arXiv subjects

Lei Duan

Publications and source records attributed to Lei Duan.

16 recordsLinked to original sources

arXiv2026

MixFormer: Linear Transformer with Mixture of Memory Experts

State Space Models (SSMs), as a mainstream research direction of linear Transformers, aim to achieve higher efficiency than standard Transformers in long-context modeling. However, existing SSMs suffer from limited input adaptivity and constrained memory capacity, leading to information loss when modeling ultra-long sequences. To address these limitations, we propose MixFormer, a novel linear Transformer that integrates a Mixture-of-Memory-Experts (MoE) mechanism. Specifically, the model maintains differentiated memory states through multiple collaborating memory experts and employs a novel Time-Aware Linear Attention (TALA) mechanism, which leverages learnable exponential decay functions and positional biases to dynamically update memory. This design enables the model to selectively reinforce important historical information while effectively mitigating memory dilution, substantially improving long-range dependency modeling. Experiments on long-sequence text and image generation tasks demonstrate that MixFormer not only achieves significant performance gains but also provides a more sustainable computational backbone for the next generation of web infrastructure.

cs.LG↗

arXiv2026

RouteTS: Frequency-Time Routing for Time Series Forecasting

Real-world time series inherently intertwine global periodic structures with localized non-stationary variations. Existing approaches process these heterogeneous dynamics within a single computational domain, incurring fundamental limitations: time-domain models suffer from periodic misalignment over long horizons, while frequency-domain models over-smooth transient spikes. We argue that the optimal computational domain is not a property of the model, but of the data itself. Based on this principle, we propose RouteTS, a unified forecasting framework that partitions the frequency spectrum via amplitude routing and delegates components to their mathematically optimal domains. Dominant frequencies are processed by a complex-valued linear predictor in the frequency domain to preserve periodic structure, while residual spectral energy is reverted to the time domain and modeled by a lightweight MLP for local variations. Extensive experiments demonstrate that RouteTS achieves competitive prediction accuracy across diverse real-world datasets, with routing decisions guided by the underlying spectral signature. Furthermore, the lightweight design of RouteTS provides significant computational efficiency advantages, offering a principled solution to the longstanding dilemma between global periodicity and local transience.

cs.LG↗

arXiv2025

Harnessing RLHF for Robust Unanswerability Recognition and Trustworthy Response Generation in LLMs

Conversational Information Retrieval (CIR) systems, while offering intuitive access to information, face a significant challenge: reliably handling unanswerable questions to prevent the generation of misleading or hallucinated content. Traditional approaches often rely on external classifiers, which can introduce inconsistencies with the core generative Large Language Models (LLMs). This paper introduces Self-Aware LLM for Unanswerability (SALU), a novel approach that deeply integrates unanswerability detection directly within the LLM's generative process. SALU is trained using a multi-task learning framework for both standard Question Answering (QA) and explicit abstention generation for unanswerable queries. Crucially, it incorporates a confidence-score-guided reinforcement learning with human feedback (RLHF) phase, which explicitly penalizes hallucinated responses and rewards appropriate abstentions, fostering intrinsic self-awareness of knowledge boundaries. Through extensive experiments on our custom-built C-IR_Answerability dataset, SALU consistently outperforms strong baselines, including hybrid LLM-classifier systems, in overall accuracy for correctly answering or abstaining from questions. Human evaluation further confirms SALU's superior reliability, achieving high scores in factuality, appropriate abstention, and, most importantly, a dramatic reduction in hallucination, demonstrating its ability to robustly "know when to say 'I don't know'."

cs.CL↗

arXiv2025

DynMoLE: Boosting Mixture of LoRA Experts Fine-Tuning with a Hybrid Routing Mechanism

Instruction-based fine-tuning of large language models (LLMs) has achieved remarkable success in various natural language processing (NLP) tasks. Parameter-efficient fine-tuning (PEFT) methods, such as Mixture of LoRA Experts (MoLE), combine the efficiency of Low-Rank Adaptation (LoRA) with the versatility of Mixture of Experts (MoE) models, demonstrating significant potential for handling multiple downstream tasks. However, the existing routing mechanisms for MoLE often involve a trade-off between computational efficiency and predictive accuracy, and they fail to fully address the diverse expert selection demands across different transformer layers. In this work, we propose DynMoLE, a hybrid routing strategy that dynamically adjusts expert selection based on the Tsallis entropy of the router's probability distribution. This approach mitigates router uncertainty, enhances stability, and promotes more equitable expert participation, leading to faster convergence and improved model performance. Additionally, we introduce an auxiliary loss based on Tsallis entropy to further guide the model toward convergence with reduced uncertainty, thereby improving training stability and performance. Our extensive experiments on commonsense reasoning benchmarks demonstrate that DynMoLE achieves substantial performance improvements, outperforming LoRA by 9.6% and surpassing the state-of-the-art MoLE method, MoLA, by 2.3%. We also conduct a comprehensive ablation study to evaluate the contributions of DynMoLE's key components.

