Jiayuan Ding

Wei Jin, Tong Zhao, Jiayuan Ding et al.

As powerful tools for representation learning on graphs, graph neural networks (GNNs) have facilitated various applications from drug discovery to recommender systems. Nevertheless, the effectiveness of GNNs is immensely challenged by issues related to data quality, such as distribution shift, abnormal features and adversarial attacks. Recent efforts have been made on tackling these issues from a modeling perspective which requires additional cost of changing model architectures or re-training model parameters. In this work, we provide a data-centric view to tackle these issues and propose a graph transformation framework named GTrans which adapts and refines graph data at test time to achieve better performance. We provide theoretical analysis on the design of the framework and discuss why adapting graph data works better than adapting the model. Extensive experiments have demonstrated the effectiveness of GTrans on three distinct scenarios for eight benchmark datasets where suboptimal data is presented. Remarkably, GTrans performs the best in most cases with improvements up to 2.8%, 8.2% and 3.8% over the best baselines on three experimental settings. Code is released at https://github.com/ChandlerBang/GTrans.

Hongzhi Wen, Jiayuan Ding, Wei Jin et al.

Recent advances in multimodal single-cell technologies have enabled simultaneous acquisitions of multiple omics data from the same cell, providing deeper insights into cellular states and dynamics. However, it is challenging to learn the joint representations from the multimodal data, model the relationship between modalities, and, more importantly, incorporate the vast amount of single-modality datasets into the downstream analyses. To address these challenges and correspondingly facilitate multimodal single-cell data analyses, three key tasks have been introduced: $\textit{modality prediction}$, $\textit{modality matching}$ and $\textit{joint embedding}$. In this work, we present a general Graph Neural Network framework $\textit{scMoGNN}$ to tackle these three tasks and show that $\textit{scMoGNN}$ demonstrates superior results in all three tasks compared with the state-of-the-art and conventional approaches. Our method is an official winner in the overall ranking of $\textit{Modality prediction}$ from NeurIPS 2021 Competition, and all implementations of our methods have been integrated into DANCE package~\url{https://github.com/OmicsML/dance}.

Juanhui Li, Harry Shomer, Jiayuan Ding et al.

Knowledge graphs (KGs) facilitate a wide variety of applications. Despite great efforts in creation and maintenance, even the largest KGs are far from complete. Hence, KG completion (KGC) has become one of the most crucial tasks for KG research. Recently, considerable literature in this space has centered around the use of Message Passing (Graph) Neural Networks (MPNNs), to learn powerful embeddings. The success of these methods is naturally attributed to the use of MPNNs over simpler multi-layer perceptron (MLP) models, given their additional message passing (MP) component. In this work, we find that surprisingly, simple MLP models are able to achieve comparable performance to MPNNs, suggesting that MP may not be as crucial as previously believed. With further exploration, we show careful scoring function and loss function design has a much stronger influence on KGC model performance. This suggests a conflation of scoring function design, loss function design, and MP in prior work, with promising insights regarding the scalability of state-of-the-art KGC methods today, as well as careful attention to more suitable MP designs for KGC tasks tomorrow. Our codes are publicly available at: https://github.com/Juanhui28/Are_MPNNs_helpful.

Yiyou Sun, Xinyang Han, Weichen Zhang et al.

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long-horizon, economically valuable, real-world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 subfields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is 2.6%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP-relevant impact.

Dylan Molho, Jiayuan Ding, Zhaoheng Li et al.

Single-cell technologies are revolutionizing the entire field of biology. The large volumes of data generated by single-cell technologies are high-dimensional, sparse, heterogeneous, and have complicated dependency structures, making analyses using conventional machine learning approaches challenging and impractical. In tackling these challenges, deep learning often demonstrates superior performance compared to traditional machine learning methods. In this work, we give a comprehensive survey on deep learning in single-cell analysis. We first introduce background on single-cell technologies and their development, as well as fundamental concepts of deep learning including the most popular deep architectures. We present an overview of the single-cell analytic pipeline pursued in research applications while noting divergences due to data sources or specific applications. We then review seven popular tasks spanning through different stages of the single-cell analysis pipeline, including multimodal integration, imputation, clustering, spatial domain identification, cell-type deconvolution, cell segmentation, and cell-type annotation. Under each task, we describe the most recent developments in classical and deep learning methods and discuss their advantages and disadvantages. Deep learning tools and benchmark datasets are also summarized for each task. Finally, we discuss the future directions and the most recent challenges. This survey will serve as a reference for biologists and computer scientists, encouraging collaborations.

