Qiaozhu Mei

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao et al. · allen-ai, amazon-science

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

Brian Ondov, Chia-Hsuan Chang, Weipeng Zhou et al.

High-dimensional biomedical data, such as cell-by-gene matrices, are increasingly generated temporally. However, Manifold Learning algorithms, like t-SNE and UMAP, cannot incorporate time-ordering in their layouts, obfuscating the dynamics of cell types or other classes. As a solution, we present IRIS, a new Manifold Learning algorithm that structures layouts both chronologically and by manifold topology. IRIS can visualize a wide range of dynamic biomedical data, including scRNA-seq, comparative metagenomics, and literature.

Cheng Li, Mingyang Zhang, Qiaozhu Mei et al. · baidu, deepmind

Personalized text generation is an emerging research area that has attracted much attention in recent years. Most studies in this direction focus on a particular domain by designing bespoke features or models. In this work, we propose a general approach for personalized text generation using large language models (LLMs). Inspired by the practice of writing education, we develop a multistage and multitask framework to teach LLMs for personalized generation. In writing instruction, the task of writing from sources is often decomposed into multiple steps that involve finding, evaluating, summarizing, synthesizing, and integrating information. Analogously, our approach to personalized text generation consists of multiple stages: retrieval, ranking, summarization, synthesis, and generation. In addition, we introduce a multitask setting that helps the model improve its generation ability further, which is inspired by the observation in education that a student's reading proficiency and writing ability are often correlated. We evaluate our approach on three public datasets, each of which covers a different and representative domain. Our results show significant improvements over a variety of baselines.

Jiaqi Ma, Xingjian Zhang, Hezheng Fan et al.

Establishing open and general benchmarks has been a critical driving force behind the success of modern machine learning techniques. As machine learning is being applied to broader domains and tasks, there is a need to establish richer and more diverse benchmarks to better reflect the reality of the application scenarios. Graph learning is an emerging field of machine learning that urgently needs more and better benchmarks. To accommodate the need, we introduce Graph Learning Indexer (GLI), a benchmark curation platform for graph learning. In comparison to existing graph learning benchmark libraries, GLI highlights two novel design objectives. First, GLI is designed to incentivize \emph{dataset contributors}. In particular, we incorporate various measures to minimize the effort of contributing and maintaining a dataset, increase the usability of the contributed dataset, as well as encourage attributions to different contributors of the dataset. Second, GLI is designed to curate a knowledge base, instead of a plain collection, of benchmark datasets. We use multiple sources of meta information to augment the benchmark datasets with \emph{rich characteristics}, so that they can be easily selected and used in downstream research or development. The source code of GLI is available at \url{https://github.com/Graph-Learning-Benchmarks/gli}.

Yutong Xie, Zhaoying Pan, Jinge Ma et al.

Recent developments in large language models (LLM) and generative AI have unleashed the astonishing capabilities of text-to-image generation systems to synthesize high-quality images that are faithful to a given reference text, known as a "prompt". These systems have immediately received lots of attention from researchers, creators, and common users. Despite the plenty of efforts to improve the generative models, there is limited work on understanding the information needs of the users of these systems at scale. We conduct the first comprehensive analysis of large-scale prompt logs collected from multiple text-to-image generation systems. Our work is analogous to analyzing the query logs of Web search engines, a line of work that has made critical contributions to the glory of the Web search industry and research. Compared with Web search queries, text-to-image prompts are significantly longer, often organized into special structures that consist of the subject, form, and intent of the generation tasks and present unique categories of information needs. Users make more edits within creation sessions, which present remarkable exploratory patterns. There is also a considerable gap between the user-input prompts and the captions of the images included in the open training data of the generative models. Our findings provide concrete implications on how to improve text-to-image generation systems for creation purposes.

Cristina Garbacea, Qiaozhu Mei

Recent advances in deep neural language models combined with the capacity of large scale datasets have accelerated the development of natural language generation systems that produce fluent and coherent texts (to various degrees of success) in a multitude of tasks and application contexts. However, controlling the output of these models for desired user and task needs is still an open challenge. This is crucial not only to customizing the content and style of the generated language, but also to their safe and reliable deployment in the real world. We present an extensive survey on the emerging topic of constrained neural language generation in which we formally define and categorize the problems of natural language generation by distinguishing between conditions and constraints (the latter being testable conditions on the output text instead of the input), present constrained text generation tasks, and review existing methods and evaluation metrics for constrained text generation. Our aim is to highlight recent progress and trends in this emerging field, informing on the most promising directions and limitations towards advancing the state-of-the-art of constrained neural language generation research.

Yuhang Zhou, Xuan Lu, Ge Gao et al.

Although remote working is increasingly adopted during the pandemic, many are concerned by the low-efficiency in the remote working. Missing in text-based communication are non-verbal cues such as facial expressions and body language, which hinders the effective communication and negatively impacts the work outcomes. Prevalent on social media platforms, emojis, as alternative non-verbal cues, are gaining popularity in the virtual workspaces well. In this paper, we study how emoji usage influences developer participation and issue resolution in virtual workspaces. To this end, we collect GitHub issues for a one-year period and apply causal inference techniques to measure the causal effect of emojis on the outcome of issues, controlling for confounders such as issue content, repository, and author information. We find that emojis can significantly reduce the resolution time of issues and attract more user participation. We also compare the heterogeneous effect on different types of issues. These findings deepen our understanding of the developer communities, and they provide design implications on how to facilitate interactions and broaden developer participation.

Xuan Lu, Wei Ai, Yixin Wang et al.

