arXiv subjects

Yu Yang

Publications and source records attributed to Yu Yang.

At least 127 records · Page 7Linked to original sources

arXiv2025

Mini-batch Coresets for Memory-efficient Language Model Training on Data Mixtures

Training with larger mini-batches improves the convergence rate and can yield superior performance. However, training with large mini-batches becomes prohibitive for Large Language Models (LLMs), due to the large GPU memory requirement. To address this problem, an effective approach is finding small mini-batch coresets that closely match the gradient of larger mini-batches. However, this approach becomes infeasible and ineffective for LLMs, due to the highly imbalanced mixture of sources in language data, use of the Adam optimizer, and the very large gradient dimensionality of LLMs. In this work, we address the above challenges by proposing Coresets for Training LLMs (CoLM). First, we show that mini-batch coresets found by gradient matching do not contain representative examples of the small sources w.h.p., and thus including all examples of the small sources in the mini-batch coresets is crucial for optimal performance. Second, we normalize the gradients by their historical exponential to find mini-batch coresets for training with Adam. Finally, we leverage zeroth-order methods to find smooth gradient of the last V-projection matrix and sparsify it to keep the dimensions with the largest normalized gradient magnitude. We apply CoLM to fine-tuning Phi-2, Phi-3, Zephyr, and Llama-3 models with LoRA on MathInstruct and SuperGLUE benchmark. Remarkably, CoLM reduces the memory requirement of fine-tuning by 2x and even outperforms training with 4x larger mini-batches. Moreover, CoLM seamlessly integrates with existing memory-efficient training methods like LoRA, further reducing the memory requirements of training LLMs. Our code is available at https://github.com/BigML-CS-UCLA/CoLM.

cs.LG↗

arXiv2025

LLM-Guided Reinforcement Learning: Addressing Training Bottlenecks through Policy Modulation

While reinforcement learning (RL) has achieved notable success in various domains, training effective policies for complex tasks remains challenging. Agents often converge to local optima and fail to maximize long-term rewards. Existing approaches to mitigate training bottlenecks typically fall into two categories: (i) Automated policy refinement, which identifies critical states from past trajectories to guide policy updates, but suffers from costly and uncertain model training; and (ii) Human-in-the-loop refinement, where human feedback is used to correct agent behavior, but this does not scale well to environments with large or continuous action spaces. In this work, we design a large language model-guided policy modulation framework that leverages LLMs to improve RL training without additional model training or human intervention. We first prompt an LLM to identify critical states from a sub-optimal agent's trajectories. Based on these states, the LLM then provides action suggestions and assigns implicit rewards to guide policy refinement. Experiments across standard RL benchmarks demonstrate that our method outperforms state-of-the-art baselines, highlighting the effectiveness of LLM-based explanations in addressing RL training bottlenecks.

cs.AI↗

arXiv2025

O$^2$-Searcher: A Searching-based Agent Model for Open-Domain Open-Ended Question Answering

Large Language Models (LLMs), despite their advancements, are fundamentally limited by their static parametric knowledge, hindering performance on tasks requiring open-domain up-to-date information. While enabling LLMs to interact with external knowledge environments is a promising solution, current efforts primarily address closed-end problems. Open-ended questions, which characterized by lacking a standard answer or providing non-unique and diverse answers, remain underexplored. To bridge this gap, we present O$^2$-Searcher, a novel search agent leveraging reinforcement learning to effectively tackle both open-ended and closed-ended questions in the open domain. O$^2$-Searcher leverages an efficient, locally simulated search environment for dynamic knowledge acquisition, effectively decoupling the external world knowledge from model's sophisticated reasoning processes. It employs a unified training mechanism with meticulously designed reward functions, enabling the agent to identify problem types and adapt different answer generation strategies. Furthermore, to evaluate performance on complex open-ended tasks, we construct O$^2$-QA, a high-quality benchmark featuring 300 manually curated, multi-domain open-ended questions with associated web page caches. Extensive experiments show that O$^2$-Searcher, using only a 3B model, significantly surpasses leading LLM agents on O$^2$-QA. It also achieves SOTA results on various closed-ended QA benchmarks against similarly-sized models, while performing on par with much larger ones.

cs.CL↗

arXiv2025

Predicting Student Dropout Risk With A Dual-Modal Abrupt Behavioral Changes Approach

