Ming Zhong

Ming Zhong, Yang Liu, Da Yin et al. · microsoft-research

Multi-dimensional evaluation is the dominant paradigm for human evaluation in Natural Language Generation (NLG), i.e., evaluating the generated text from multiple explainable dimensions, such as coherence and fluency. However, automatic evaluation in NLG is still dominated by similarity-based metrics, and we lack a reliable framework for a more comprehensive evaluation of advanced models. In this paper, we propose a unified multi-dimensional evaluator UniEval for NLG. We re-frame NLG evaluation as a Boolean Question Answering (QA) task, and by guiding the model with different questions, we can use one evaluator to evaluate from multiple dimensions. Furthermore, thanks to the unified Boolean QA format, we are able to introduce an intermediate learning phase that enables UniEval to incorporate external knowledge from multiple related tasks and gain further improvement. Experiments on three typical NLG tasks show that UniEval correlates substantially better with human judgments than existing metrics. Specifically, compared to the top-performing unified evaluators, UniEval achieves a 23% higher correlation on text summarization, and over 43% on dialogue response generation. Also, UniEval demonstrates a strong zero-shot learning ability for unseen evaluation dimensions and tasks. Source code, data and all pre-trained evaluators are available on our GitHub repository (https://github.com/maszhongming/UniEval).

Siru Ouyang, Shuohang Wang, Yang Liu et al. · stanford

Recent progress in Large Language Models (LLMs) has produced models that exhibit remarkable performance across a variety of NLP tasks. However, it remains unclear whether the existing focus of NLP research accurately captures the genuine requirements of human users. This paper provides a comprehensive analysis of the divergence between current NLP research and the needs of real-world NLP applications via a large-scale collection of user-GPT conversations. We analyze a large-scale collection of real user queries to GPT. We compare these queries against existing NLP benchmark tasks and identify a significant gap between the tasks that users frequently request from LLMs and the tasks that are commonly studied in academic research. For example, we find that tasks such as ``design'' and ``planning'' are prevalent in user interactions but are largely neglected or different from traditional NLP benchmarks. We investigate these overlooked tasks, dissect the practical challenges they pose, and provide insights toward a roadmap to make LLMs better aligned with user needs.

Yizhu Jiao, Ming Zhong, Sha Li et al.

Large language models with instruction-following capabilities open the door to a wider group of users. However, when it comes to information extraction - a classic task in natural language processing - most task-specific systems cannot align well with long-tail ad hoc extraction use cases for non-expert users. To address this, we propose a novel paradigm, termed On-Demand Information Extraction, to fulfill the personalized demands of real-world users. Our task aims to follow the instructions to extract the desired content from the associated text and present it in a structured tabular format. The table headers can either be user-specified or inferred contextually by the model. To facilitate research in this emerging area, we present a benchmark named InstructIE, inclusive of both automatically generated training data, as well as the human-annotated test set. Building on InstructIE, we further develop an On-Demand Information Extractor, ODIE. Comprehensive evaluations on our benchmark reveal that ODIE substantially outperforms the existing open-source models of similar size. Our code and dataset are released on https://github.com/yzjiao/On-Demand-IE.

Yizhu Jiao, Sha Li, Yiqing Xie et al.

The argument role in event extraction refers to the relation between an event and an argument participating in it. Despite the great progress in event extraction, existing studies still depend on roles pre-defined by domain experts. These studies expose obvious weakness when extending to emerging event types or new domains without available roles. Therefore, more attention and effort needs to be devoted to automatically customizing argument roles. In this paper, we define this essential but under-explored task: open-vocabulary argument role prediction. The goal of this task is to infer a set of argument roles for a given event type. We propose a novel unsupervised framework, RolePred for this task. Specifically, we formulate the role prediction problem as an in-filling task and construct prompts for a pre-trained language model to generate candidate roles. By extracting and analyzing the candidate arguments, the event-specific roles are further merged and selected. To standardize the research of this task, we collect a new event extraction dataset from WikiPpedia including 142 customized argument roles with rich semantics. On this dataset, RolePred outperforms the existing methods by a large margin. Source code and dataset are available on our GitHub repository: https://github.com/yzjiao/RolePred