Hongzhi Wen, Wenzhuo Tang, Wei Jin et al.

Spatially resolved transcriptomics brings exciting breakthroughs to single-cell analysis by providing physical locations along with gene expression. However, as a cost of the extremely high spatial resolution, the cellular level spatial transcriptomic data suffer significantly from missing values. While a standard solution is to perform imputation on the missing values, most existing methods either overlook spatial information or only incorporate localized spatial context without the ability to capture long-range spatial information. Using multi-head self-attention mechanisms and positional encoding, transformer models can readily grasp the relationship between tokens and encode location information. In this paper, by treating single cells as spatial tokens, we study how to leverage transformers to facilitate spatial tanscriptomics imputation. In particular, investigate the following two key questions: (1) $\textit{how to encode spatial information of cells in transformers}$, and (2) $\textit{ how to train a transformer for transcriptomic imputation}$. By answering these two questions, we present a transformer-based imputation framework, SpaFormer, for cellular-level spatial transcriptomic data. Extensive experiments demonstrate that SpaFormer outperforms existing state-of-the-art imputation algorithms on three large-scale datasets while maintaining superior computational efficiency.

Wenzhuo Tang, Hongzhi Wen, Renming Liu et al.

The recent development of multimodal single-cell technology has made the possibility of acquiring multiple omics data from individual cells, thereby enabling a deeper understanding of cellular states and dynamics. Nevertheless, the proliferation of multimodal single-cell data also introduces tremendous challenges in modeling the complex interactions among different modalities. The recently advanced methods focus on constructing static interaction graphs and applying graph neural networks (GNNs) to learn from multimodal data. However, such static graphs can be suboptimal as they do not take advantage of the downstream task information; meanwhile GNNs also have some inherent limitations when deeply stacking GNN layers. To tackle these issues, in this work, we investigate how to leverage transformers for multimodal single-cell data in an end-to-end manner while exploiting downstream task information. In particular, we propose a scMoFormer framework which can readily incorporate external domain knowledge and model the interactions within each modality and cross modalities. Extensive experiments demonstrate that scMoFormer achieves superior performance on various benchmark datasets. Remarkably, scMoFormer won a Kaggle silver medal with the rank of 24/1221 (Top 2%) without ensemble in a NeurIPS 2022 competition. Our implementation is publicly available at Github.

Binxu Li, Tiankai Yan, Yuanting Pan et al.

Multi-Modal Large Language Models (MLLMs), despite being successful, exhibit limited generality and often fall short when compared to specialized models. Recently, LLM-based agents have been developed to address these challenges by selecting appropriate specialized models as tools based on user inputs. However, such advancements have not been extensively explored within the medical domain. To bridge this gap, this paper introduces the first agent explicitly designed for the medical field, named \textbf{M}ulti-modal \textbf{Med}ical \textbf{Agent} (MMedAgent). We curate an instruction-tuning dataset comprising six medical tools solving seven tasks across five modalities, enabling the agent to choose the most suitable tools for a given task. Comprehensive experiments demonstrate that MMedAgent achieves superior performance across a variety of medical tasks compared to state-of-the-art open-source methods and even the closed-source model, GPT-4o. Furthermore, MMedAgent exhibits efficiency in updating and integrating new medical tools. Codes and models are all available.

Haoyu Han, Yu Wang, Harry Shomer et al.

Retrieval-augmented generation (RAG) is a powerful technique that enhances downstream task execution by retrieving additional information, such as knowledge, skills, and tools from external sources. Graph, by its intrinsic "nodes connected by edges" nature, encodes massive heterogeneous and relational information, making it a golden resource for RAG in tremendous real-world applications. As a result, we have recently witnessed increasing attention on equipping RAG with Graph, i.e., GraphRAG. However, unlike conventional RAG, where the retriever, generator, and external data sources can be uniformly designed in the neural-embedding space, the uniqueness of graph-structured data, such as diverse-formatted and domain-specific relational knowledge, poses unique and significant challenges when designing GraphRAG for different domains. Given the broad applicability, the associated design challenges, and the recent surge in GraphRAG, a systematic and up-to-date survey of its key concepts and techniques is urgently desired. Following this motivation, we present a comprehensive and up-to-date survey on GraphRAG. Our survey first proposes a holistic GraphRAG framework by defining its key components, including query processor, retriever, organizer, generator, and data source. Furthermore, recognizing that graphs in different domains exhibit distinct relational patterns and require dedicated designs, we review GraphRAG techniques uniquely tailored to each domain. Finally, we discuss research challenges and brainstorm directions to inspire cross-disciplinary opportunities. Our survey repository is publicly maintained at https://github.com/Graph-RAG/GraphRAG/.