Timely prediction of students at high risk of dropout is critical for early intervention and improving educational outcomes. However, in offline educational settings, poor data quality, limited scale, and high heterogeneity often hinder the application of advanced machine learning models. Furthermore, while educational theories provide valuable insights into dropout phenomena, the lack of quantifiable metrics for key indicators limits their use in data-driven modeling. Through data analysis and a review of educational literature, we identified abrupt changes in student behavior as key early signals of dropout risk. To address this, we propose the Dual-Modal Multiscale Sliding Window (DMSW) Model, which integrates academic performance and behavioral data to dynamically capture behavior patterns using minimal data. The DMSW model improves prediction accuracy by 15% compared to traditional methods, enabling educators to identify high-risk students earlier, provide timely support, and foster a more inclusive learning environment. Our analysis highlights key behavior patterns, offering practical insights for preventive strategies and tailored support. These findings bridge the gap between theory and practice in dropout prediction, giving educators an innovative tool to enhance student retention and outcomes.

cs.AI↗

arXiv2025

CIS: Composable Instruction Set for Data Streaming Applications

The enhanced efficiency of hardware accelerators, including Single Instruction Multiple Data (SIMD) architectures and Coarse-Grained Reconfigurable Architectures (CGRAs), is driving significant advancements in Artificial Intelligence and Machine Learning (AI/ML) applications. These applications frequently involve data streaming operations comprised of numerous vector calculations inherently amenable to parallelization. However, despite considerable progress in hardware accelerator design, their potential remains constrained by conventional instruction set architectures (ISAs). Traditional ISAs, primarily designed for microprocessors and accelerators, emphasize computation while often neglecting instruction composability and inter-instruction cooperation. This limitation results in rigid ISAs that are difficult to extend and suffer from large control overhead in their hardware implementations. To address this, we present a novel composable instruction set (CIS) architecture, designed with both spatial and temporal composability, making it well-suited for data streaming applications. The proposed CIS utilizes a small instruction set, yet efficiently implements complex, multi-level loop structures essential for accelerating data streaming workloads. Furthermore, CIS adopts a resource-centric approach, facilitating straightforward extension through the integration of new hardware resources, enabling the creation of custom, heterogeneous computing platforms. Our results comparing performance between the proposed CIS and other state-of-the-art ISAs demonstrate that a CIS-based architecture significantly outperforms existing solutions, achieving near-optimal processing element (PE) utilization.

cs.AR↗

arXiv2025

Challenges and Opportunities in Improving Worst-Group Generalization in Presence of Spurious Features

Deep neural networks often exploit *spurious* features that are present in the majority of examples within a class during training. This leads to *poor worst-group test accuracy*, i.e., poor accuracy for minority groups that lack these spurious features. Despite the growing body of recent efforts to address spurious correlations (SC), several challenging settings remain unexplored.In this work, we propose studying methods to mitigate SC in settings with: 1) spurious features that are learned more slowly, 2) a larger number of classes, and 3) a larger number of groups. We introduce two new datasets, Animals and SUN, to facilitate this study and conduct a systematic benchmarking of 8 state-of-the-art (SOTA) methods across a total of 5 vision datasets, training over 5,000 models. Through this, we highlight how existing group inference methods struggle in the presence of spurious features that are learned later in training. Additionally, we demonstrate how all existing methods struggle in settings with more groups and/or classes. Finally, we show the importance of careful model selection (hyperparameter tuning) in extracting optimal performance, especially in the more challenging settings we introduced, and propose more cost-efficient strategies for model selection. Overall, through extensive and systematic experiments, this work uncovers a suite of new challenges and opportunities for improving worst-group generalization in the presence of spurious features. Our datasets, methods and scripts available at https://github.com/BigML-CS-UCLA/SpuCo.

cs.LG↗

arXiv2025

GLINT-RU: Gated Lightweight Intelligent Recurrent Units for Sequential Recommender Systems

Transformer-based models have gained significant traction in sequential recommender systems (SRSs) for their ability to capture user-item interactions effectively. However, these models often suffer from high computational costs and slow inference. Meanwhile, existing efficient SRS approaches struggle to embed high-quality semantic and positional information into latent representations. To tackle these challenges, this paper introduces GLINT-RU, a lightweight and efficient SRS leveraging a single-layer dense selective Gated Recurrent Units (GRU) module to accelerate inference. By incorporating a dense selective gate, GLINT-RU adaptively captures temporal dependencies and fine-grained positional information, generating high-quality latent representations. Additionally, a parallel mixing block infuses fine-grained positional features into user-item interactions, enhancing both recommendation quality and efficiency. Extensive experiments on three datasets demonstrate that GLINT-RU achieves superior prediction accuracy and inference speed, outperforming baselines based on RNNs, Transformers, MLPs, and SSMs. These results establish GLINT-RU as a powerful and efficient solution for SRSs.