Yuning Mao, Ming Zhong, Jiawei Han

Scientific extreme summarization (TLDR) aims to form ultra-short summaries of scientific papers. Previous efforts on curating scientific TLDR datasets failed to scale up due to the heavy human annotation and domain expertise required. In this paper, we propose a simple yet effective approach to automatically extracting TLDR summaries for scientific papers from their citation texts. Based on the proposed approach, we create a new benchmark CiteSum without human annotation, which is around 30 times larger than the previous human-curated dataset SciTLDR. We conduct a comprehensive analysis of CiteSum, examining its data characteristics and establishing strong baselines. We further demonstrate the usefulness of CiteSum by adapting models pre-trained on CiteSum (named CITES) to new tasks and domains with limited supervision. For scientific extreme summarization, CITES outperforms most fully-supervised methods on SciTLDR without any fine-tuning and obtains state-of-the-art results with only 128 examples. For news extreme summarization, CITES achieves significant gains on XSum over its base model (not pre-trained on CiteSum), e.g., +7.2 ROUGE-1 zero-shot performance and state-of-the-art few-shot performance. For news headline generation, CITES performs the best among unsupervised and zero-shot methods on Gigaword. Our dataset and code can be found at https://github.com/morningmoni/CiteSum.

Negar Arabzadeh, Ali Ahmadvand, Julia Kiseleva et al. · microsoft-research

The recent increase in the volume of online meetings necessitates automated tools for managing and organizing the material, especially when an attendee has missed the discussion and needs assistance in quickly exploring it. In this work, we propose a novel end-to-end framework for generating interactive questionnaires for preference-based meeting exploration. As a result, users are supplied with a list of suggested questions reflecting their preferences. Since the task is new, we introduce an automatic evaluation strategy. Namely, it measures how much the generated questions via questionnaire are answerable to ensure factual correctness and covers the source meeting for the depth of possible exploration.

Yulong Chen, Ming Zhong, Xuefeng Bai et al. · cambridge

We propose the shared task of cross-lingual conversation summarization, \emph{ConvSumX Challenge}, opening new avenues for researchers to investigate solutions that integrate conversation summarization and machine translation. This task can be particularly useful due to the emergence of online meetings and conferences. We construct a new benchmark, covering 2 real-world scenarios and 3 language directions, including a low-resource language. We hope that \emph{ConvSumX} can motivate researches to go beyond English and break the barrier for non-English speakers to benefit from recent advances of conversation summarization.

Chenxin An, Ming Zhong, Zhiyong Wu et al.

Traditional training paradigms for extractive and abstractive summarization systems always only use token-level or sentence-level training objectives. However, the output summary is always evaluated from summary-level which leads to the inconsistency in training and evaluation. In this paper, we propose a Contrastive Learning based re-ranking framework for one-stage summarization called COLO. By modeling a contrastive objective, we show that the summarization model is able to directly generate summaries according to the summary-level score without additional modules and parameters. Extensive experiments demonstrate that COLO boosts the extractive and abstractive results of one-stage systems on CNN/DailyMail benchmark to 44.58 and 46.33 ROUGE-1 score while preserving the parameter efficiency and inference efficiency. Compared with state-of-the-art multi-stage systems, we save more than 100 GPU training hours and obtaining 3~8 speed-up ratio during inference while maintaining comparable results.

Dugang Liu, Pengxiang Cheng, Hong Zhu et al.

Tabular data is one of the most common data storage formats behind many real-world web applications such as retail, banking, and e-commerce. The success of these web applications largely depends on the ability of the employed machine learning model to accurately distinguish influential features from all the predetermined features in tabular data. Intuitively, in practical business scenarios, different instances should correspond to different sets of influential features, and the set of influential features of the same instance may vary in different scenarios. However, most existing methods focus on global feature selection assuming that all instances have the same set of influential features, and few methods considering instance-wise feature selection ignore the variability of influential features in different scenarios. In this paper, we first introduce a new perspective based on the influence function for instance-wise feature selection, and give some corresponding theoretical insights, the core of which is to use the influence function as an indicator to measure the importance of an instance-wise feature. We then propose a new solution for discovering instance-wise influential features in tabular data (DIWIFT), where a self-attention network is used as a feature selection model and the value of the corresponding influence function is used as an optimization objective to guide the model. Benefiting from the advantage of the influence function, i.e., its computation does not depend on a specific architecture and can also take into account the data distribution in different scenarios, our DIWIFT has better flexibility and robustness. Finally, we conduct extensive experiments on both synthetic and real-world datasets to validate the effectiveness of our DIWIFT.