While many organizations have shifted to working remotely during the COVID-19 pandemic, how the remote workforce and the remote teams are influenced by and would respond to this and future shocks remain largely unknown. Software developers have relied on remote collaborations long before the pandemic, working in virtual teams (GitHub repositories). The dynamics of these repositories through the pandemic provide a unique opportunity to understand how remote teams react under shock. This work presents a systematic analysis. We measure the overall effect of the early pandemic on public GitHub repositories by comparing their sizes and productivity with the counterfactual outcomes forecasted as if there were no pandemic. We find that the productivity level and the number of active members of these teams vary significantly during different periods of the pandemic. We then conduct a finer-grained investigation and study the heterogeneous effects of the shock on individual teams. We find that the resilience of a team is highly correlated to certain properties of the team before the pandemic. Through a bootstrapped regression analysis, we reveal which types of teams are robust or fragile to the shock.

Jin Huang, Xingjian Zhang, Qiaozhu Mei et al.

Large language models (LLMs) are gaining increasing attention for their capability to process graphs with rich text attributes, especially in a zero-shot fashion. Recent studies demonstrate that LLMs obtain decent text classification performance on common text-rich graph benchmarks, and the performance can be improved by appending encoded structural information as natural languages into prompts. We aim to understand why the incorporation of structural information inherent in graph data can improve the prediction performance of LLMs. First, we rule out the concern of data leakage by curating a novel leakage-free dataset and conducting a comparative analysis alongside a previously widely-used dataset. Second, as past work usually encodes the ego-graph by describing the graph structure in natural language, we ask the question: do LLMs understand the graph structure in accordance with the intent of the prompt designers? Third, we investigate why LLMs can improve their performance after incorporating structural information. Our exploration of these questions reveals that (i) there is no substantial evidence that the performance of LLMs is significantly attributed to data leakage; (ii) instead of understanding prompts as graph structures as intended by the prompt designers, LLMs tend to process prompts more as contextual paragraphs and (iii) the most efficient elements of the local neighborhood included in the prompt are phrases that are pertinent to the node label, rather than the graph structure.

Zhuowan Li, Cheng Li, Mingyang Zhang et al.

Retrieval Augmented Generation (RAG) has been a powerful tool for Large Language Models (LLMs) to efficiently process overly lengthy contexts. However, recent LLMs like Gemini-1.5 and GPT-4 show exceptional capabilities to understand long contexts directly. We conduct a comprehensive comparison between RAG and long-context (LC) LLMs, aiming to leverage the strengths of both. We benchmark RAG and LC across various public datasets using three latest LLMs. Results reveal that when resourced sufficiently, LC consistently outperforms RAG in terms of average performance. However, RAG's significantly lower cost remains a distinct advantage. Based on this observation, we propose Self-Route, a simple yet effective method that routes queries to RAG or LC based on model self-reflection. Self-Route significantly reduces the computation cost while maintaining a comparable performance to LC. Our findings provide a guideline for long-context applications of LLMs using RAG and LC.

Cheng Li, Mingyang Zhang, Qiaozhu Mei et al.

Facilitated by large language models (LLMs), personalized text generation has become a rapidly growing research direction. Most existing studies focus on designing specialized models for a particular domain, or they require fine-tuning the LLMs to generate personalized text. We consider a typical scenario in which the large language model, which generates personalized output, is frozen and can only be accessed through APIs. Under this constraint, all one can do is to improve the input text (i.e., text prompts) sent to the LLM, a procedure that is usually done manually. In this paper, we propose a novel method to automatically revise prompts for personalized text generation. The proposed method takes the initial prompts generated by a state-of-the-art, multistage framework for personalized generation and rewrites a few critical components that summarize and synthesize the personal context. The prompt rewriter employs a training paradigm that chains together supervised learning (SL) and reinforcement learning (RL), where SL reduces the search space of RL and RL facilitates end-to-end training of the rewriter. Using datasets from three representative domains, we demonstrate that the rewritten prompts outperform both the original prompts and the prompts optimized via supervised learning or reinforcement learning alone. In-depth analysis of the rewritten prompts shows that they are not only human readable, but also able to guide manual revision of prompts when there is limited resource to employ reinforcement learning to train the prompt rewriter, or when it is costly to deploy an automatic prompt rewriter for inference.

Ting Wei Li, Qiaozhu Mei, Jiaqi Ma

Graph Neural Networks (GNNs) have achieved remarkable success in various applications, but their performance can be sensitive to specific data properties of the graph datasets they operate on. Current literature on understanding the limitations of GNNs has primarily employed a $\textit{model-driven}$ approach that leverage heuristics and domain knowledge from network science or graph theory to model the GNN behaviors, which is time-consuming and highly subjective. In this work, we propose a $\textit{metadata-driven}$ approach to analyze the sensitivity of GNNs to graph data properties, motivated by the increasing availability of graph learning benchmarks. We perform a multivariate sparse regression analysis on the metadata derived from benchmarking GNN performance across diverse datasets, yielding a set of salient data properties. To validate the effectiveness of our data-driven approach, we focus on one identified data property, the degree distribution, and investigate how this property influences GNN performance through theoretical analysis and controlled experiments. Our theoretical findings reveal that datasets with more balanced degree distribution exhibit better linear separability of node representations, thus leading to better GNN performance. We also conduct controlled experiments using synthetic datasets with varying degree distributions, and the results align well with our theoretical findings. Collectively, both the theoretical analysis and controlled experiments verify that the proposed metadata-driven approach is effective in identifying critical data properties for GNNs.

Qiaozhu Mei, Yutong Xie, Walter Yuan et al.

We administer a Turing Test to AI Chatbots. We examine how Chatbots behave in a suite of classic behavioral games that are designed to elicit characteristics such as trust, fairness, risk-aversion, cooperation, \textit{etc.}, as well as how they respond to a traditional Big-5 psychological survey that measures personality traits. ChatGPT-4 exhibits behavioral and personality traits that are statistically indistinguishable from a random human from tens of thousands of human subjects from more than 50 countries. Chatbots also modify their behavior based on previous experience and contexts ``as if'' they were learning from the interactions, and change their behavior in response to different framings of the same strategic situation. Their behaviors are often distinct from average and modal human behaviors, in which case they tend to behave on the more altruistic and cooperative end of the distribution. We estimate that they act as if they are maximizing an average of their own and partner's payoffs.