cs.IR↗

arXiv2025

AutoRedTeamer: Autonomous Red Teaming with Lifelong Attack Integration

As large language models (LLMs) become increasingly capable, security and safety evaluation are crucial. While current red teaming approaches have made strides in assessing LLM vulnerabilities, they often rely heavily on human input and lack comprehensive coverage of emerging attack vectors. This paper introduces AutoRedTeamer, a novel framework for fully automated, end-to-end red teaming against LLMs. AutoRedTeamer combines a multi-agent architecture with a memory-guided attack selection mechanism to enable continuous discovery and integration of new attack vectors. The dual-agent framework consists of a red teaming agent that can operate from high-level risk categories alone to generate and execute test cases and a strategy proposer agent that autonomously discovers and implements new attacks by analyzing recent research. This modular design allows AutoRedTeamer to adapt to emerging threats while maintaining strong performance on existing attack vectors. We demonstrate AutoRedTeamer's effectiveness across diverse evaluation settings, achieving 20% higher attack success rates on HarmBench against Llama-3.1-70B while reducing computational costs by 46% compared to existing approaches. AutoRedTeamer also matches the diversity of human-curated benchmarks in generating test cases, providing a comprehensive, scalable, and continuously evolving framework for evaluating the security of AI systems.

cs.CR↗

arXiv2025

Dynamic Load Balancing for EV Charging Stations Using Reinforcement Learning and Demand Prediction

This paper presents a method for load balancing and dynamic pricing in electric vehicle (EV) charging networks, utilizing reinforcement learning (RL) to enhance network performance. The proposed framework integrates a pre-trained graph neural network to predict demand elasticity and inform pricing decisions. The spatio-temporal EV charging demand prediction (EVCDP) dataset from Shenzhen is utilized to capture the geographic and temporal characteristics of the charging stations. The RL model dynamically adjusts prices at individual stations based on occupancy, maximum station capacity, and demand forecasts, ensuring an equitable network load distribution while preventing station overloads. By leveraging spatially-aware demand predictions and a carefully designed reward function, the framework achieves efficient load balancing and adaptive pricing strategies that respond to localized demand and global network dynamics, ensuring improved network stability and user satisfaction. The efficacy of the approach is validated through simulations on the dataset, showing significant improvements in load balancing and reduced overload as the RL agent iteratively interacts with the environment and learns to dynamically adjust pricing strategies based on real-time demand patterns and station constraints. The findings highlight the potential of adaptive pricing and load-balancing strategies to address the complexities of EV infrastructure, paving the way for scalable and user-centric solutions.

eess.SY↗

arXiv2025

The subpath number of cactus graphs

The subpath number of a graph G is defined as the total number of subpaths in G, and it is closely related to the number of subtrees, a well-studied topic in graph theory. This paper is a continuation of our previous paper [5], where we investigated the subpath number and identified extremal graphs within the classes of trees, unicyclic graphs, bipartite graphs, and cycle chains. Here, we focus on the subpath number of cactus graphs and characterize all maximal and minimal cacti with n vertices and k cycles. We prove that maximal cacti are cycle chains in which all interior cycles are triangles, while the two end-cycles differ in length by at most one. In contrast, minimal cacti consist of k triangles, all sharing a common vertex, with the remaining vertices forming a tree attached to this joint vertex. By comparing extremal cacti with respect to the subpath number to those that are extremal for the subtree number and the Wiener index, we demonstrate that the subpath number does not correlate with either of these quantities, as their corresponding extremal graphs differ.

math.CO↗

arXiv2025

Optimal Batched Best Arm Identification

We study the batched best arm identification (BBAI) problem, where the learner's goal is to identify the best arm while switching the policy as less as possible. In particular, we aim to find the best arm with probability $1-δ$ for some small constant $δ>0$ while minimizing both the sample complexity (total number of arm pulls) and the batch complexity (total number of batches). We propose the three-batch best arm identification (Tri-BBAI) algorithm, which is the first batched algorithm that achieves the optimal sample complexity in the asymptotic setting (i.e., $δ\rightarrow 0$) and runs in $3$ batches in expectation. Based on Tri-BBAI, we further propose the almost optimal batched best arm identification (Opt-BBAI) algorithm, which is the first algorithm that achieves the near-optimal sample and batch complexity in the non-asymptotic setting (i.e., $δ$ is finite), while enjoying the same batch and sample complexity as Tri-BBAI when $δ$ tends to zero. Moreover, in the non-asymptotic setting, the complexity of previous batch algorithms is usually conditioned on the event that the best arm is returned (with a probability of at least $1-δ$), which is potentially unbounded in cases where a sub-optimal arm is returned. In contrast, the complexity of Opt-BBAI does not rely on such an event. This is achieved through a novel procedure that we design for checking whether the best arm is eliminated, which is of independent interest.