Ting Zhang, Zhuang Chen, Ming Zhong et al.

Emotion recognition in conversation, which aims to predict the emotion for all utterances, has attracted considerable research attention in recent years. It is a challenging task since the recognition of the emotion in one utterance involves many complex factors, such as the conversational context, the speaker's background, and the subtle difference between emotion labels. In this paper, we propose a novel framework which mimics the thinking process when modeling these factors. Specifically, we first comprehend the conversational context with a history-oriented prompt to selectively gather information from predecessors of the target utterance. We then model the speaker's background with an experience-oriented prompt to retrieve the similar utterances from all conversations. We finally differentiate the subtle label semantics with a paraphrasing mechanism to elicit the intrinsic label related knowledge. We conducted extensive experiments on three benchmarks. The empirical results demonstrate the superiority of our proposed framework over the state-of-the-art baselines.

Jinchao Feng, Mauro Maggioni, Patrick Martin et al.

Dynamical systems across many disciplines are modeled as interacting particles or agents, with interaction rules that depend on a very small number of variables (e.g. pairwise distances, pairwise differences of phases, etc...), functions of the state of pairs of agents. Yet, these interaction rules can generate self-organized dynamics, with complex emergent behaviors (clustering, flocking, swarming, etc.). We propose a learning technique that, given observations of states and velocities along trajectories of the agents, yields both the variables upon which the interaction kernel depends and the interaction kernel itself, in a nonparametric fashion. This yields an effective dimension reduction which avoids the curse of dimensionality from the high-dimensional observation data (states and velocities of all the agents). We demonstrate the learning capability of our method to a variety of first-order interacting systems.

Guowei Wu, Weike Pan, Zhong Ming

Graph neural networks (GNNs) have gained wide popularity in recommender systems due to their capability to capture higher-order structure information among the nodes of users and items. However, these methods need to collect personal interaction data between a user and the corresponding items and then model them in a central server, which would break the privacy laws such as GDPR. So far, no existing work can construct a global graph without leaking each user's private interaction data (i.e., his or her subgraph). In this paper, we are the first to design a novel lossless federated recommendation framework based on GNN, which achieves full-graph training with complete high-order structure information, enabling the training process to be equivalent to the corresponding un-federated counterpart. In addition, we use LightGCN to instantiate an example of our framework and show its equivalence.

Jiaxiang Liu, Yunhan Xing, Xiaomu Shi et al.

As a new programming paradigm, deep neural networks (DNNs) have been increasingly deployed in practice, but the lack of robustness hinders their applications in safety-critical domains. While there are techniques for verifying DNNs with formal guarantees, they are limited in scalability and accuracy. In this paper, we present a novel abstraction-refinement approach for scalable and exact DNN verification. Specifically, we propose a novel abstraction to break down the size of DNNs by over-approximation. The result of verifying the abstract DNN is always conclusive if no spurious counterexample is reported. To eliminate spurious counterexamples introduced by abstraction, we propose a novel counterexample-guided refinement that refines the abstract DNN to exclude a given spurious counterexample while still over-approximating the original one. Our approach is orthogonal to and can be integrated with many existing verification techniques. For demonstration, we implement our approach using two promising and exact tools Marabou and Planet as the underlying verification engines, and evaluate on widely-used benchmarks ACAS Xu, MNIST and CIFAR-10. The results show that our approach can boost their performance by solving more problems and reducing up to 86.3% and 78.0% verification time, respectively. Compared to the most relevant abstraction-refinement approach, our approach is 11.6-26.6 times faster.

Yinghui Pan, Hanyi Zhang, Yifeng Zeng et al.