Yaqing Wang, Jiepu Jiang, Mingyang Zhang et al.

Personalized text generation presents a specialized mechanism for delivering content that is specific to a user's personal context. While the research progress in this area has been rapid, evaluation still presents a challenge. Traditional automated metrics such as BLEU and ROUGE primarily measure lexical similarity to human-written references, and are not able to distinguish personalization from other subtle semantic aspects, thus falling short of capturing the nuances of personalized generated content quality. On the other hand, human judgments are costly to obtain, especially in the realm of personalized evaluation. Inspired by these challenges, we explore the use of large language models (LLMs) for evaluating personalized text generation, and examine their ability to understand nuanced user context. We present AuPEL, a novel evaluation method that distills three major semantic aspects of the generated text: personalization, quality and relevance, and automatically measures these aspects. To validate the effectiveness of AuPEL, we design carefully controlled experiments and compare the accuracy of the evaluation judgments made by LLMs versus that of judgements made by human annotators, and conduct rigorous analyses of the consistency and sensitivity of the proposed metric. We find that, compared to existing evaluation metrics, AuPEL not only distinguishes and ranks models based on their personalization abilities more accurately, but also presents commendable consistency and efficiency for this task. Our work suggests that using LLMs as the evaluators of personalized text generation is superior to traditional text similarity metrics, even though interesting new challenges still remain.

Yachuan Liu, Liang Chen, Jindong Wang et al.

Do large language models (LLMs) genuinely understand the semantics of the language, or just memorize the training data? The recent concern on potential data contamination of LLMs has raised awareness of the community to conduct research on LLMs evaluation. In this paper, we propose MSTemp, an approach that creates meta semantic templates to evaluate the semantic understanding ability of LLMs. The core of MSTemp is not to perform evaluation directly on existing benchmark datasets, but to generate new out-of-distribution (OOD) evaluation sets using existing datasets as seeds. Specifically, for a given sentence, MSTemp leverages another language model to generate new samples while preserving its semantics. The new samples are called semantic templates to the original sentence. Then, MSTemp generates evaluation samples via sentence parsing and random word replacement on the semantic templates. MSTemp is highly flexible, dynamic, and cost-effective. Our initial experiments show that MSTemp-generated samples can significantly reduce the performance of LLMs using existing datasets as seeds. We hope this initial work can shed light on future research of LLMs evaluation.

Jingyi Qiu, Qiaozhu Mei

Major deployed generative AI advertising systems preserve a visible boundary between commercial content and AI-generated responses. Yet empirical research shows that ads woven directly into large language model (LLM) outputs often go undetected by users. We argue that generative AI fundamentally changes advertising: rather than placing products into discrete slots, it enables interventions on the generative process itself, which induce commercial influence through less observable channels. This reframes generative AI advertising as a problem of trustworthy intervention rather than content placement. We introduce a taxonomy organized by influence tier, corresponding to interventions on progressively more latent variables: product mentions, information framing, behavioral redirection, and long-term preference shaping; and show how these tiers instantiate across modalities and system architectures, including retrieval-augmented generation and agentic pipelines where upstream decisions can sharply constrain downstream outcomes. Both major deployed systems and designed mechanisms concentrate on the most observable and easiest-to-govern tier, while the forms of commercial influence most consequential for user autonomy remain poorly understood and lack frameworks for detection, measurement, or disclosure. The central challenge is whether commercial influence in generative systems can be made trustworthy, i.e., attributable, measurable, contestable, and aligned with user welfare.

Yutong Xie, Yijun Pan, Hua Xu et al.

Artificial Intelligence has proven to be a transformative tool for advancing scientific research across a wide range of disciplines. However, a significant gap still exists between AI and scientific communities, limiting the full potential of AI methods in driving broad scientific discovery. Existing efforts in identifying and bridging this gap have often relied on qualitative examination of small samples of literature, offering a limited perspective on the broader AI4Science landscape. In this work, we present a large-scale analysis of the AI4Science literature, starting by using large language models to identify scientific problems and AI methods in publications from top science and AI venues. Leveraging this new dataset, we quantitatively highlight key disparities between AI methods and scientific problems, revealing substantial opportunities for deeper AI integration across scientific disciplines. Furthermore, we explore the potential and challenges of facilitating collaboration between AI and scientific communities through the lens of link prediction. Our findings and tools aim to promote more impactful interdisciplinary collaborations and accelerate scientific discovery through deeper and broader AI integration. Our code and dataset are available at: https://github.com/charles-pyj/Bridging-AI-and-Science.

Yutong Xie, Zhuoheng Li, Xiyuan Wang et al.

Despite their success in numerous fields, the potential of foundation models for modeling and understanding human behavior remains largely unexplored. We introduce Be.FM, one of the first open foundation models designed for human behavior modeling. Built upon open-source large language models and fine-tuned on a diverse range of behavioral data, Be.FM can be used to understand and predict human decision-making. We construct a comprehensive set of benchmark tasks for testing the capabilities of behavioral foundation models. Our results demonstrate that Be.FM can predict behaviors, infer characteristics of individuals and populations, generate insights about contexts, and apply behavioral science knowledge.

Alan Zhu, Jiaqi Ma, Qiaozhu Mei

As large graph datasets become increasingly common across many fields, sampling is often needed to reduce the graphs into manageable sizes. This procedure raises critical questions about representativeness as no sample can capture the properties of the original graph perfectly, and different parts of the graph are not evenly affected by the loss. Recent work has shown that the distances from the non-sampled nodes to the sampled nodes can be a quantitative indicator of bias and fairness in graph machine learning. However, to our knowledge, there is no method for evaluating how a sampling method affects the distribution of shortest-path distances without actually performing the sampling and shortest-path calculation. In this paper, we present an accurate and efficient framework for estimating the distribution of shortest-path distances to the sample, applicable to a wide range of sampling methods and graph structures. Our framework is faster than empirical methods and only requires the specification of degree distributions. We also extend our framework to handle graphs with community structures. While this introduces a decrease in accuracy, we demonstrate that our framework remains highly accurate on downstream comparison-based tasks. Code is publicly available at https://github.com/az1326/shortest_paths.