cs.CL↗

arXiv2024

Deep Causal Inference for Point-referenced Spatial Data with Continuous Treatments

Causal reasoning is often challenging with spatial data, particularly when handling high-dimensional inputs. To address this, we propose a neural network (NN) based framework integrated with an approximate Gaussian process to manage spatial interference and unobserved confounding. Additionally, we adopt a generalized propensity-score-based approach to address partially observed outcomes when estimating causal effects with continuous treatments. We evaluate our framework using synthetic, semi-synthetic, and real-world data inferred from satellite imagery. Our results demonstrate that NN-based models significantly outperform linear spatial regression models in estimating causal effects. Furthermore, in real-world case studies, NN-based models offer more reasonable predictions of causal effects, facilitating decision-making in relevant applications.

cs.LG↗

arXiv2024

SCOP: A Sequence-Structure Contrast-Aware Framework for Protein Function Prediction

Improving the ability to predict protein function can potentially facilitate research in the fields of drug discovery and precision medicine. Technically, the properties of proteins are directly or indirectly reflected in their sequence and structure information, especially as the protein function is largely determined by its spatial properties. Existing approaches mostly focus on protein sequences or topological structures, while rarely exploiting the spatial properties and ignoring the relevance between sequence and structure information. Moreover, obtaining annotated data to improve protein function prediction is often time-consuming and costly. To this end, this work proposes a novel contrast-aware pre-training framework, called SCOP, for protein function prediction. We first design a simple yet effective encoder to integrate the protein topological and spatial features under the structure view. Then a convolutional neural network is utilized to learn the protein features under the sequence view. Finally, we pretrain SCOP by leveraging two types of auxiliary supervision to explore the relevance between these two views and thus extract informative representations to better predict protein function. Experimental results on four benchmark datasets and one self-built dataset demonstrate that SCOP provides more specific results, while using less pre-training data.

q-bio.BM↗

arXiv2024

mLoRA: Fine-Tuning LoRA Adapters via Highly-Efficient Pipeline Parallelism in Multiple GPUs

Transformer-based, pre-trained large language models (LLMs) have demonstrated outstanding performance across diverse domains, particularly in the emerging {\em pretrain-then-finetune} paradigm. Low-Rank Adaptation (LoRA), a parameter-efficient fine-tuning method, is commonly used to adapt a base LLM to multiple downstream tasks. Further, LLM platforms enable developers to fine-tune multiple models and develop various domain-specific applications simultaneously. However, existing model parallelism schemes suffer from high communication overhead and inefficient GPU utilization when training multiple LoRA tasks across GPUs and machines. In this paper, we present mLoRA, a parallelism-efficient fine-tuning system designed for training multiple LoRA across GPUs and machines. mLoRA introduces a novel LoRA-aware pipeline parallelism scheme that efficiently pipelines independent LoRA adapters and their distinct fine-tuning stages across GPUs and machines, along with a new LoRA-efficient operator to enhance GPU utilization during pipelined LoRA training. Our extensive evaluation shows that mLoRA can significantly reduce average fine-tuning task completion time, e.g., by 30\%, compared to state-of-the-art methods like FSDP. More importantly, mLoRA enables simultaneous fine-tuning of larger models, e.g., two Llama-2-13B models on four NVIDIA RTX A6000 48GB GPUs, which is not feasible for FSDP due to high memory requirements. Hence, mLoRA not only increases fine-tuning efficiency but also makes it more accessible on cost-effective GPUs. mLoRA has been deployed in AntGroup's production environment.

cs.LG↗

arXiv2024

MixLoRA: Enhancing Large Language Models Fine-Tuning with LoRA-based Mixture of Experts

