Searcharxiv⌕ Search

arXiv subjects

Yifan Gao

Publications and source records attributed to Yifan Gao.

At least 73 records · Page 4Linked to original sources
arXiv2025

M+: Extending MemoryLLM with Scalable Long-Term Memory

Equipping large language models (LLMs) with latent-space memory has attracted increasing attention as they can extend the context window of existing language models. However, retaining information from the distant past remains a challenge. For example, MemoryLLM (Wang et al., 2024a), as a representative work with latent-space memory, compresses past information into hidden states across all layers, forming a memory pool of 1B parameters. While effective for sequence lengths up to 16k tokens, it struggles to retain knowledge beyond 20k tokens. In this work, we address this limitation by introducing M+, a memory-augmented model based on MemoryLLM that significantly enhances long-term information retention. M+ integrates a long-term memory mechanism with a co-trained retriever, dynamically retrieving relevant information during text generation. We evaluate M+ on diverse benchmarks, including long-context understanding and knowledge retention tasks. Experimental results show that M+ significantly outperforms MemoryLLM and recent strong baselines, extending knowledge retention from under 20k to over 160k tokens with similar GPU memory overhead. We open-source our code at https://github.com/wangyu-ustc/MemoryLLM

cs.CL↗
arXiv2025

Observation of ferron transport in ferroelectrics

Ferroelectrics feature spontaneous electric dipolar order reconfigurable via electric fields. Recent theoretical studies of the collective excitations of this electric dipolar order give rise to the hope that "ferron" quasiparticles may complement the magnons of magnetic materials in information and heat management technologies. Yet direct experimental evidence of ferron transport remains elusive. Here we demonstrate efficient ferron injection and detection enabled by ferromagnetic metal contacts, achieving nonlocal signal transmission over micrometer distances in a prototypical ferroelectric PMN-PT. The transmission efficiency can be switched by external magnetic fields that couple to the contacts and gate electric fields that control the ferron excitations. Ferron-based devices open new power saving strategies that employ ferroelectric materials in a future sustainable information society.

physics.app-ph↗
arXiv2025

Gradient Boosting Decision Tree with LSTM for Investment Prediction

This paper proposes a hybrid framework combining LSTM (Long Short-Term Memory) networks with LightGBM and CatBoost for stock price prediction. The framework processes time-series financial data and evaluates performance using seven models: Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), Bidirectional LSTM (BiLSTM), vanilla LSTM, XGBoost, LightGBM, and standard Neural Networks (NNs). Key metrics, including MAE, R-squared, MSE, and RMSE, are used to establish benchmarks across different time scales. Building on these benchmarks, we develop an ensemble model that combines the strengths of sequential and tree-based approaches. Experimental results show that the proposed framework improves accuracy by 10 to 15 percent compared to individual models and reduces error during market changes. This study highlights the potential of ensemble methods for financial forecasting and provides a flexible design for integrating new machine learning techniques.

cs.LG↗
arXiv2025

SETransformer: A Hybrid Attention-Based Architecture for Robust Human Activity Recognition

Human Activity Recognition (HAR) using wearable sensor data has become a central task in mobile computing, healthcare, and human-computer interaction. Despite the success of traditional deep learning models such as CNNs and RNNs, they often struggle to capture long-range temporal dependencies and contextual relevance across multiple sensor channels. To address these limitations, we propose SETransformer, a hybrid deep neural architecture that combines Transformer-based temporal modeling with channel-wise squeeze-and-excitation (SE) attention and a learnable temporal attention pooling mechanism. The model takes raw triaxial accelerometer data as input and leverages global self-attention to capture activity-specific motion dynamics over extended time windows, while adaptively emphasizing informative sensor channels and critical time steps. We evaluate SETransformer on the WISDM dataset and demonstrate that it significantly outperforms conventional models including LSTM, GRU, BiLSTM, and CNN baselines. The proposed model achieves a validation accuracy of 84.68\% and a macro F1-score of 84.64\%, surpassing all baseline architectures by a notable margin. Our results show that SETransformer is a competitive and interpretable solution for real-world HAR tasks, with strong potential for deployment in mobile and ubiquitous sensing applications.