Xingjian Zhang, Yutong Xie, Jin Huang et al.

Scientific innovation relies on detailed workflows, which include critical steps such as analyzing literature, generating ideas, validating these ideas, interpreting results, and inspiring follow-up research. However, scientific publications that document these workflows are extensive and unstructured. This makes it difficult for both human researchers and AI systems to effectively navigate and explore the space of scientific innovation. To address this issue, we introduce MASSW, a comprehensive text dataset on Multi-Aspect Summarization of Scientific Workflows. MASSW includes more than 152,000 peer-reviewed publications from 17 leading computer science conferences spanning the past 50 years. Using Large Language Models (LLMs), we automatically extract five core aspects from these publications -- context, key idea, method, outcome, and projected impact -- which correspond to five key steps in the research workflow. These structured summaries facilitate a variety of downstream tasks and analyses. The quality of the LLM-extracted summaries is validated by comparing them with human annotations. We demonstrate the utility of MASSW through multiple novel machine-learning tasks that can be benchmarked using this new dataset, which make various types of predictions and recommendations along the scientific workflow. MASSW holds significant potential for researchers to create and benchmark new AI methods for optimizing scientific workflows and fostering scientific innovation in the field. Our dataset is openly available at \url{https://github.com/xingjian-zhang/massw}.

Weijing Tang, Jiaqi Ma, Qiaozhu Mei et al.

In this paper, we propose a flexible model for survival analysis using neural networks along with scalable optimization algorithms. One key technical challenge for directly applying maximum likelihood estimation (MLE) to censored data is that evaluating the objective function and its gradients with respect to model parameters requires the calculation of integrals. To address this challenge, we recognize that the MLE for censored data can be viewed as a differential-equation constrained optimization problem, a novel perspective. Following this connection, we model the distribution of event time through an ordinary differential equation and utilize efficient ODE solvers and adjoint sensitivity analysis to numerically evaluate the likelihood and the gradients. Using this approach, we are able to 1) provide a broad family of continuous-time survival distributions without strong structural assumptions, 2) obtain powerful feature representations using neural networks, and 3) allow efficient estimation of the model in large-scale applications using stochastic gradient descent. Through both simulation studies and real-world data examples, we demonstrate the effectiveness of the proposed method in comparison to existing state-of-the-art deep learning survival analysis models. The implementation of the proposed SODEN approach has been made publicly available at https://github.com/jiaqima/SODEN.

Zixuan Ke, Weize Kong, Cheng Li et al.

Large Language Models (LLMs) have demonstrated superior results across a wide range of tasks, and Retrieval-augmented Generation (RAG) is an effective way to enhance the performance by locating relevant information and placing it into the context window of the LLM. However, the relationship between retrievers and LLMs in a RAG is still under-investigated. Most existing work treats the retriever and the LLM as independent components and leaves a gap between retrieving human-"friendly" information and assembling a LLM-"friendly" context. In this work, we examine a novel bridge mechanism. We validate the ranking and selection assumptions of retrievers in the context of RAG and propose a framework that chains together supervised and reinforcement learning to train a bridge model that optimizes the connection between the retriever and the LLM. Empirical results demonstrate the effectiveness of our method in both question-answering and personalized generation tasks.

Alireza Salemi, Cheng Li, Mingyang Zhang et al.

Personalized text generation requires a unique ability of large language models (LLMs) to learn from context that they often do not encounter during their standard training. One way to encourage LLMs to better use personalized context for generating outputs that better align with the user's expectations is to instruct them to reason over the user's past preferences, background knowledge, or writing style. To achieve this, we propose Reasoning-Enhanced Self-Training for Personalized Text Generation (REST-PG), a framework that trains LLMs to reason over personal data during response generation. REST-PG first generates reasoning paths to train the LLM's reasoning abilities and then employs Expectation-Maximization Reinforced Self-Training to iteratively train the LLM based on its own high-reward outputs. We evaluate REST-PG on the LongLaMP benchmark, consisting of four diverse personalized long-form text generation tasks. Our experiments demonstrate that REST-PG achieves significant improvements over state-of-the-art baselines, with an average relative performance gain of 14.5% on the benchmark.

Yutong Xie, Yiyao Liu, Zhuang Ma et al.

The deployment of large language models (LLMs) in diverse applications requires a thorough understanding of their decision-making strategies and behavioral patterns. As a supplement to a recent study on the behavioral Turing test, this paper presents a comprehensive analysis of five leading LLM-based chatbot families as they navigate a series of behavioral economics games. By benchmarking these AI chatbots, we aim to uncover and document both common and distinct behavioral patterns across a range of scenarios. The findings provide valuable insights into the strategic preferences of each LLM, highlighting potential implications for their deployment in critical decision-making roles.

Teng Ye, Jingnan Zheng, Junhui Jin et al.

While small businesses are increasingly turning to online crowdfunding platforms for essential funding, over 40% of these campaigns may fail to raise any money, especially those from low socio-economic areas. We utilize the latest advancements in AI technology to identify crucial factors that influence the success of crowdfunding campaigns and to improve their fundraising outcomes by strategically optimizing these factors. Our best-performing machine learning model accurately predicts the fundraising outcomes of 81.0% of campaigns, primarily based on their textual descriptions. Interpreting the machine learning model allows us to provide actionable suggestions on improving the textual description before launching a campaign. We demonstrate that by augmenting just three aspects of the narrative using a large language model, a campaign becomes more preferable to 83% human evaluators, and its likelihood of securing financial support increases by 11.9%. Our research uncovers the effective strategies for crafting descriptions for small business fundraising campaigns and opens up a new realm in integrating large language models into crowdfunding methodologies.

Xingjian Zhang, Sophia Moylan, Ziyang Xiong et al.