Fine-tuning Large Language Models (LLMs) is a common practice to adapt pre-trained models for specific applications. While methods like LoRA have effectively addressed GPU memory constraints during fine-tuning, their performance often falls short, especially in multi-task scenarios. In contrast, Mixture-of-Expert (MoE) models, such as Mixtral 8x7B, demonstrate remarkable performance in multi-task learning scenarios while maintaining a reduced parameter count. However, the resource requirements of these MoEs remain challenging, particularly for consumer-grade GPUs with less than 24GB memory. To tackle these challenges, we propose MixLoRA, an approach to construct a resource-efficient sparse MoE model based on LoRA. MixLoRA inserts multiple LoRA-based experts within the feed-forward network block of a frozen pre-trained dense model and employs a commonly used top-k router. Unlike other LoRA-based MoE methods, MixLoRA enhances model performance by utilizing independent attention-layer LoRA adapters. Additionally, an auxiliary load balance loss is employed to address the imbalance problem of the router. Our evaluations show that MixLoRA improves about 9% accuracy compared to state-of-the-art PEFT methods in multi-task learning scenarios. We also propose a new high-throughput framework to alleviate the computation and memory bottlenecks during the training and inference of MOE models. This framework reduces GPU memory consumption by 40% and token computation latency by 30% during both training and inference.

cs.CL↗

arXiv2024

Fast Asymmetric Factorization for Large Scale Multiple Kernel Clustering

Kernel methods are extensively employed for nonlinear data clustering, yet their effectiveness heavily relies on selecting suitable kernels and associated parameters, posing challenges in advance determination. In response, Multiple Kernel Clustering (MKC) has emerged as a solution, allowing the fusion of information from multiple base kernels for clustering. However, both early fusion and late fusion methods for large-scale MKC encounter challenges in memory and time constraints, necessitating simultaneous optimization of both aspects. To address this issue, we propose Efficient Multiple Kernel Concept Factorization (EMKCF), which constructs a new sparse kernel matrix inspired by local regression to achieve memory efficiency. EMKCF learns consensus and individual representations by extending orthogonal concept factorization to handle multiple kernels for time efficiency. Experimental results demonstrate the efficiency and effectiveness of EMKCF on benchmark datasets compared to state-of-the-art methods. The proposed method offers a straightforward, scalable, and effective solution for large-scale MKC tasks.

cs.LG↗

arXiv2024

Augmenting Ground-Level PM2.5 Prediction via Kriging-Based Pseudo-Label Generation

Fusing abundant satellite data with sparse ground measurements constitutes a major challenge in climate modeling. To address this, we propose a strategy to augment the training dataset by introducing unlabeled satellite images paired with pseudo-labels generated through a spatial interpolation technique known as ordinary kriging, thereby making full use of the available satellite data resources. We show that the proposed data augmentation strategy helps enhance the performance of the state-of-the-art convolutional neural network-random forest (CNN-RF) model by a reasonable amount, resulting in a noteworthy improvement in spatial correlation and a reduction in prediction error.

cs.LG↗

arXiv2022

A ferrotoroidic candidate with well-separated spin chains

The search of novel quasi one-dimensional (1D) materials is one of the important aspects in the field of material science. Toroidal moment, the order parameter of ferrotoroidic order, can be generated by a head-to-tail configuration of magnetic moment. It has been theoretically proposed that one-dimensional (1D) dimerized and antiferromagnetic-like spin chain hosts ferrotoroidicity and has the toroidal moment composed of only two antiparallel spins. Here, we report a ferrotoroidic candidate of Ba6Cr2S10 with such a theoretical model of spin chain. The structure consists of unique dimerized face-sharing CrS6 octahedral chains along the c axis. An antiferromagnetic-like ordering at ~10 K breaks both space- and time-reversal symmetries and the magnetic point group of mm'2' allows three ferroic orders in Ba6Cr2S10: (anti)ferromagnetic, ferroelectric and ferrotoroidic orders. Our investigation reveals that Ba6Cr2S10 is a rare ferrotoroidic candidate with quasi 1D spin chain, which can be considered as a starting point for the further exploration of the physics and applications of ferrotoroidicity.

cond-mat.str-el↗

arXiv2021

Probing the superconducting gap structure in the noncentrosymmetric topological superconductor ZrRuAs

The superconducting gap structure of a topological crystalline insulator (TCI) candidate ZrRuAs ($T^{\rm on}_{\rm c}$ = 7.9(1) K) with a noncentrosymmetric crystal structure has been investigated using muon spin rotation/relaxation ($μ$SR) measurements in transverse-field (TF) and zero-field (ZF) geometries. We also present the results of magnetization, electrical resistivity and heat capacity measurements on ZrRuAs, which reveal bulk superconductivity below 7.9~K. The temperature dependence of the effective penetration depth obtained from the analysis of the TF-$μ$SR spectra below $T_{\rm c}$ is well described by an isotropic $s$-wave gap model as also inferred from an analysis of the heat capacity in the superconducting state. ZF $μ$SR data do not show any significant change in the muon spin relaxation rate above and below the superconducting transition temperature indicating that time-reversal symmetry is preserved in the superconducting state of this material.

cond-mat.supr-con↗

arXiv2020

A New Quasi One-Dimensional Compound Ba3TiTe5 and Superconductivity Induced by Pressure