cs.LG↗
arXiv2025

EcomScriptBench: A Multi-task Benchmark for E-commerce Script Planning via Step-wise Intention-Driven Product Association

Goal-oriented script planning, or the ability to devise coherent sequences of actions toward specific goals, is commonly employed by humans to plan for typical activities. In e-commerce, customers increasingly seek LLM-based assistants to generate scripts and recommend products at each step, thereby facilitating convenient and efficient shopping experiences. However, this capability remains underexplored due to several challenges, including the inability of LLMs to simultaneously conduct script planning and product retrieval, difficulties in matching products caused by semantic discrepancies between planned actions and search queries, and a lack of methods and benchmark data for evaluation. In this paper, we step forward by formally defining the task of E-commerce Script Planning (EcomScript) as three sequential subtasks. We propose a novel framework that enables the scalable generation of product-enriched scripts by associating products with each step based on the semantic similarity between the actions and their purchase intentions. By applying our framework to real-world e-commerce data, we construct the very first large-scale EcomScript dataset, EcomScriptBench, which includes 605,229 scripts sourced from 2.4 million products. Human annotations are then conducted to provide gold labels for a sampled subset, forming an evaluation benchmark. Extensive experiments reveal that current (L)LMs face significant challenges with EcomScript tasks, even after fine-tuning, while injecting product purchase intentions improves their performance.

cs.CL↗
arXiv2025

Chemical design of monolayer altermagnets

The crystal-symmetry-paired spin-momentum locking (CSML) arisen from the intrinsic crystal symmetry connecting different magnetic sublattices in altermagnets enables many exotic spintronics properties such as unconventional piezomagnetism and noncollinear spin current. However, the shortage of monolayer altermagnets restricts further exploration of dimensionally confined phenomena and applications of nanostructured devices. Here, we propose general chemical design principles inspired by sublattice symmetry of layered altermagnet V$_2$(Se,Te)$_2$O through symmetry-preserving structural modification and valence-adaptive chemical substitutions. In total, we construct 2600 candidates across four structural frameworks, M$_2$A$_2$B$_{1,0}$ and their Janus derivatives. High-throughput calculations identify 670 potential altermagnets with Néel-ordered ground states, among which 91 ones exhibiting CSML Dirac cones that enable spin-polarized ultra-fast transport. These materials also feature different ground-state magnetic orderings and demonstrate diverse electronic behaviors, ranging from semiconductors, metals, half-metals, to Dirac semimetals. This work not only reveals abundant monolayer altermagnets, but also establishes a rational principle for their design, opening gates for exploration of confined magnetism and spintronics in atomically thin systems.

cond-mat.mtrl-sci↗
arXiv2025

WeGA: Weakly-Supervised Global-Local Affinity Learning Framework for Lymph Node Metastasis Prediction in Rectal Cancer

Accurate lymph node metastasis (LNM) assessment in rectal cancer is essential for treatment planning, yet current MRI-based evaluation shows unsatisfactory accuracy, leading to suboptimal clinical decisions. Developing automated systems also faces significant obstacles, primarily the lack of node-level annotations. Previous methods treat lymph nodes as isolated entities rather than as an interconnected system, overlooking valuable spatial and contextual information. To solve this problem, we present WeGA, a novel weakly-supervised global-local affinity learning framework that addresses these challenges through three key innovations: 1) a dual-branch architecture with DINOv2 backbone for global context and residual encoder for local node details; 2) a global-local affinity extractor that aligns features across scales through cross-attention fusion; and 3) a regional affinity loss that enforces structural coherence between classification maps and anatomical regions. Experiments across one internal and two external test centers demonstrate that WeGA outperforms existing methods, achieving AUCs of 0.750, 0.822, and 0.802 respectively. By effectively modeling the relationships between individual lymph nodes and their collective context, WeGA provides a more accurate and generalizable approach for lymph node metastasis prediction, potentially enhancing diagnostic precision and treatment selection for rectal cancer patients.