Scientific knowledge bases accelerate discovery by curating findings from primary literature into structured, queryable formats for both human researchers and emerging AI systems. Maintaining these resources requires expert curators to search relevant papers, reconcile evidence across documents, and produce ontology-grounded annotations - a workflow that existing benchmarks, focused on isolated subtasks like named entity recognition or relation extraction, do not capture. We present FlyBench to evaluate AI agents on end-to-end agentic ontology curation from scientific literature. Given only a gene symbol, agents must search and read from a corpus of 16,898 full-text papers to produce structured annotations: Gene Ontology terms describing function, expression patterns, and historical synonyms linking decades of nomenclature. The benchmark includes 7,397 expert-curated annotations across 100 genes drawn from FlyBase, the Drosophila (fruit fly) knowledge base. We evaluate four baseline agent architectures: memorization, fixed pipeline, single-agent, and multi-agent. We find that architectural choices significantly impact performance, with multi-agent designs outperforming simpler alternatives, yet scaling backbone models yields diminishing returns. All baselines leave substantial room for improvement. Our analysis surfaces several findings to guide future development; for example, agents primarily use retrieval to confirm parametric knowledge rather than discover new information. We hope FlyBench will drive progress on retrieval-augmented scientific reasoning, a capability with broad applications across scientific domains.

Xingjian Zhang, Tianhong Gao, Suliang Jin et al.

Large language models (LLMs) are increasingly used as raters for evaluation tasks. However, their reliability is often limited for subjective tasks, when human judgments involve subtle reasoning beyond annotation labels. Thinking traces, the reasoning behind a judgment, are highly informative but challenging to collect and curate. We present a human-LLM collaborative framework to infer thinking traces from label-only annotations. The proposed framework uses a simple and effective rejection sampling method to reconstruct these traces at scale. These inferred thinking traces are applied to two complementary tasks: (1) fine-tuning open LLM raters; and (2) synthesizing clearer annotation guidelines for proprietary LLM raters. Across multiple datasets, our methods lead to significantly improved LLM-human agreement. Additionally, the refined annotation guidelines increase agreement among different LLM models. These results suggest that LLMs can serve as practical proxies for otherwise unrevealed human thinking traces, enabling label-only corpora to be extended into thinking-trace-augmented resources that enhance the reliability of LLM raters.

Alireza Salemi, Cheng Li, Mingyang Zhang et al. · deepmind, gatech

Personalization is essential for adapting question answering (QA) systems to user-specific information needs, thereby improving both accuracy and user satisfaction. However, personalized QA remains relatively underexplored due to challenges such as inferring preferences from long, noisy, and implicit contexts, and generating responses that are simultaneously correct, contextually appropriate, and aligned with user expectations and background knowledge. To address these challenges, we propose Pathways of Thoughts (PoT), an inference-stage method that applies to any large language model (LLM) without requiring task-specific fine-tuning. The approach models the reasoning of an LLM as an iterative decision process, where the model dynamically selects among cognitive operations such as reasoning, revision, personalization, and clarification. This enables exploration of multiple reasoning trajectories, producing diverse candidate responses that capture different perspectives. PoT then aggregates and reweights these candidates according to inferred user preferences, yielding a final personalized response that benefits from the complementary strengths of diverse reasoning paths. Experiments on the LaMP-QA benchmark for personalized QA show that PoT consistently outperforms competitive baselines, achieving up to a 13.1% relative improvement. Human evaluation corroborates these results, with annotators preferring outputs from PoT in 66% of cases and reporting ties in only 15% of cases.

Yachuan Liu, Xiaochun Wei, Lin Shi et al.

Large language models (LLMs) face significant challenges in ex-ante reasoning, where analysis, inference, or predictions must be made without access to information from future events. Even with explicit prompts enforcing temporal cutoffs, LLMs often generate outputs influenced by internalized knowledge of events beyond the specified cutoff. This paper introduces a novel task and benchmark designed to evaluate the ability of LLMs to reason while adhering to such temporal constraints. The benchmark includes a variety of tasks: stock prediction, Wikipedia event prediction, scientific publication prediction, and Question Answering (QA), designed to assess factual knowledge under temporal cutoff constraints. We use leakage rate to quantify models' reliance on future information beyond cutoff timestamps. Experimental results reveal that LLMs struggle to consistently adhere to temporal cutoffs across common prompting strategies and tasks, demonstrating persistent challenges in ex-ante reasoning. This benchmark provides a potential evaluation framework to advance the development of LLMs' temporal reasoning ability for time-sensitive applications.

Yutong Xie, Qiaozhu Mei, Walter Yuan et al.

By varying prompts to a large language model, we can elicit the full range of human behaviors in a variety of different scenarios in classic economic games. By analyzing which prompts elicit which behaviors, we can categorize and compare different strategic situations, which can also help provide insight into what different economic scenarios induce people to think about. We discuss how this provides a first step towards a non-standard method of inferring (deciphering) the motivations behind the human behaviors. We also show how this deciphering process can be used to categorize differences in the behavioral tendencies of different populations.

Hua Shen, Tiffany Knearem, Reshmi Ghosh et al.

Recent advances in general-purpose AI underscore the urgent need to align AI systems with human goals and values. Yet, the lack of a clear, shared understanding of what constitutes "alignment" limits meaningful progress and cross-disciplinary collaboration. In this position paper, we argue that the research community should explicitly define and critically reflect on "alignment" to account for the bidirectional and dynamic relationship between humans and AI. Through a systematic review of over 400 papers spanning HCI, NLP, ML, and more, we examine how alignment is currently defined and operationalized. Building on this analysis, we introduce the Bidirectional Human-AI Alignment framework, which not only incorporates traditional efforts to align AI with human values but also introduces the critical, underexplored dimension of aligning humans with AI -- supporting cognitive, behavioral, and societal adaptation to rapidly advancing AI technologies. Our findings reveal significant gaps in current literature, especially in long-term interaction design, human value modeling, and mutual understanding. We conclude with three central challenges and actionable recommendations to guide future research toward more nuanced, reciprocal, and human-AI alignment approaches.