We report systematical studies of a new quasi-one-dimensional (1D) compound Ba3TiTe5 and the high-pressure induced superconductivity therein. Ba3TiTe5 was synthesized at high pressure and high temperature. It crystallizes into a hexagonal structure (P63/mcm), which consists of infinite face-sharing octahedral TiTe6 chains and Te chains along the c axis, exhibiting a strong 1D characteristic structure. The first-principles calculations demonstrate that Ba3TiTe5 is a well-defined 1D conductor and thus, it can be considered a starting point to explore the exotic physics induced by pressure via enhancing the interchain hopping to move the 1D conductor to a high dimensional metal. For Ba3TiTe5, high-pressure techniques were employed to study the emerging physics dependent on interchain hopping, such as the Umklapp scattering effect, spin/charge density wave (SDW/CDW), superconductivity and non-Fermi Liquid behavior. Finally, a complete phase diagram was plotted. The superconductivity emerges from 8.8 GPa, near which the Umklapp gap is mostly suppressed. Tc is enhanced and reaches the maximum ~6 K at about 36.7 GPa, where the spin/charge density wave (SDW/CDW) is completely suppressed, and a non-Fermi Liquid behavior appears. Our results suggest that the appearance of superconductivity is associated with the fluctuation due to the suppression of Umklapp gap and the enhancement of Tc is related with the fluctuation of the SDW/CDW.

cond-mat.supr-con↗

arXiv2018

Towards Highly Accurate and Stable Face Alignment for High-Resolution Videos

In recent years, heatmap regression based models have shown their effectiveness in face alignment and pose estimation. However, Conventional Heatmap Regression (CHR) is not accurate nor stable when dealing with high-resolution facial videos, since it finds the maximum activated location in heatmaps which are generated from rounding coordinates, and thus leads to quantization errors when scaling back to the original high-resolution space. In this paper, we propose a Fractional Heatmap Regression (FHR) for high-resolution video-based face alignment. The proposed FHR can accurately estimate the fractional part according to the 2D Gaussian function by sampling three points in heatmaps. To further stabilize the landmarks among continuous video frames while maintaining the precise at the same time, we propose a novel stabilization loss that contains two terms to address time delay and non-smooth issues, respectively. Experiments on 300W, 300-VW and Talking Face datasets clearly demonstrate that the proposed method is more accurate and stable than the state-of-the-art models.

cs.CV↗

arXiv2012

User Taglines: Alternative Presentations of Expertise and Interest in Social Media

Web applications are increasingly showing recommended users from social media along with some descriptions, an attempt to show relevancy - why they are being shown. For example, Twitter search for a topical keyword shows expert twitterers on the side for 'whom to follow'. Google+ and Facebook also recommend users to follow or add to friend circle. Popular Internet newspaper- The Huffington Post shows Twitter influencers/ experts on the side of an article for authoritative relevant tweets. The state of the art shows user profile bios as summary for Twitter experts, but it has issues with length constraint imposed by user interface (UI) design, missing bio and sometimes funny profile bio. Alternatively, applications can use human generated user summary, but it will not scale. Therefore, we study the problem of automatic generation of informative expertise summary or taglines for Twitter experts in space constraint imposed by UI design. We propose three methods for expertise summary generation- Occupation-Pattern based, Link-Triangulation based and User-Classification based, with use of knowledge-enhanced computing approaches. We also propose methods for final summary selection for users with multiple candidates of generated summaries. We evaluate the proposed approaches by user-study using a number of experiments. Our results show promising quality of 92.8% good summaries with majority agreement in the best case and 70% with majority agreement in the worst case. Our approaches also outperform the state of the art up to 88%. This study has implications in the area of expert profiling, user presentation and application design for engaging user experience.

cs.SI↗

arXiv1997

Probabilistic Event Categorization

This paper describes the automation of a new text categorization task. The categories assigned in this task are more syntactically, semantically, and contextually complex than those typically assigned by fully automatic systems that process unseen test data. Our system for assigning these categories is a probabilistic classifier, developed with a recent method for formulating a probabilistic model from a predefined set of potential features. This paper focuses on feature selection. It presents a number of fully automatic features. It identifies and evaluates various approaches to organizing collocational properties into features, and presents the results of experiments covarying type of organization and type of property. We find that one organization is not best for all kinds of properties, so this is an experimental parameter worth investigating in NLP systems. In addition, the results suggest a way to take advantage of properties that are low frequency but strongly indicative of a class. The problems of recognizing and organizing the various kinds of contextual information required to perform a linguistically complex categorization task have rarely been systematically investigated in NLP.

cmp-lg↗