Yuping Lin, Jiayuan Ding, Yue Xing et al.

KV cache growth is a major bottleneck for long-context inference in large language models. Existing methods are often dominated by binary eviction or representation approximation, which may underutilize tokens that are not critical for exact retention but are still reconstructable. We present VECTOR, a plug-and-play augmentation for eviction-based pipelines that introduces three-way token routing: retention, approximation, and eviction. VECTOR combines an importance signal from the base scorer with a reconstructability signal from an offline-calibrated regression-based value estimation. By leveraging reconstructability, VECTOR recovers useful value information that would otherwise be irreversibly lost under binary eviction, while preserving key vectors for attention routing stability. Experimental results show that VECTOR improves quality-memory trade-offs under medium-to-high compression, with especially clear gains in stricter budget regimes.

Hongchao Fang, Sicheng Wang, Meng Zhou et al.

Pretrained language models such as BERT, GPT have shown great effectiveness in language understanding. The auxiliary predictive tasks in existing pretraining approaches are mostly defined on tokens, thus may not be able to capture sentence-level semantics very well. To address this issue, we propose CERT: Contrastive self-supervised Encoder Representations from Transformers, which pretrains language representation models using contrastive self-supervised learning at the sentence level. CERT creates augmentations of original sentences using back-translation. Then it finetunes a pretrained language encoder (e.g., BERT) by predicting whether two augmented sentences originate from the same sentence. CERT is simple to use and can be flexibly plugged into any pretraining-finetuning NLP pipeline. We evaluate CERT on 11 natural language understanding tasks in the GLUE benchmark where CERT outperforms BERT on 7 tasks, achieves the same performance as BERT on 2 tasks, and performs worse than BERT on 2 tasks. On the averaged score of the 11 tasks, CERT outperforms BERT. The data and code are available at https://github.com/UCSD-AI4H/CERT

Mayank Kejriwal, Jiayuan Ding, Runqi Shao et al.

In this paper, we describe and study the indicator mining problem in the online sex advertising domain. We present an in-development system, FlagIt (Flexible and adaptive generation of Indicators from text), which combines the benefits of both a lightweight expert system and classical semi-supervision (heuristic re-labeling) with recently released state-of-the-art unsupervised text embeddings to tag millions of sentences with indicators that are highly correlated with human trafficking. The FlagIt technology stack is open source. On preliminary evaluations involving five indicators, FlagIt illustrates promising performance compared to several alternatives. The system is being actively developed, refined and integrated into a domain-specific search system used by over 200 law enforcement agencies to combat human trafficking, and is being aggressively extended to mine at least six more indicators with minimal programming effort. FlagIt is a good example of a system that operates in limited label settings, and that requires creative combinations of established machine learning techniques to produce outputs that could be used by real-world non-technical analysts.

Yuping Lin, Pengfei He, Yue Xing et al.

Supervised fine-tuning (SFT) induces new behaviors in large language models, yet imposes no structural constraint on how these behaviors are distributed within the model. Existing behavior interpretation methods, such as circuit attribution approaches, identify sparse subnetworks correlated with SFT-induced behaviors post-hoc. However, such correlations do not imply *causal necessity*, limiting the ability to selectively control SFT-induced behaviors at inference time. We pursue an alternative by asking: can an SFT-induced behavior be deliberately compressed into a sparse, mechanistically necessary subnetwork, termed a *carrier*, while remaining controllable at inference time without weight modification? We propose (a) **Loss-Constrained Dual Descent (LCDD)**, which constructs such carriers by jointly optimizing routing masks and model weights under an explicit utility budget, and (b) **SFT-Eraser**, a soft prompt optimized via activation matching on extracted carrier channels, to reverse the SFT-induced behavior. Across safety, fixed-response, and style behaviors on multiple model families, LCDD yields sparse carriers that preserve target behaviors while enabling strong reversion when triggered by SFT-Eraser. Ablations further establish that the sparse structure is the key precondition for reversal: the same trigger optimization fails on standard SFT models, confirming that structure rather than trigger design is the operative factor. These results provide direct evidence that the learned carriers are causally necessary for the behaviors, pointing to a new direction for systematically localizing and selectively suppressing SFT-induced behaviors in deployed models.