Modelling other agents' behaviors plays an important role in decision models for interactions among multiple agents. To optimise its own decisions, a subject agent needs to model what other agents act simultaneously in an uncertain environment. However, modelling insufficiency occurs when the agents are competitive and the subject agent can not get full knowledge about other agents. Even when the agents are collaborative, they may not share their true behaviors due to their privacy concerns. In this article, we investigate into diversifying behaviors of other agents in the subject agent's decision model prior to their interactions. Starting with prior knowledge about other agents' behaviors, we use a linear reduction technique to extract representative behavioral features from the known behaviors. We subsequently generate their new behaviors by expanding the features and propose two diversity measurements to select top-K behaviors. We demonstrate the performance of the new techniques in two well-studied problem domains. This research will contribute to intelligent systems dealing with unknown unknowns in an open artificial intelligence world.

Da Yin, Xiao Liu, Fan Yin et al.

Instruction tuning has emerged to enhance the capabilities of large language models (LLMs) to comprehend instructions and generate appropriate responses. Existing methods either manually annotate or employ LLM (e.g., GPT-series) to generate data for instruction tuning. However, they often overlook associating instructions with existing annotated datasets. In this paper, we propose Dynosaur, a dynamic growth paradigm for the automatic curation of instruction-tuning data. Based on the metadata of existing datasets, we use LLMs to automatically construct instruction-tuning data by identifying relevant data fields and generating appropriate instructions. By leveraging the existing annotated datasets, Dynosaur offers several advantages: 1) it reduces the API cost for generating instructions (e.g., it costs less than $12 USD by calling GPT-3.5-turbo for generating 800K instruction tuning samples; 2) it provides high-quality data for instruction tuning (e.g., it performs better than Alpaca and Flan on Super-NI and Longform with comparable data sizes); and 3) it supports the continuous improvement of models by generating instruction-tuning data when a new annotated dataset becomes available. We further investigate a continual learning scheme for learning with the ever-growing instruction-tuning dataset, and demonstrate that replaying tasks with diverse instruction embeddings not only helps mitigate forgetting issues but generalizes to unseen tasks better. Code and data are available at https://github.com/WadeYin9712/Dynosaur.

Ming Zhong, Yang Liu, Suyu Ge et al.

Text summarization is a user-preference based task, i.e., for one document, users often have different priorities for summary. As a key aspect of customization in summarization, granularity is used to measure the semantic coverage between the summary and source document. However, developing systems that can generate summaries with customizable semantic coverage is still an under-explored topic. In this paper, we propose the first unsupervised multi-granularity summarization framework, GranuSum. We take events as the basic semantic units of the source documents and propose to rank these events by their salience. We also develop a model to summarize input documents with given events as anchors and hints. By inputting different numbers of events, GranuSum is capable of producing multi-granular summaries in an unsupervised manner. Meanwhile, we annotate a new benchmark GranuDUC that contains multiple summaries at different granularities for each document cluster. Experimental results confirm the substantial superiority of GranuSum on multi-granularity summarization over strong baselines. Further, by exploiting the event information, GranuSum also exhibits state-of-the-art performance under the conventional unsupervised abstractive setting. Dataset for this paper can be found at: https://github.com/maszhongming/GranuDUC

Tianbao Xie, Chen Henry Wu, Peng Shi et al.

Structured knowledge grounding (SKG) leverages structured knowledge to complete user requests, such as semantic parsing over databases and question answering over knowledge bases. Since the inputs and outputs of SKG tasks are heterogeneous, they have been studied separately by different communities, which limits systematic and compatible research on SKG. In this paper, we overcome this limitation by proposing the UnifiedSKG framework, which unifies 21 SKG tasks into a text-to-text format, aiming to promote systematic SKG research, instead of being exclusive to a single task, domain, or dataset. We use UnifiedSKG to benchmark T5 with different sizes and show that T5, with simple modifications when necessary, achieves state-of-the-art performance on almost all of the 21 tasks. We further demonstrate that multi-task prefix-tuning improves the performance on most tasks, largely improving the overall performance. UnifiedSKG also facilitates the investigation of zero-shot and few-shot learning, and we show that T0, GPT-3, and Codex struggle in zero-shot and few-shot learning for SKG. We also use UnifiedSKG to conduct a series of controlled experiments on structured knowledge encoding variants across SKG tasks. UnifiedSKG is easily extensible to more tasks, and it is open-sourced at https://github.com/hkunlp/unifiedskg.