eess.IV↗
arXiv2025

Multiple points on the boundaries of Brownian loop-soup clusters

For a Brownian loop soup with intensity $c\in(0,1]$ in the unit disk, we show that almost surely, the set of simple (resp. double) points on any portion of boundary of any of its clusters has Hausdorff dimension $2-ξ_c(2)$ (resp. $2-ξ_c(4)$), where $ξ_c(k)$ is the generalized disconnection exponent computed in arxiv:1901.05436. As a consequence, when the dimension is positive, such points are a.s. dense on every boundary of every cluster. There are a.s. no triple points on the cluster boundaries. As an intermediate result, we establish a separation lemma for Brownian loop soups, which is a powerful tool for obtaining sharp estimates on non-intersection and non-disconnection probabilities in the setting of loop soups. In particular, it allows us to define a family of generalized intersection exponents $ξ_c(k, λ)$, and show that $ξ_c(k)$ is the limit as $λ\searrow 0$ of $ξ_c(k, λ)$.

math.PR↗
arXiv2025

PosterMaker: Towards High-Quality Product Poster Generation with Accurate Text Rendering

Product posters, which integrate subject, scene, and text, are crucial promotional tools for attracting customers. Creating such posters using modern image generation methods is valuable, while the main challenge lies in accurately rendering text, especially for complex writing systems like Chinese, which contains over 10,000 individual characters. In this work, we identify the key to precise text rendering as constructing a character-discriminative visual feature as a control signal. Based on this insight, we propose a robust character-wise representation as control and we develop TextRenderNet, which achieves a high text rendering accuracy of over 90%. Another challenge in poster generation is maintaining the fidelity of user-specific products. We address this by introducing SceneGenNet, an inpainting-based model, and propose subject fidelity feedback learning to further enhance fidelity. Based on TextRenderNet and SceneGenNet, we present PosterMaker, an end-to-end generation framework. To optimize PosterMaker efficiently, we implement a two-stage training strategy that decouples text rendering and background generation learning. Experimental results show that PosterMaker outperforms existing baselines by a remarkable margin, which demonstrates its effectiveness.

cs.CV↗
arXiv2025

Up-to-constants estimates on four-arm events for simple conformal loop ensemble

We prove up-to-constants estimates for a general class of four-arm events in simple conformal loop ensembles, i.e. CLE$_κ$ for $κ\in (8/3,4]$. The four-arm events that we consider can be created by either one or two loops, with no constraint on the topology of the crossings. Our result is a key input in our series of works arxiv:2409.16230 and arxiv:2409.16273 on percolation of the two-sided level sets in the discrete Gaussian free field (and level sets in the occupation field of the random walk loop soup). In order to get rid of all constraints on the topology of the crossings, we rely on the Brownian loop-soup representation of simple CLE [Ann. Math. 176 (2012) 1827-1917], and a "cluster version" of a separation lemma for the Brownian loop soup. As a corollary, we also obtain up-to-constants estimates for a general version of four-arm events for SLE$_κ$ for $κ\in (8/3,4]$. This fixes (in the case of four arms and $κ\in(8/3,4]$) an essential gap in [Ann. Probab. 46 (2018) 2863-2907] and improves some estimates therein.

math.PR↗
arXiv2025

Scaling Laws for Predicting Downstream Performance in LLMs

Precise estimation of downstream performance in large language models (LLMs) prior to training is essential for guiding their development process. Scaling laws analysis utilizes the statistics of a series of significantly smaller sampling language models (LMs) to predict the performance of the target LLM. For downstream performance prediction, the critical challenge lies in the emergent abilities in LLMs that occur beyond task-specific computational thresholds. In this work, we focus on the pre-training loss as a more computation-efficient metric for performance estimation. Our two-stage approach FLP consists of first estimating a function that maps computational resources (e.g., FLOPs) to the pre-training Loss using a series of fully-converged sampling models, followed by mapping the pre-training loss to downstream task Performance using the intermediate models with emerged performance. In our experiments, this FLP solution accurately predicts the performance of LLMs with 7B and 13B parameters using a series of sampling LMs up to 3B, achieving error margins of 5% and 10%, respectively, and significantly outperforming the FLOPs-to-Performance approach. Further, we present FLP-M, a fundamental approach for performance prediction that addresses the practical need to integrate datasets from multiple sources during pre-training. FLP-M extends the power law analytical function to predict domain-specific pre-training loss based on FLOPs across data sources, and employs a two-layer neural network to model the non-linear relationship between multiple domain-specific loss and downstream performance. By utilizing a 3B LLM trained on a specific ratio and a series of smaller sampling LMs, FLP-M can effectively forecast the performance of 3B and 7B LLMs across various data mixtures for most benchmarks within 10% error margins.

