arXiv subjects

Xian Li

Publications and source records attributed to Xian Li.

At least 73 records · Page 4Linked to original sources

arXiv2025

SWEET-RL: Training Multi-Turn LLM Agents on Collaborative Reasoning Tasks

Large language model (LLM) agents need to perform multi-turn interactions in real-world tasks. However, existing multi-turn RL algorithms for optimizing LLM agents fail to perform effective credit assignment over multiple turns while leveraging the generalization capabilities of LLMs and it remains unclear how to develop such algorithms. To study this, we first introduce a new benchmark, ColBench, where an LLM agent interacts with a human collaborator over multiple turns to solve realistic tasks in backend programming and frontend design. Building on this benchmark, we propose a novel RL algorithm, SWEET-RL (RL with Step-WisE Evaluation from Training-time information), that uses a carefully designed optimization objective to train a critic model with access to additional training-time information. The critic provides step-level rewards for improving the policy model. Our experiments demonstrate that SWEET-RL achieves a 6% absolute improvement in success and win rates on ColBench compared to other state-of-the-art multi-turn RL algorithms, enabling Llama-3.1-8B to match or exceed the performance of GPT4-o in realistic collaborative content creation.

cs.LG↗

arXiv2025

Subbagging Variable Selection for Big Data

This article introduces a subbagging (subsample aggregating) approach for variable selection in regression within the context of big data. The proposed subbagging approach not only ensures that variable selection is scalable given the constraints of available computational resources, but also preserves the statistical efficiency of the resulting estimator. In particular, we propose a subbagging loss function that aggregates the least-squares approximations of the loss function for each subsample. Subsequently, we penalize the subbagging loss function via an adaptive LASSO-type regularizer, and obtain a regularized estimator to achieve variable selection. We then demonstrate that the regularized estimator exhibits $\sqrt{N}$-consistency and possesses the oracle properties, where $N$ represents the size of the full sample in the big data. In addition, we propose a subbagging Bayesian information criterion to select the regularization parameter, ensuring that the regularized estimator achieves selection consistency. Simulation experiments are conducted to demonstrate the numerical performance. A U.S. census dataset is analyzed to illustrate the usefulness and computational scalability of the subbagging variable selection method.

stat.ME↗

arXiv2025

On Robust Aggregation for Distributed Data

When data are stored across multiple locations, directly pooling all the data together for statistical analysis may be impossible due to communication costs and privacy concerns. Distributed computing systems allow the analysis of such data, by getting local servers to separately process their own statistical analyses and using a central processor to aggregate the local statistical results. Naive aggregation of local statistics using simple or weighted averages, is vulnerable to contamination within a distributed computing system. This paper develops and investigates a Huber-type aggregation method for locally computed M-estimators to handle contamination in the local estimates. Our implementation of this aggregation method requires estimating the asymptotic variance-covariance matrix of the M-estimator, which we accomplish using a robust spatial median approach. Theoretically, the Huber-type aggregation achieves the same convergence rate as if all the data were pooled. We establish its asymptotic normality for making inferences, including justifying a two-step approach for detecting contamination in the distributed computing system. Extensive simulation studies are conducted to validate the theoretical results and the usefulness of our proposed approach is demonstrated on U.S. airline data.

stat.ME↗

arXiv2025

BIG-AOME: Designing Bodily Interaction Gamification towards Anti-sedentary Online Meeting Environments

Online meetings have become an integral part of daily life, but prolonged screen time poses significant health risks. While various interventions address sedentary lifestyles, few focus on mitigating sedentary behavior during online meetings. Design opportunities in this context remain underexplored. This study investigates the design of gamified bodily interactions as anti-sedentary measures during online meetings using a research through design approach. In collaboration with 11 users, we co-designed and iterated three prototypes, resulting in the BIG-AOME (Bodily Interaction Gamification towards Anti-sedentary Online Meeting Environments) framework. User studies with 15 participants across three groups evaluated these prototypes through semi-structured interviews analyzed using Hsieh's qualitative content analysis. Findings show that gamified bodily interactions encourage physical movement while reducing awkwardness during online meetings. Participants valued the social engagement fostered by cooperative and competitive elements in these games, enhancing social dynamics while encouraging physical movement. Such games can also serve as online icebreakers or playful decision-making tools. This study offers a comprehensive analysis of design dimensions within the BIG-AOME framework, including body engagement, attention, bodily interplay, timeliness, and virtual/physical environments, highlighting the potential of anti-sedentary bodily interactions to mitigate sedentary behavior and enhance social connections in online meetings.