cs.LG↗

arXiv2025

Invitation to the subpath number

In this paper we count all the subpaths of a given graph G; including the subpaths of length zero, and we call this quantity the subpath number of G. The subpath number is related to the extensively studied number of subtrees, as it can be considered as counting subtrees with the additional requirement of maximum degree being two. We first give the explicit formula for the subpath number of trees and unicyclic graphs. We show that among connected graphs on the same number of vertices, the minimum of the subpath number is attained for any tree and the maximum for the complete graph. Further, we show that the complete bipartite graph with partite sets of almost equal size maximizes the subpath number among all bipartite graphs. The explicit formula for cycle chains, i.e. graphs in which two consecutive cycles share a single edge, is also given. This family of graphs includes the unbranched catacondensed benzenoids which implies a possible application of the result in chemistry. The paper is concluded with several directions for possible further research where several conjectures are provided.

math.CO↗

arXiv2025

Demand Forecasting for Electric Vehicle Charging Stations using Multivariate Time-Series Analysis

As the number of electric vehicles (EVs) continues to grow, the demand for charging stations is also increasing, leading to challenges such as long wait times and insufficient infrastructure. High-precision forecasting of EV charging demand is crucial for efficient station management, to address some of these challenges. This paper presents an approach to predict the charging demand at 15-minute intervals for the day ahead using a multivariate long short-term memory (LSTM) network with an attention mechanism. Additionally, the model leverages explainable AI techniques to evaluate the influence of various factors on the predictions, including weather conditions, day of the week, month, and any holiday. SHapley Additive exPlanations (SHAP) are used to quantify the contribution of each feature to the final forecast, providing deeper insights into how these factors affect prediction accuracy. As a result, the framework offers enhanced decision-making for infrastructure planning. The efficacy of the proposed method is demonstrated by simulations using the test data collected from the EV charging stations at California State University, Long Beach.

eess.SY↗

arXiv2025

Modeling Behavior Change for Multi-model At-Risk Students Early Prediction (extended version)

In the educational domain, identifying students at risk of dropping out is essential for allowing educators to intervene effectively, improving both academic outcomes and overall student well-being. Data in educational settings often originate from diverse sources, such as assignments, grades, and attendance records. However, most existing research relies on online learning data and just extracting the quantitative features. While quantification eases processing, it also leads to a significant loss of original information. Moreover, current models primarily identify students with consistently poor performance through simple and discrete behavioural patterns, failing to capture the complex continuity and non-linear changes in student behaviour. We have developed an innovative prediction model, Multimodal- ChangePoint Detection (MCPD), utilizing the textual teacher remark data and numerical grade data from middle schools. Our model achieves a highly integrated and intelligent analysis by using independent encoders to process two data types, fusing the encoded feature. The model further refines its analysis by leveraging a changepoint detection module to pinpoint crucial behavioral changes, which are integrated as dynamic weights through a simple attention mechanism. Experimental validations indicate that our model achieves an accuracy range of 70- 75%, with an average outperforming baseline algorithms by approximately 5-10%. Additionally, our algorithm demonstrates a certain degree of transferability, maintaining high accuracy when adjusted and retrained with different definitions of at-risk, proving its broad applicability.

cs.CY↗

arXiv2025

Gensor: A Graph-based Construction Tensor Compilation Method for Deep Learning

High-performance deep learning depends on efficient tensor programs. In recent years, automatic tensor program optimization, also known as tensor compilation, has emerged as the primary approach to generating efficient tensor programs. However, how to generate kernels with higher performance in a shorter time is still the key challenge. In this paper, we present Gensor, a graph-based construction tensor compilation method for deep learning, to further improve the performance of construction tensor compilation. Unlike existing tree-based methods, Gensor abstracts construction space into a graph structure. Gensor then explores the construction space with Markov analysis. Gensor takes tensor programs as states and models scheduling primitives as transition actions between these states. Therefore, the process of tensor program construction optimization is abstracted as a graph traversal process. This approach expands the optimization space, improving operator performance while ensuring rapid optimization. Extensive experiments with typical operators demonstrate that Gensor significantly outperforms the state-of-the-art methods on GPUs for both cloud servers and edge devices. As a result, Gensor can generate operator kernels in seconds, with performance increasing by 18\% on average, reaching a maximum of 30\%. It also achieves high speedup for end-to-end models like ResNet-50 and GPT-2, with an average acceleration of 20\%.

