of runtime and memory consumption. In addition, existing algorithms were developed to deal with simple sequence data, working with only one event at a time. Complex sequence data, which represent multiple events occurring simultaneously, are also commonly observed in real life. In this paper, we propose a novel algorithm, fast utility-driven contiguous sequential pattern mining (FUCPM), to address the CSPM problem. FUCPM adopts a compact sequence information list and instance chain structures to store the necessary information of the database and candidate patterns. For further efficiency, we develop the global unpromising items pruning and local unpromising items pruning strategies, based on sequence-weighted utilization and item-extension utilization, to reduce the search space. Extensive experiments on real-world and synthetic datasets demonstrate that FUCPM outperforms the state-of-the-art algorithms and is scalable enough to handle complex sequence data.

cs.DB↗

arXiv2021

Explainable Fuzzy Utility Mining on Sequences

Fuzzy systems have good modeling capabilities in several data science scenarios, and can provide human-explainable intelligence models with explainability and interpretability. In contrast to transaction data, which have been extensively studied, sequence data are more common in real-life applications. To obtain a human-explainable data intelligence model for decision making, in this study, we investigate explainable fuzzy-theoretic utility mining on multi-sequences. Meanwhile, a more normative formulation of the problem of fuzzy utility mining on sequences is formulated. By exploring fuzzy set theory for utility mining, we propose a novel method termed pattern growth fuzzy utility mining (PGFUM) for mining fuzzy high-utility sequences with linguistic meaning. In the case of sequence data, PGFUM reflects the fuzzy quantity and utility regions of sequences. To improve the efficiency and feasibility of PGFUM, we develop two compressed data structures with explainable fuzziness. Furthermore, one existing and two new upper bounds on the explainable fuzzy utility of candidates are adopted in three proposed pruning strategies to substantially reduce the search space and thus expedite the mining process. Finally, the proposed PGFUM algorithm is compared with PFUS, which is the only currently available method for the same task, through extensive experimental evaluation. It is demonstrated that PGFUM achieves not only human-explainable mining results that contain the original nature of revealable intelligibility, but also high efficiency in terms of runtime and memory cost.

cs.DB↗

arXiv2021

TUSQ: Targeted High-Utility Sequence Querying

Significant efforts have been expended in the research and development of a database management system (DBMS) that has a wide range of applications for managing an enormous collection of multisource, heterogeneous, complex, or growing data. Besides the primary function (i.e., create, delete, and update), a practical and impeccable DBMS can interact with users through information selection, that is, querying with their targets. Previous querying algorithms, such as frequent itemset querying and sequential pattern querying (SPQ) have focused on the measurement of frequency, which does not involve the concept of utility, which is helpful for users to discover more informative patterns. To apply the querying technology for wider applications, we incorporate utility into target-oriented SPQ and formulate the task of targeted utility-oriented sequence querying. To address the proposed problem, we develop a novel algorithm, namely targeted high-utility sequence querying (TUSQ), based on two novel upper bounds suffix remain utility and terminated descendants utility as well as a vertical Last Instance Table structure. For further efficiency, TUSQ relies on a projection technology utilizing a compact data structure called the targeted chain. An extensive experimental study conducted on several real and synthetic datasets shows that the proposed algorithm outperformed the designed baseline algorithm in terms of runtime, memory consumption, and candidate filtering.

cs.DB↗

arXiv2020

TKUS: Mining Top-K High-Utility Sequential Patterns

High-utility sequential pattern mining (HUSPM) has recently emerged as a focus of intense research interest. The main task of HUSPM is to find all subsequences, within a quantitative sequential database, that have high utility with respect to a user-defined minimum utility threshold. However, it is difficult to specify the minimum utility threshold, especially when database features, which are invisible in most cases, are not understood. To handle this problem, top-k HUSPM was proposed. Up to now, only very preliminary work has been conducted to capture top-k HUSPs, and existing strategies require improvement in terms of running time, memory consumption, unpromising candidate filtering, and scalability. Moreover, no systematic problem statement has been defined. In this paper, we formulate the problem of top-k HUSPM and propose a novel algorithm called TKUS. To improve efficiency, TKUS adopts a projection and local search mechanism and employs several schemes, including the Sequence Utility Raising, Terminate Descendants Early, and Eliminate Unpromising Items strategies, which allow it to greatly reduce the search space. Finally, experimental results demonstrate that TKUS can achieve sufficiently good top-k HUSPM performance compared to state-of-the-art algorithm TKHUS-Span.

cs.DB↗

arXiv2020

On-shelf Utility Mining of Sequence Data

Utility mining has emerged as an important and interesting topic owing to its wide application and considerable popularity. However, conventional utility mining methods have a bias toward items that have longer on-shelf time as they have a greater chance to generate a high utility. To eliminate the bias, the problem of on-shelf utility mining (OSUM) is introduced. In this paper, we focus on the task of OSUM of sequence data, where the sequential database is divided into several partitions according to time periods and items are associated with utilities and several on-shelf time periods. To address the problem, we propose two methods, OSUM of sequence data (OSUMS) and OSUMS+, to extract on-shelf high-utility sequential patterns. For further efficiency, we also designed several strategies to reduce the search space and avoid redundant calculation with two upper bounds time prefix extension utility (TPEU) and time reduced sequence utility (TRSU). In addition, two novel data structures were developed for facilitating the calculation of upper bounds and utilities. Substantial experimental results on certain real and synthetic datasets show that the two methods outperform the state-of-the-art algorithm. In conclusion, OSUMS may consume a large amount of memory and is unsuitable for cases with limited memory, while OSUMS+ has wider real-life applications owing to its high efficiency.

cs.DB↗