Minhao Jiang, Ken Ziyu Liu, Ming Zhong et al.

Language models pre-trained on web-scale corpora demonstrate impressive capabilities on diverse downstream tasks. However, there is increasing concern whether such capabilities might arise from evaluation datasets being included in the pre-training corpus -- a phenomenon known as \textit{data contamination} -- in a manner that artificially increases performance. There has been little understanding of how this potential contamination might influence LMs' performance on downstream tasks. In this paper, we explore the impact of data contamination at the pre-training stage by pre-training a series of GPT-2 models \textit{from scratch}. We highlight the effect of both text contamination (\textit{i.e.}\ input text of the evaluation samples) and ground-truth contamination (\textit{i.e.}\ the prompts asked on the input and the desired outputs) from evaluation data. We also investigate the effects of repeating contamination for various downstream tasks. Additionally, we examine the prevailing n-gram-based definitions of contamination within current LLM reports, pinpointing their limitations and inadequacy. Our findings offer new insights into data contamination's effects on language model capabilities and underscore the need for independent, comprehensive contamination assessments in LLM studies.

Ming Zhong, Yelong Shen, Shuohang Wang et al.

Low-Rank Adaptation (LoRA) is extensively utilized in text-to-image models for the accurate rendition of specific elements like distinct characters or unique styles in generated images. Nonetheless, existing methods face challenges in effectively composing multiple LoRAs, especially as the number of LoRAs to be integrated grows, thus hindering the creation of complex imagery. In this paper, we study multi-LoRA composition through a decoding-centric perspective. We present two training-free methods: LoRA Switch, which alternates between different LoRAs at each denoising step, and LoRA Composite, which simultaneously incorporates all LoRAs to guide more cohesive image synthesis. To evaluate the proposed approaches, we establish ComposLoRA, a new comprehensive testbed as part of this research. It features a diverse range of LoRA categories with 480 composition sets. Utilizing an evaluation framework based on GPT-4V, our findings demonstrate a clear improvement in performance with our methods over the prevalent baseline, particularly evident when increasing the number of LoRAs in a composition. The code, benchmarks, LoRA weights, and all evaluation details are available on our project website: https://maszhongming.github.io/Multi-LoRA-Composition.

Yunfeng Ning, Mayi Xu, Jintao Wen et al.

LLMs often suffer from hallucinations and outdated or incomplete knowledge. RAG is proposed to address these issues by integrating external knowledge like that in KGs into LLMs. However, leveraging private KGs in RAG systems poses significant privacy risks due to the black-box nature of LLMs and potential insecure data transmission, especially when using third-party LLM APIs lacking transparency and control. In this paper, we investigate the privacy-protected RAG scenario for the first time, where entities in KGs are anonymous for LLMs, thus preventing them from accessing entity semantics. Due to the loss of semantics of entities, previous RAG systems cannot retrieve question-relevant knowledge from KGs by matching questions with the meaningless identifiers of anonymous entities. To realize an effective RAG system in this scenario, two key challenges must be addressed: (1) How can anonymous entities be converted into retrievable information. (2) How to retrieve question-relevant anonymous entities. Hence, we propose a novel ARoG framework including relation-centric abstraction and structure-oriented abstraction strategies. For challenge (1), the first strategy abstracts entities into high-level concepts by dynamically capturing the semantics of their adjacent relations. It supplements meaningful semantics which can further support the retrieval process. For challenge (2), the second strategy transforms unstructured natural language questions into structured abstract concept paths. These paths can be more effectively aligned with the abstracted concepts in KGs, thereby improving retrieval performance. To guide LLMs to effectively retrieve knowledge from KGs, the two strategies strictly protect privacy from being exposed to LLMs. Experiments on three datasets demonstrate that ARoG achieves strong performance and privacy-robustness.

Yichen Yan, Ming Zhong, Qi Zhu et al.