Weize Kong, Spurthi Amba Hombaiah, Mingyang Zhang et al.

Prompt engineering is critical for the development of LLM-based applications. However, it is usually done manually in a "trial and error" fashion that can be time consuming, ineffective, and sub-optimal. Even for the prompts which seemingly work well, there is always a lingering question: can the prompts be made better with further modifications? To address these problems, we investigate automated prompt engineering in this paper. Specifically, we propose PRewrite, an automated method to rewrite an under-optimized prompt to a more effective prompt. We instantiate the prompt rewriter using a LLM. The rewriter LLM is trained using reinforcement learning to optimize the performance on a given downstream task. We conduct experiments on diverse benchmark datasets, which demonstrates the effectiveness of PRewrite.

Yachuan Liu, Bohan Zhang, Qiaozhu Mei et al.

Recent work has shown that standard training via empirical risk minimization (ERM) can produce models that achieve high accuracy on average but low accuracy on underrepresented groups due to the prevalence of spurious features. A predominant approach to tackle this group robustness problem minimizes the worst group error (akin to a minimax strategy) on the training data, hoping it will generalize well on the testing data. However, this is often suboptimal, especially when the out-of-distribution (OOD) test data contains previously unseen groups. Inspired by ideas from the information retrieval and learning-to-rank literature, this paper first proposes to use Discounted Cumulative Gain (DCG) as a metric of model quality for facilitating better hyperparameter tuning and model selection. Being a ranking-based metric, DCG weights multiple poorly-performing groups (instead of considering just the group with the worst performance). As a natural next step, we build on our results to propose a ranking-based training method called Discounted Rank Upweighting (DRU), which differentially reweights a ranked list of poorly-performing groups in the training data to learn models that exhibit strong OOD performance on the test data. Results on several synthetic and real-world datasets highlight the superior generalization ability of our group-ranking-based (akin to soft-minimax) approach in selecting and learning models that are robust to group distributional shifts.

Jiaqi Ma, Ziqiao Ma, Joyce Chai et al.

We study the problem of semi-supervised learning with Graph Neural Networks (GNNs) in an active learning setup. We propose GraphPart, a novel partition-based active learning approach for GNNs. GraphPart first splits the graph into disjoint partitions and then selects representative nodes within each partition to query. The proposed method is motivated by a novel analysis of the classification error under realistic smoothness assumptions over the graph and the node features. Extensive experiments on multiple benchmark datasets demonstrate that the proposed method outperforms existing active learning methods for GNNs under a wide range of annotation budget constraints. In addition, the proposed method does not introduce additional hyperparameters, which is crucial for model training, especially in the active learning setting where a labeled validation set may not be available.

Jiaqi Ma, Xingjian Zhang, Qiaozhu Mei

Multinomial Logit (MNL) is one of the most popular discrete choice models and has been widely used to model ranking data. However, there is a long-standing technical challenge of learning MNL from many real-world ranking data: exact calculation of the MNL likelihood of \emph{partial rankings} is generally intractable. In this work, we develop a scalable method for approximating the MNL likelihood of general partial rankings in polynomial time complexity. We also extend the proposed method to learn mixture of MNL. We demonstrate that the proposed methods are particularly helpful for applications to choice-based network formation modeling, where the formation of new edges in a network is viewed as individuals making choices of their friends over a candidate set. The problem of learning mixture of MNL models from partial rankings naturally arises in such applications. And the proposed methods can be used to learn MNL models from network data without the strong assumption that temporal orders of all the edge formation are available. We conduct experiments on both synthetic and real-world network data to demonstrate that the proposed methods achieve more accurate parameter estimation and better fitness of data compared to conventional methods.

Yutong Xie, Ziqiao Xu, Jiaqi Ma et al.

Forming a molecular candidate set that contains a wide range of potentially effective compounds is crucial to the success of drug discovery. While most databases and machine-learning-based generation models aim to optimize particular chemical properties, there is limited literature on how to properly measure the coverage of the chemical space by those candidates included or generated. This problem is challenging due to the lack of formal criteria to select good measures of the chemical space. In this paper, we propose a novel evaluation framework for measures of the chemical space based on two analyses: an axiomatic analysis with three intuitive axioms that a good measure should obey, and an empirical analysis on the correlation between a measure and a proxy gold standard. Using this framework, we are able to identify #Circles, a new measure of chemical space coverage, which is superior to existing measures both analytically and empirically. We further evaluate how well the existing databases and generation models cover the chemical space in terms of #Circles. The results suggest that many generation models fail to explore a larger space over existing databases, which leads to new opportunities for improving generation models by encouraging exploration.

Jiaqi Ma, Junwei Deng, Qiaozhu Mei

Despite enormous successful applications of graph neural networks (GNNs), theoretical understanding of their generalization ability, especially for node-level tasks where data are not independent and identically-distributed (IID), has been sparse. The theoretical investigation of the generalization performance is beneficial for understanding fundamental issues (such as fairness) of GNN models and designing better learning methods. In this paper, we present a novel PAC-Bayesian analysis for GNNs under a non-IID semi-supervised learning setup. Moreover, we analyze the generalization performances on different subgroups of unlabeled nodes, which allows us to further study an accuracy-(dis)parity-style (un)fairness of GNNs from a theoretical perspective. Under reasonable assumptions, we demonstrate that the distance between a test subgroup and the training set can be a key factor affecting the GNN performance on that subgroup, which calls special attention to the training node selection for fair learning. Experiments across multiple GNN models and datasets support our theoretical results.