cs.HC↗

arXiv2025

LLM Pretraining with Continuous Concepts

Next token prediction has been the standard training objective used in large language model pretraining. Representations are learned as a result of optimizing for token-level perplexity. We propose Continuous Concept Mixing (CoCoMix), a novel pretraining framework that combines discrete next token prediction with continuous concepts. Specifically, CoCoMix predicts continuous concepts learned from a pretrained sparse autoencoder and mixes them into the model's hidden state by interleaving with token hidden representations. Through experiments on multiple benchmarks, including language modeling and downstream reasoning tasks, we show that CoCoMix is more sample efficient and consistently outperforms standard next token prediction, knowledge distillation and inserting pause tokens. We find that combining both concept learning and interleaving in an end-to-end framework is critical to performance gains. Furthermore, CoCoMix enhances interpretability and steerability by allowing direct inspection and modification of the predicted concept, offering a transparent way to guide the model's internal reasoning process.

cs.LG↗

arXiv2025

LPRnet: A self-supervised registration network for LiDAR and photogrammetric point clouds

LiDAR and photogrammetry are active and passive remote sensing techniques for point cloud acquisition, respectively, offering complementary advantages and heterogeneous. Due to the fundamental differences in sensing mechanisms, spatial distributions and coordinate systems, their point clouds exhibit significant discrepancies in density, precision, noise, and overlap. Coupled with the lack of ground truth for large-scale scenes, integrating the heterogeneous point clouds is a highly challenging task. This paper proposes a self-supervised registration network based on a masked autoencoder, focusing on heterogeneous LiDAR and photogrammetric point clouds. At its core, the method introduces a multi-scale masked training strategy to extract robust features from heterogeneous point clouds under self-supervision. To further enhance registration performance, a rotation-translation embedding module is designed to effectively capture the key features essential for accurate rigid transformations. Building upon the robust representations, a transformer-based architecture seamlessly integrates local and global features, fostering precise alignment across diverse point cloud datasets. The proposed method demonstrates strong feature extraction capabilities for both LiDAR and photogrammetric point clouds, addressing the challenges of acquiring ground truth at the scene level. Experiments conducted on two real-world datasets validate the effectiveness of the proposed method in solving heterogeneous point cloud registration problems.

cs.CV↗

arXiv2025

Your Fix Is My Exploit: Enabling Comprehensive DL Library API Fuzzing with Large Language Models

Deep learning (DL) libraries, widely used in AI applications, often contain vulnerabilities like buffer overflows and use-after-free errors. Traditional fuzzing struggles with the complexity and API diversity of DL libraries such as TensorFlow and PyTorch, which feature over 1,000 APIs. Testing all these APIs is challenging due to complex inputs and varied usage patterns. While large language models (LLMs) show promise in code understanding and generation, existing LLM-based fuzzers lack deep knowledge of API edge cases and struggle with test input generation. To address this, we propose DFUZZ, an LLM-driven fuzzing approach for DL libraries. DFUZZ leverages two insights: (1) LLMs can reason about error-triggering edge cases from API code and apply this knowledge to untested APIs, and (2) LLMs can accurately synthesize test programs to automate API testing. By providing LLMs with a "white-box view" of APIs, DFUZZ enhances reasoning and generation for comprehensive fuzzing. Experimental results show that DFUZZ outperforms state-of-the-art fuzzers in API coverage for TensorFlow and PyTorch, uncovering 37 bugs, with 8 fixed and 19 under developer investigation.

cs.SE↗

arXiv2024

An Enhanced Classification Method Based on Adaptive Multi-Scale Fusion for Long-tailed Multispectral Point Clouds