Jie Ren, Han Xu, Pengfei He et al.

Generative AI has witnessed rapid advancement in recent years, expanding their capabilities to create synthesized content such as text, images, audio, and code. The high fidelity and authenticity of contents generated by these Deep Generative Models (DGMs) have sparked significant copyright concerns. There have been various legal debates on how to effectively safeguard copyrights in DGMs. This work delves into this issue by providing a comprehensive overview of copyright protection from a technical perspective. We examine from two distinct viewpoints: the copyrights pertaining to the source data held by the data owners and those of the generative models maintained by the model builders. For data copyright, we delve into methods data owners can protect their content and DGMs can be utilized without infringing upon these rights. For model copyright, our discussion extends to strategies for preventing model theft and identifying outputs generated by specific models. Finally, we highlight the limitations of existing techniques and identify areas that remain unexplored. Furthermore, we discuss prospective directions for the future of copyright protection, underscoring its importance for the sustainable and ethical development of Generative AI.

Subhabrata Mukherjee, Paul Gamble, Markel Sanz Ausin et al.

We develop Polaris, the first safety-focused LLM constellation for real-time patient-AI healthcare conversations. Unlike prior LLM works in healthcare focusing on tasks like question answering, our work specifically focuses on long multi-turn voice conversations. Our one-trillion parameter constellation system is composed of several multibillion parameter LLMs as co-operative agents: a stateful primary agent that focuses on driving an engaging conversation and several specialist support agents focused on healthcare tasks performed by nurses to increase safety and reduce hallucinations. We develop a sophisticated training protocol for iterative co-training of the agents that optimize for diverse objectives. We train our models on proprietary data, clinical care plans, healthcare regulatory documents, medical manuals, and other medical reasoning documents. We align our models to speak like medical professionals, using organic healthcare conversations and simulated ones between patient actors and experienced nurses. This allows our system to express unique capabilities such as rapport building, trust building, empathy and bedside manner. Finally, we present the first comprehensive clinician evaluation of an LLM system for healthcare. We recruited over 1100 U.S. licensed nurses and over 130 U.S. licensed physicians to perform end-to-end conversational evaluations of our system by posing as patients and rating the system on several measures. We demonstrate Polaris performs on par with human nurses on aggregate across dimensions such as medical safety, clinical readiness, conversational quality, and bedside manner. Additionally, we conduct a challenging task-based evaluation of the individual specialist support agents, where we demonstrate our LLM agents significantly outperform a much larger general-purpose LLM (GPT-4) as well as from its own medium-size class (LLaMA-2 70B).

Subhabrata Mukherjee, Markel Sanz Ausin, Kriti Aggarwal et al.

Healthcare conversational AI agents shouldn't be optimized only for clean benchmark accuracy in production-first regime; they must be optimized for the lived reality of patient conversations, where audio is imperfect, intent is indirect, language shifts mid-call, and compliance hinges on how guidance is delivered. We present a production-validated framework grounded in real-time signals from 115M+ live patient-AI interactions and clinician-led testing (7K+ licensed clinicians; 500K+ test calls). These in-the-wild cues -- paralinguistics, turn-taking dynamics, clarification triggers, escalation markers, multilingual continuity, and workflow confirmations -- reveal failure modes that curated data misses and provide actionable training and evaluation signals for safety and reliability. We further show why healthcare-grade safety cannot rely on a single LLM: long-horizon dialogue and limited attention demand redundancy via governed orchestration, independent checks, and verification. Many apparent "reasoning" errors originate upstream, motivating vertical integration across contextual ASR, clarification/repair, ambient speech handling, and latency-aware model/hardware choices. Treating interaction intelligence (tone, pacing, empathy, clarification, turn-taking) as first-class safety variables, we drive measurable gains in safety, documentation, task completion, and equity in building the safest generative AI solution for autonomous patient-facing care. Deployed across more than 10 million real patient calls, Polaris attains a clinical safety score of 99.9%, while significantly improving patient experience with average patient rating of 8.95 and reducing ASR errors by 50% over enterprise ASR. These results establish real-world interaction intelligence as a critical -- and previously underexplored -- determinant of safety and reliability in patient-facing clinical AI systems.