Jiaqi Ma, Junwei Deng, Qiaozhu Mei

Graph neural networks (GNNs) have attracted increasing interests. With broad deployments of GNNs in real-world applications, there is an urgent need for understanding the robustness of GNNs under adversarial attacks, especially in realistic setups. In this work, we study the problem of attacking GNNs in a restricted and realistic setup, by perturbing the features of a small set of nodes, with no access to model parameters and model predictions. Our formal analysis draws a connection between this type of attacks and an influence maximization problem on the graph. This connection not only enhances our understanding on the problem of adversarial attack on GNNs, but also allows us to propose a group of effective and practical attack strategies. Our experiments verify that the proposed attack strategies significantly degrade the performance of three popular GNN models and outperform baseline adversarial attack strategies.

Xuan Lu, Wei Ai, Zhenpeng Chen et al.

Emotions at work have long been identified as critical signals of work motivations, status, and attitudes, and as predictors of various work-related outcomes. When more and more employees work remotely, these emotional signals of workers become harder to observe through daily, face-to-face communications. The use of online platforms to communicate and collaborate at work provides an alternative channel to monitor the emotions of workers. This paper studies how emojis, as non-verbal cues in online communications, can be used for such purposes and how the emotional signals in emoji usage can be used to predict future behavior of workers. In particular, we present how the developers on GitHub use emojis in their work-related activities. We show that developers have diverse patterns of emoji usage, which can be related to their working status including activity levels, types of work, types of communications, time management, and other behavioral patterns. Developers who use emojis in their posts are significantly less likely to dropout from the online work platform. Surprisingly, solely using emoji usage as features, standard machine learning models can predict future dropouts of developers at a satisfactory accuracy. Features related to the general use and the emotions of emojis appear to be important factors, while they do not rule out paths through other purposes of emoji use.

Jiaqi Ma, Bo Chang, Xuefei Zhang et al.

Graph-structured data are ubiquitous. However, graphs encode diverse types of information and thus play different roles in data representation. In this paper, we distinguish the \textit{representational} and the \textit{correlational} roles played by the graphs in node-level prediction tasks, and we investigate how Graph Neural Network (GNN) models can effectively leverage both types of information. Conceptually, the representational information provides guidance for the model to construct better node features; while the correlational information indicates the correlation between node outcomes conditional on node features. Through a simulation study, we find that many popular GNN models are incapable of effectively utilizing the correlational information. By leveraging the idea of the copula, a principled way to describe the dependence among multivariate random variables, we offer a general solution. The proposed Copula Graph Neural Network (CopulaGNN) can take a wide range of GNN models as base models and utilize both representational and correlational information stored in the graphs. Experimental results on two types of regression tasks verify the effectiveness of the proposed method.

Teng Ye, Wei Ai, Lingyu Zhang et al.

Millions of drivers worldwide have enjoyed financial benefits and work schedule flexibility through a ride-sharing economy, but meanwhile they have suffered from the lack of a sense of identity and career achievement. Equipped with social identity and contest theories, financially incentivized team competitions have been an effective instrument to increase drivers' productivity, job satisfaction, and retention, and to improve revenue over cost for ride-sharing platforms. While these competitions are overall effective, the decisive factors behind the treatment effects and how they affect the outcomes of individual drivers have been largely mysterious. In this study, we analyze data collected from more than 500 large-scale team competitions organized by a leading ride-sharing platform, building machine learning models to predict individual treatment effects. Through a careful investigation of features and predictors, we are able to reduce out-sample prediction error by more than 24%. Through interpreting the best-performing models, we discover many novel and actionable insights regarding how to optimize the design and the execution of team competitions on ride-sharing platforms. A simulated analysis demonstrates that by simply changing a few contest design options, the average treatment effect of a real competition is expected to increase by as much as 26%. Our procedure and findings shed light on how to analyze and optimize large-scale online field experiments in general.

Cristina Garbacea, Mengtian Guo, Samuel Carton et al.

Text simplification reduces the language complexity of professional content for accessibility purposes. End-to-end neural network models have been widely adopted to directly generate the simplified version of input text, usually functioning as a blackbox. We show that text simplification can be decomposed into a compact pipeline of tasks to ensure the transparency and explainability of the process. The first two steps in this pipeline are often neglected: 1) to predict whether a given piece of text needs to be simplified, and 2) if yes, to identify complex parts of the text. The two tasks can be solved separately using either lexical or deep learning methods, or solved jointly. Simply applying explainable complexity prediction as a preliminary step, the out-of-sample text simplification performance of the state-of-the-art, black-box simplification models can be improved by a large margin.

Cristina Garbacea, Qiaozhu Mei

Recent advances in neural network-based generative modeling have reignited the hopes in having computer systems capable of seamlessly conversing with humans and able to understand natural language. Neural architectures have been employed to generate text excerpts to various degrees of success, in a multitude of contexts and tasks that fulfil various user needs. Notably, high capacity deep learning models trained on large scale datasets demonstrate unparalleled abilities to learn patterns in the data even in the lack of explicit supervision signals, opening up a plethora of new possibilities regarding producing realistic and coherent texts. While the field of natural language generation is evolving rapidly, there are still many open challenges to address. In this survey we formally define and categorize the problem of natural language generation. We review particular application tasks that are instantiations of these general formulations, in which generating natural language is of practical importance. Next we include a comprehensive outline of methods and neural architectures employed for generating diverse texts. Nevertheless, there is no standard way to assess the quality of text produced by these generative models, which constitutes a serious bottleneck towards the progress of the field. To this end, we also review current approaches to evaluating natural language generation systems. We hope this survey will provide an informative overview of formulations, methods, and assessments of neural natural language generation.

Jiaqi Ma, Xinyang Yi, Weijing Tang et al.

We investigate the Plackett-Luce (PL) model based listwise learning-to-rank (LTR) on data with partitioned preference, where a set of items are sliced into ordered and disjoint partitions, but the ranking of items within a partition is unknown. Given $N$ items with $M$ partitions, calculating the likelihood of data with partitioned preference under the PL model has a time complexity of $O(N+S!)$, where $S$ is the maximum size of the top $M-1$ partitions. This computational challenge restrains most existing PL-based listwise LTR methods to a special case of partitioned preference, top-$K$ ranking, where the exact order of the top $K$ items is known. In this paper, we exploit a random utility model formulation of the PL model, and propose an efficient numerical integration approach for calculating the likelihood and its gradients with a time complexity $O(N+S^3)$. We demonstrate that the proposed method outperforms well-known LTR baselines and remains scalable through both simulation experiments and applications to real-world eXtreme Multi-Label classification tasks.