Multispectral point cloud (MPC) captures 3D spatial-spectral information from the observed scene, which can be used for scene understanding and has a wide range of applications. However, most of the existing classification methods were extensively tested on indoor datasets, and when applied to outdoor datasets they still face problems including sparse labeled targets, differences in land-covers scales, and long-tailed distributions. To address the above issues, an enhanced classification method based on adaptive multi-scale fusion for MPCs with long-tailed distributions is proposed. In the training set generation stage, a grid-balanced sampling strategy is designed to reliably generate training samples from sparse labeled datasets. In the feature learning stage, a multi-scale feature fusion module is proposed to fuse shallow features of land-covers at different scales, addressing the issue of losing fine features due to scale variations in land-covers. In the classification stage, an adaptive hybrid loss module is devised to utilize multi-classification heads with adaptive weights to balance the learning ability of different classes, improving the classification performance of small classes due to various-scales and long-tailed distributions in land-covers. Experimental results on three MPC datasets demonstrate the effectiveness of the proposed method compared with the state-of-the-art methods.

cs.CV↗

arXiv2024

Towards Full Delegation: Designing Ideal Agentic Behaviors for Travel Planning

How are LLM-based agents used in the future? While many of the existing work on agents has focused on improving the performance of a specific family of objective and challenging tasks, in this work, we take a different perspective by thinking about full delegation: agents take over humans' routine decision-making processes and are trusted by humans to find solutions that fit people's personalized needs and are adaptive to ever-changing context. In order to achieve such a goal, the behavior of the agents, i.e., agentic behaviors, should be evaluated not only on their achievements (i.e., outcome evaluation), but also how they achieved that (i.e., procedure evaluation). For this, we propose APEC Agent Constitution, a list of criteria that an agent should follow for good agentic behaviors, including Accuracy, Proactivity, Efficiency and Credibility. To verify whether APEC aligns with human preferences, we develop APEC-Travel, a travel planning agent that proactively extracts hidden personalized needs via multi-round dialog with travelers. APEC-Travel is constructed purely from synthetic data generated by Llama3.1-405B-Instruct with a diverse set of travelers' persona to simulate rich distribution of dialogs. Iteratively fine-tuned to follow APEC Agent Constitution, APEC-Travel surpasses baselines by 20.7% on rule-based metrics and 9.1% on LLM-as-a-Judge scores across the constitution axes.

cs.CL↗

arXiv2024

Data Diversity Matters for Robust Instruction Tuning

Recent works have shown that by curating high quality and diverse instruction tuning datasets, we can significantly improve instruction-following capabilities. However, creating such datasets is difficult and most works rely on manual curation or proprietary language models. Automatic data curation is difficult as it is still not clear how we can define diversity for instruction tuning, how diversity and quality depend on one other, and how we can optimize dataset quality and diversity. To resolve these issue, we propose a new algorithm, Quality-Diversity Instruction Tuning (QDIT). QDIT provides a simple method to simultaneously control dataset diversity and quality, allowing us to conduct an in-depth study on the effect of diversity and quality on instruction tuning performance. From this study we draw two key insights (1) there is a natural tradeoff between data diversity and quality and (2) increasing data diversity significantly improves the worst case instruction following performance, therefore improving robustness. We validate the performance of QDIT on several large scale instruction tuning datasets, where we find it can substantially improve worst and average case performance compared to quality-driven data selection.

cs.CL↗

arXiv2024

Shopping MMLU: A Massive Multi-Task Online Shopping Benchmark for Large Language Models

Online shopping is a complex multi-task, few-shot learning problem with a wide and evolving range of entities, relations, and tasks. However, existing models and benchmarks are commonly tailored to specific tasks, falling short of capturing the full complexity of online shopping. Large Language Models (LLMs), with their multi-task and few-shot learning abilities, have the potential to profoundly transform online shopping by alleviating task-specific engineering efforts and by providing users with interactive conversations. Despite the potential, LLMs face unique challenges in online shopping, such as domain-specific concepts, implicit knowledge, and heterogeneous user behaviors. Motivated by the potential and challenges, we propose Shopping MMLU, a diverse multi-task online shopping benchmark derived from real-world Amazon data. Shopping MMLU consists of 57 tasks covering 4 major shopping skills: concept understanding, knowledge reasoning, user behavior alignment, and multi-linguality, and can thus comprehensively evaluate the abilities of LLMs as general shop assistants. With Shopping MMLU, we benchmark over 20 existing LLMs and uncover valuable insights about practices and prospects of building versatile LLM-based shop assistants. Shopping MMLU can be publicly accessed at https://github.com/KL4805/ShoppingMMLU. In addition, with Shopping MMLU, we host a competition in KDD Cup 2024 with over 500 participating teams. The winning solutions and the associated workshop can be accessed at our website https://amazon-kddcup24.github.io/.