cs.CL↗
arXiv2025

IHEval: Evaluating Language Models on Following the Instruction Hierarchy

The instruction hierarchy, which establishes a priority order from system messages to user messages, conversation history, and tool outputs, is essential for ensuring consistent and safe behavior in language models (LMs). Despite its importance, this topic receives limited attention, and there is a lack of comprehensive benchmarks for evaluating models' ability to follow the instruction hierarchy. We bridge this gap by introducing IHEval, a novel benchmark comprising 3,538 examples across nine tasks, covering cases where instructions in different priorities either align or conflict. Our evaluation of popular LMs highlights their struggle to recognize instruction priorities. All evaluated models experience a sharp performance decline when facing conflicting instructions, compared to their original instruction-following performance. Moreover, the most competitive open-source model only achieves 48% accuracy in resolving such conflicts. Our results underscore the need for targeted optimization in the future development of LMs.

cs.CL↗
arXiv2025

Leveraging Multimodal Methods and Spontaneous Speech for Alzheimer's Disease Identification

Cognitive impairment detection through spontaneous speech is a promising avenue for early diagnosis of Alzheimer's disease (AD) and mild cognitive impairment (MCI), where timely intervention can significantly improve patient outcomes. The PROCESS Grand Challenge at ICASSP 2025 addresses these tasks by promoting innovative classification and regression methods for detecting cognitive decline. In this paper, we propose a multimodal fusion strategy that combines interpretable linguistic features with temporal embeddings extracted from pre-trained models. Our approach achieves an F1-score of 0.649 for the classification task (predicting healthy, MCI, dementia) and an RMSE of 2.628 for the regression task (MMSE score prediction), securing the top overall ranking in the competition.

cs.SD↗
arXiv2025

A Dual-Stage Time-Context Network for Speech-Based Alzheimer's Disease Detection

Alzheimer's disease (AD) is a progressive neurodegenerative disorder that leads to irreversible cognitive decline in memory and communication. Early detection of AD through speech analysis is crucial for delaying disease progression. However, existing methods mainly use pre-trained acoustic models for feature extraction but have limited ability to model both local and global patterns in long-duration speech. In this letter, we introduce a Dual-Stage Time-Context Network (DSTC-Net) for speech-based AD detection, integrating local acoustic features with global conversational context in long-duration recordings.We first partition each long-duration recording into fixed-length segments to reduce computational overhead and preserve local temporal details.Next, we feed these segments into an Intra-Segment Temporal Attention (ISTA) module, where a bidirectional Long Short-Term Memory (BiLSTM) network with frame-level attention extracts enhanced local features.Subsequently, a Cross-Segment Context Attention (CSCA) module applies convolution-based context modeling and adaptive attention to unify global patterns across all segments.Extensive experiments on the ADReSSo dataset show that our DSTC-Net outperforms state-of-the-art models, reaching 83.10% accuracy and 83.15% F1.

cs.SD↗
arXiv2025

Hephaestus: Improving Fundamental Agent Capabilities of Large Language Models through Continual Pre-Training

Due to the scarcity of agent-oriented pre-training data, LLM-based autonomous agents typically rely on complex prompting or extensive fine-tuning, which often fails to introduce new capabilities while preserving strong generalizability. We introduce Hephaestus-Forge, the first large-scale pre-training corpus designed to enhance the fundamental capabilities of LLM agents in API function calling, intrinsic reasoning and planning, and adapting to environmental feedback. Hephaestus-Forge comprises 103B agent-specific data encompassing 76,537 APIs, including both tool documentation to introduce knowledge of API functions and function calling trajectories to strengthen intrinsic reasoning. To explore effective training protocols, we investigate scaling laws to identify the optimal recipe in data mixing ratios. By continual pre-training on Hephaestus-Forge, Hephaestus outperforms small- to medium-scale open-source LLMs and rivals commercial LLMs on three agent benchmarks, demonstrating the effectiveness of our pre-training corpus in enhancing fundamental agentic capabilities and generalization of LLMs to new tasks or environments.

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↗