Xinnan Dai, Kai Guo, Chung-Hsiang Lo et al.

Large Language Models (LLMs) demonstrate remarkable reasoning capabilities, yet the structural mechanisms underlying these abilities remain under explored. In this work, we introduce GraphGhost, a unified framework that represents neuron activations and their signal propagation as graphs, explaining how LLMs capture structural semantics from sequential inputs and generate outputs through structurally consistent mechanisms. This graph-based perspective enables us to employ graph algorithms such as PageRank to characterize the properties of LLMs, revealing both shared and model-specific reasoning behaviors across diverse datasets. We further identify the activated neurons within GraphGhost and evaluate them through structural interventions, showing that edits to key neuron nodes can trigger reasoning collapse, altering both logical flow and semantic understanding. Together, these contributions position GraphGhost as a powerful tool for analyzing, intervening in, and ultimately understanding the structural foundations of reasoning in LLMs.

Pengfei He, Zhenwei Dai, Bing He et al.

Large language model (LLM)-based agents increasingly rely on tool use to complete real-world tasks. While existing works evaluate the LLMs' tool use capability, they largely focus on the final answers yet overlook the detailed tool usage trajectory, i.e., whether tools are selected, parameterized, and ordered correctly. We introduce TRAJECT-Bench, a trajectory-aware benchmark to comprehensively evaluate LLMs' tool use capability through diverse tasks with fine-grained evaluation metrics. TRAJECT-Bench pairs high-fidelity, executable tools across practical domains with tasks grounded in production-style APIs, and synthesizes trajectories that vary in breadth (parallel calls) and depth (interdependent chains). Besides final accuracy, TRAJECT-Bench also reports trajectory-level diagnostics, including tool selection and argument correctness, and dependency/order satisfaction. Analyses reveal failure modes such as similar tool confusion and parameter-blind selection, and scaling behavior with tool diversity and trajectory length where the bottleneck of transiting from short to mid-length trajectories is revealed, offering actionable guidance for LLMs' tool use.

Jiayuan Ding, Tong Xiang, Zijing Ou et al.

Recent years have witnessed the dramatic growth of paper volumes with plenty of new research papers published every day, especially in the area of computer science. How to glean papers worth reading from the massive literature to do a quick survey or keep up with the latest advancement about a specific research topic has become a challenging task. Existing academic search engines such as Google Scholar return relevant papers by individually calculating the relevance between each paper and query. However, such systems usually omit the prerequisite chains of a research topic and cannot form a meaningful reading path. In this paper, we introduce a new task named Reading Path Generation (RPG) which aims at automatically producing a path of papers to read for a given query. To serve as a research benchmark, we further propose SurveyBank, a dataset consisting of large quantities of survey papers in the field of computer science as well as their citation relationships. Each survey paper contains key phrases extracted from its title and multi-level reading lists inferred from its references. Furthermore, we propose a graph-optimization-based approach for reading path generation which takes the relationship between papers into account. Extensive evaluations demonstrate that our approach outperforms other baselines. A Real-time Reading Path Generation System (RePaGer) has been also implemented with our designed model. To the best of our knowledge, we are the first to target this important research problem. Our source code of RePaGer system and SurveyBank dataset can be found on here.

Jiayuan Ding, Mayank Kejriwal

Emotion Cause Extraction (ECE) aims to identify emotion causes from a document after annotating the emotion keywords. Some baselines have been proposed to address this problem, such as rule-based, commonsense based and machine learning methods. We show, however, that a simple random selection approach toward ECE that does not require observing the text achieves similar performance compared to the baselines. We utilized only position information relative to the emotion cause to accomplish this goal. Since position information alone without observing the text resulted in higher F-measure, we therefore uncovered a bias in the ECE single genre Sina-news benchmark. Further analysis showed that an imbalance of emotional cause location exists in the benchmark, with a majority of cause clauses immediately preceding the central emotion clause. We examine the bias from a linguistic perspective, and show that high accuracy rate of current state-of-art deep learning models that utilize location information is only evident in datasets that contain such position biases. The accuracy drastically reduced when a dataset with balanced location distribution is introduced. We therefore conclude that it is the innate bias in this benchmark that caused high accuracy rate of these deep learning models in ECE. We hope that the case study in this paper presents both a cautionary lesson, as well as a template for further studies, in interpreting the superior fit of deep learning models without checking for bias.