Multimodal large language models (MLLMs) rely heavily on instruction tuning to align vision and language capabilities, yet the computational cost of training on large-scale datasets remains a major bottleneck. Existing data selection methods aim to mitigate this by selecting important and diverse subsets, but they often suffer from two critical drawbacks: high computational overhead from processing the entire dataset and suboptimal data selection due to separate treatment of importance and diversity. We introduce CoIDO, a novel dual-objective framework that jointly optimizes data importance and diversity to overcome these challenges. Unlike existing approaches that require costly evaluations across the whole dataset, CoIDO employs a lightweight plug-in scorer. This scorer is trained on just a small random sample of data to learn the distribution of the candidate set, drastically reducing computational demands. By leveraging a homoscedastic uncertainty-based formulation, CoIDO effectively balances importance and diversity during training, enabling efficient and scalable data selection. In our experiments, we trained the CoIDO scorer using only 20 percent of randomly sampled data. Once trained, CoIDO was applied to the entire dataset to select a 20 percent subset for instruction tuning. On the widely used LLaVA-1.5-7B model across ten downstream tasks, this selected subset achieved an impressive 98.2 percent of the performance of full-data fine-tuning, on average.

Ziheng Guo, James Greene, Ming Zhong

Stochastic differential equations (SDEs) are a ubiquitous modeling framework that finds applications in physics, biology, engineering, social science, and finance. Due to the availability of large-scale data sets, there is growing interest in learning mechanistic models from observations with stochastic noise. In this work, we present a nonparametric framework to learn both the drift and diffusion terms in systems of SDEs where the stochastic noise is singular. Specifically, inspired by second-order equations from classical physics, we consider systems which possess structured noise, i.e. noise with a singular covariance matrix. We provide an algorithm for constructing estimators given trajectory data and demonstrate the effectiveness of our methods via a number of examples from physics and biology. As the developed framework is most naturally applicable to systems possessing a high degree of dimensionality reduction (i.e. symmetry), we also apply it to the high dimensional Cucker-Smale flocking model studied in collective dynamics and show that it is able to accurately infer the low dimensional interaction kernel from particle data.

Muhui Jiang, Tianyi Xu, Yajin Zhou et al.

Emulator is widely used to build dynamic analysis frameworks due to its fine-grained tracing capability, full system monitoring functionality, and scalability of running on different operating systemsand architectures. However, whether the emulator is consistent with real devices is unknown. To understand this problem, we aim to automatically locate inconsistent instructions, which behave differently between emulators and real devices. We target ARM architecture, which provides machine readable specification. Based on the specification, we propose a test case generator by designing and implementing the first symbolic execution engine for ARM architecture specification language (ASL). We generate 2,774,649 representative instruction streams and conduct differential testing with these instruction streams between four ARM real devices in different architecture versions (i.e., ARMv5, ARMv6, ARMv7-a, and ARMv8-a) and the state-of-the-art emulators (i.e., QEMU). We locate 155,642 inconsistent instruction streams, which cover 30% of all instruction encodings and 47.8% of the instructions. We find undefined implementation in ARM manual and implementation bugs of QEMU are the major causes of inconsistencies. Furthermore, we discover four QEMU bugs, which are confirmed and patched by thedevelopers, covering 13 instruction encodings including the most commonly used ones (e.g.,STR,BLX). With the inconsistent instructions, we build three security applications and demonstrate thecapability of these instructions on detecting emulators, anti-emulation, and anti-fuzzing.

Guotong Xue, Ming Zhong, Jianxin Li et al.

Since many real world networks are evolving over time, such as social networks and user-item networks, there are increasing research efforts on dynamic network embedding in recent years. They learn node representations from a sequence of evolving graphs but not only the latest network, for preserving both structural and temporal information from the dynamic networks. Due to the lack of comprehensive investigation of them, we give a survey of dynamic network embedding in this paper. Our survey inspects the data model, representation learning technique, evaluation and application of current related works and derives common patterns from them. Specifically, we present two basic data models, namely, discrete model and continuous model for dynamic networks. Correspondingly, we summarize two major categories of dynamic network embedding techniques, namely, structural-first and temporal-first that are adopted by most related works. Then we build a taxonomy that refines the category hierarchy by typical learning models. The popular experimental data sets and applications are also summarized. Lastly, we have a discussion of several distinct research topics in dynamic network embedding.