cs.LG↗

arXiv2024

To the Globe (TTG): Towards Language-Driven Guaranteed Travel Planning

Travel planning is a challenging and time-consuming task that aims to find an itinerary which satisfies multiple, interdependent constraints regarding flights, accommodations, attractions, and other travel arrangements. In this paper, we propose To the Globe (TTG), a real-time demo system that takes natural language requests from users, translates it to symbolic form via a fine-tuned Large Language Model, and produces optimal travel itineraries with Mixed Integer Linear Programming solvers. The overall system takes ~5 seconds to reply to the user request with guaranteed itineraries. To train TTG, we develop a synthetic data pipeline that generates user requests, flight and hotel information in symbolic form without human annotations, based on the statistics of real-world datasets, and fine-tune an LLM to translate NL user requests to their symbolic form, which is sent to the symbolic solver to compute optimal itineraries. Our NL-symbolic translation achieves ~91% exact match in a backtranslation metric (i.e., whether the estimated symbolic form of generated natural language matches the groundtruth), and its returned itineraries have a ratio of 0.979 compared to the optimal cost of the ground truth user request. When evaluated by users, TTG achieves consistently high Net Promoter Scores (NPS) of 35-40% on generated itinerary.

cs.CL↗

arXiv2024

Multi-head Sequence Tagging Model for Grammatical Error Correction

To solve the Grammatical Error Correction (GEC) problem , a mapping between a source sequence and a target one is needed, where the two differ only on few spans. For this reason, the attention has been shifted to the non-autoregressive or sequence tagging models. In which, the GEC has been simplified from Seq2Seq to labeling the input tokens with edit commands chosen from a large edit space. Due to this large number of classes and the limitation of the available datasets, the current sequence tagging approaches still have some issues handling a broad range of grammatical errors just by being laser-focused on one single task. To this end, we simplified the GEC further by dividing it into seven related subtasks: Insertion, Deletion, Merge, Substitution, Transformation, Detection, and Correction, with Correction being our primary focus. A distinct classification head is dedicated to each of these subtasks. the novel multi-head and multi-task learning model is proposed to effectively utilize training data and harness the information from related task training signals. To mitigate the limited number of available training samples, a new denoising autoencoder is used to generate a new synthetic dataset to be used for pretraining. Additionally, a new character-level transformation is proposed to enhance the sequence-to-edit function and improve the model's vocabulary coverage. Our single/ensemble model achieves an F0.5 of 74.4/77.0, and 68.6/69.1 on BEA-19 (test) and CoNLL-14 (test) respectively. Moreover, evaluated on JFLEG test set, the GLEU scores are 61.6 and 61.7 for the single and ensemble models, respectively. It mostly outperforms recently published state-of-the-art results by a considerable margin.

cs.CL↗

arXiv2024

RNR: Teaching Large Language Models to Follow Roles and Rules

Instruction fine-tuning (IFT) elicits instruction following capabilities and steers the behavior of large language models (LLMs) via supervised learning. However, existing models trained on open-source IFT datasets only have the ability to follow instructions from users, and often fail to follow complex role and rules specified by developers, a.k.a. system prompts. The ability to follow these roles and rules is essential for deployment, as it ensures that the model safely interacts with users within developer defined guidelines. To improve such role and rule following ability, we propose \model, an automated data generation pipeline that generates diverse roles and rules from existing IFT instructions, along with corresponding responses. This data can then be used to train models that follow complex system prompts. The models are evaluated on our newly created benchmarks for role and rule following ability, as well as standard instruction-following benchmarks and general NLP tasks. Our framework significantly improves role and rule following capability in LLMs, as evidenced by over 25% increase in pass-rate on rule adherence, i.e. following all requirements, in our experiments with the Alpaca and Ultrachat datasets. Moreover, our models achieves this increase without any regression on popular instruction following benchmarks.