cs.DC↗

arXiv2025

Near-Optimal Online Learning for Multi-Agent Submodular Coordination: Tight Approximation and Communication Efficiency

Coordinating multiple agents to collaboratively maximize submodular functions in unpredictable environments is a critical task with numerous applications in machine learning, robot planning and control. The existing approaches, such as the OSG algorithm, are often hindered by their poor approximation guarantees and the rigid requirement for a fully connected communication graph. To address these challenges, we firstly present a $\textbf{MA-OSMA}$ algorithm, which employs the multi-linear extension to transfer the discrete submodular maximization problem into a continuous optimization, thereby allowing us to reduce the strict dependence on a complete graph through consensus techniques. Moreover, $\textbf{MA-OSMA}$ leverages a novel surrogate gradient to avoid sub-optimal stationary points. To eliminate the computationally intensive projection operations in $\textbf{MA-OSMA}$, we also introduce a projection-free $\textbf{MA-OSEA}$ algorithm, which effectively utilizes the KL divergence by mixing a uniform distribution. Theoretically, we confirm that both algorithms achieve a regret bound of $\widetilde{O}(\sqrt{\frac{C_{T}T}{1-β}})$ against a $(\frac{1-e^{-c}}{c})$-approximation to the best comparator in hindsight, where $C_{T}$ is the deviation of maximizer sequence, $β$ is the spectral gap of the network and $c$ is the joint curvature of submodular objectives. This result significantly improves the $(\frac{1}{1+c})$-approximation provided by the state-of-the-art OSG algorithm. Finally, we demonstrate the effectiveness of our proposed algorithms through simulation-based multi-target tracking.

cs.MA↗

arXiv2025

Combinatorial Optimization Perspective based Framework for Multi-behavior Recommendation

In real-world recommendation scenarios, users engage with items through various types of behaviors. Leveraging diversified user behavior information for learning can enhance the recommendation of target behaviors (e.g., buy), as demonstrated by recent multi-behavior methods. The mainstream multi-behavior recommendation framework consists of two steps: fusion and prediction. Recent approaches utilize graph neural networks for multi-behavior fusion and employ multi-task learning paradigms for joint optimization in the prediction step, achieving significant success. However, these methods have limited perspectives on multi-behavior fusion, which leads to inaccurate capture of user behavior patterns in the fusion step. Moreover, when using multi-task learning for prediction, the relationship between the target task and auxiliary tasks is not sufficiently coordinated, resulting in negative information transfer. To address these problems, we propose a novel multi-behavior recommendation framework based on the combinatorial optimization perspective, named COPF. Specifically, we treat multi-behavior fusion as a combinatorial optimization problem, imposing different constraints at various stages of each behavior to restrict the solution space, thus significantly enhancing fusion efficiency (COGCN). In the prediction step, we improve both forward and backward propagation during the generation and aggregation of multiple experts to mitigate negative transfer caused by differences in both feature and label distributions (DFME). Comprehensive experiments on three real-world datasets indicate the superiority of COPF. Further analyses also validate the effectiveness of the COGCN and DFME modules. Our code is available at https://github.com/1918190/COPF.

cs.IR↗

arXiv2025

Mixture of Knowledge Minigraph Agents for Literature Review Generation

Literature reviews play a crucial role in scientific research for understanding the current state of research, identifying gaps, and guiding future studies on specific topics. However, the process of conducting a comprehensive literature review is yet time-consuming. This paper proposes a novel framework, collaborative knowledge minigraph agents (CKMAs), to automate scholarly literature reviews. A novel prompt-based algorithm, the knowledge minigraph construction agent (KMCA), is designed to identify relations between concepts from academic literature and automatically constructs knowledge minigraphs. By leveraging the capabilities of large language models on constructed knowledge minigraphs, the multiple path summarization agent (MPSA) efficiently organizes concepts and relations from different viewpoints to generate literature review paragraphs. We evaluate CKMAs on three benchmark datasets. Experimental results show the effectiveness of the proposed method, further revealing promising applications of LLMs in scientific research.

cs.CL↗