Jiaqi Ma, Shuangrui Ding, Qiaozhu Mei

We study the black-box attacks on graph neural networks (GNNs) under a novel and realistic constraint: attackers have access to only a subset of nodes in the network, and they can only attack a small number of them. A node selection step is essential under this setup. We demonstrate that the structural inductive biases of GNN models can be an effective source for this type of attacks. Specifically, by exploiting the connection between the backward propagation of GNNs and random walks, we show that the common gradient-based white-box attacks can be generalized to the black-box setting via the connection between the gradient and an importance score similar to PageRank. In practice, we find attacks based on this importance score indeed increase the classification loss by a large margin, but they fail to significantly increase the mis-classification rate. Our theoretical and empirical analyses suggest that there is a discrepancy between the loss and mis-classification rate, as the latter presents a diminishing-return pattern when the number of attacked nodes increases. Therefore, we propose a greedy procedure to correct the importance score that takes into account of the diminishing-return pattern. Experimental results show that the proposed procedure can significantly increase the mis-classification rate of common GNNs on real-world data without access to model parameters nor predictions.

Jiaqi Ma, Qiaozhu Mei

Machine learning on graph structured data has attracted much research interest due to its ubiquity in real world data. However, how to efficiently represent graph data in a general way is still an open problem. Traditional methods use handcraft graph features in a tabular form but suffer from the defects of domain expertise requirement and information loss. Graph representation learning overcomes these defects by automatically learning the continuous representations from graph structures, but they require abundant training labels, which are often hard to fulfill for graph-level prediction problems. In this work, we demonstrate that, if available, the domain expertise used for designing handcraft graph features can improve the graph-level representation learning when training labels are scarce. Specifically, we proposed a multi-task knowledge distillation method. By incorporating network-theory-based graph metrics as auxiliary tasks, we show on both synthetic and real datasets that the proposed multi-task learning method can improve the prediction performance of the original learning task, especially when the training data size is small.

Zhenpeng Chen, Yanbin Cao, Xuan Lu et al.

Sentiment analysis has various application scenarios in software engineering (SE), such as detecting developers' emotions in commit messages and identifying their opinions on Q&A forums. However, commonly used out-of-the-box sentiment analysis tools cannot obtain reliable results on SE tasks and the misunderstanding of technical jargon is demonstrated to be the main reason. Then, researchers have to utilize labeled SE-related texts to customize sentiment analysis for SE tasks via a variety of algorithms. However, the scarce labeled data can cover only very limited expressions and thus cannot guarantee the analysis quality. To address such a problem, we turn to the easily available emoji usage data for help. More specifically, we employ emotional emojis as noisy labels of sentiments and propose a representation learning approach that uses both Tweets and GitHub posts containing emojis to learn sentiment-aware representations for SE-related texts. These emoji-labeled posts can not only supply the technical jargon, but also incorporate more general sentiment patterns shared across domains. They as well as labeled data are used to learn the final sentiment classifier. Compared to the existing sentiment analysis methods used in SE, the proposed approach can achieve significant improvement on representative benchmark datasets. By further contrast experiments, we find that the Tweets make a key contribution to the power of our approach. This finding informs future research not to unilaterally pursue the domain-specific resource, but try to transform knowledge from the open domain through ubiquitous signals such as emojis.

Jiaqi Ma, Weijing Tang, Ji Zhu et al.

We consider a family of problems that are concerned about making predictions for the majority of unlabeled, graph-structured data samples based on a small proportion of labeled samples. Relational information among the data samples, often encoded in the graph/network structure, is shown to be helpful for these semi-supervised learning tasks. However, conventional graph-based regularization methods and recent graph neural networks do not fully leverage the interrelations between the features, the graph, and the labels. In this work, we propose a flexible generative framework for graph-based semi-supervised learning, which approaches the joint distribution of the node features, labels, and the graph structure. Borrowing insights from random graph models in network science literature, this joint distribution can be instantiated using various distribution families. For the inference of missing labels, we exploit recent advances of scalable variational inference techniques to approximate the Bayesian posterior. We conduct thorough experiments on benchmark datasets for graph-based semi-supervised learning. Results show that the proposed methods outperform the state-of-the-art models in most settings.

Cristina Garbacea, Samuel Carton, Shiyan Yan et al.

We conduct a large-scale, systematic study to evaluate the existing evaluation methods for natural language generation in the context of generating online product reviews. We compare human-based evaluators with a variety of automated evaluation procedures, including discriminative evaluators that measure how well machine-generated text can be distinguished from human-written text, as well as word overlap metrics that assess how similar the generated text compares to human-written references. We determine to what extent these different evaluators agree on the ranking of a dozen of state-of-the-art generators for online product reviews. We find that human evaluators do not correlate well with discriminative evaluators, leaving a bigger question of whether adversarial accuracy is the correct objective for natural language generation. In general, distinguishing machine-generated text is challenging even for human evaluators, and human decisions correlate better with lexical overlaps. We find lexical diversity an intriguing metric that is indicative of the assessments of different evaluators. A post-experiment survey of participants provides insights into how to evaluate and improve the quality of natural language generation systems.

Samuel Carton, Qiaozhu Mei, Paul Resnick

We introduce an adversarial method for producing high-recall explanations of neural text classifier decisions. Building on an existing architecture for extractive explanations via hard attention, we add an adversarial layer which scans the residual of the attention for remaining predictive signal. Motivated by the important domain of detecting personal attacks in social media comments, we additionally demonstrate the importance of manually setting a semantically appropriate `default' behavior for the model by explicitly manipulating its bias term. We develop a validation set of human-annotated personal attacks to evaluate the impact of these changes.