cs.CL↗

arXiv2024

Better Alignment with Instruction Back-and-Forth Translation

We propose a new method, instruction back-and-forth translation, to construct high-quality synthetic data grounded in world knowledge for aligning large language models (LLMs). Given documents from a web corpus, we generate and curate synthetic instructions using the backtranslation approach proposed by Li et al.(2023a), and rewrite the responses to improve their quality further based on the initial documents. Fine-tuning with the resulting (backtranslated instruction, rewritten response) pairs yields higher win rates on AlpacaEval than using other common instruction datasets such as Humpback, ShareGPT, Open Orca, Alpaca-GPT4 and Self-instruct. We also demonstrate that rewriting the responses with an LLM outperforms direct distillation, and the two generated text distributions exhibit significant distinction in embedding space. Further analysis shows that our backtranslated instructions are of higher quality than other sources of synthetic instructions, while our responses are more diverse and complex than those obtained from distillation. Overall we find that instruction back-and-forth translation combines the best of both worlds -- making use of the information diversity and quantity found on the web, while ensuring the quality of the responses which is necessary for effective alignment.

cs.CL↗

arXiv2024

Self-Taught Evaluators

Model-based evaluation is at the heart of successful model development -- as a reward model for training, and as a replacement for human evaluation. To train such evaluators, the standard approach is to collect a large amount of human preference judgments over model responses, which is costly and the data becomes stale as models improve. In this work, we present an approach that aims to im-prove evaluators without human annotations, using synthetic training data only. Starting from unlabeled instructions, our iterative self-improvement scheme generates contrasting model outputs and trains an LLM-as-a-Judge to produce reasoning traces and final judgments, repeating this training at each new iteration using the improved predictions. Without any labeled preference data, our Self-Taught Evaluator can improve a strong LLM (Llama3-70B-Instruct) from 75.4 to 88.3 (88.7 with majority vote) on RewardBench. This outperforms commonly used LLM judges such as GPT-4 and matches the performance of the top-performing reward models trained with labeled examples.

cs.CL↗

arXiv2024

Inductive or Deductive? Rethinking the Fundamental Reasoning Abilities of LLMs

Reasoning encompasses two typical types: deductive reasoning and inductive reasoning. Despite extensive research into the reasoning capabilities of Large Language Models (LLMs), most studies have failed to rigorously differentiate between inductive and deductive reasoning, leading to a blending of the two. This raises an essential question: In LLM reasoning, which poses a greater challenge - deductive or inductive reasoning? While the deductive reasoning capabilities of LLMs, (i.e. their capacity to follow instructions in reasoning tasks), have received considerable attention, their abilities in true inductive reasoning remain largely unexplored. To investigate into the true inductive reasoning capabilities of LLMs, we propose a novel framework, SolverLearner. This framework enables LLMs to learn the underlying function (i.e., $y = f_w(x)$), that maps input data points $(x)$ to their corresponding output values $(y)$, using only in-context examples. By focusing on inductive reasoning and separating it from LLM-based deductive reasoning, we can isolate and investigate inductive reasoning of LLMs in its pure form via SolverLearner. Our observations reveal that LLMs demonstrate remarkable inductive reasoning capabilities through SolverLearner, achieving near-perfect performance with ACC of 1 in most cases. Surprisingly, despite their strong inductive reasoning abilities, LLMs tend to relatively lack deductive reasoning capabilities, particularly in tasks involving ``counterfactual'' reasoning.

cs.AI↗

arXiv2024

FlowGPT: Exploring Domains, Output Modalities, and Goals of Community-Generated AI Chatbots

The advent of Generative AI and Large Language Models has not only enhanced the intelligence of interactive applications but also catalyzed the formation of communities passionate about customizing these AI capabilities. FlowGPT, an emerging platform for sharing AI prompts and use cases, exemplifies this trend, attracting many creators who develop and share chatbots with a broader community. Despite its growing popularity, there remains a significant gap in understanding the types and purposes of the AI tools created and shared by community members. In this study, we delve into FlowGPT and present our preliminary findings on the domain, output modality, and goals of chatbots. We aim to highlight common types of AI applications and identify future directions for research in AI-sharing communities.

cs.HC↗