arXiv subjects

Yanyu Liu

Publications and source records attributed to Yanyu Liu.

14 recordsLinked to original sources

arXiv2025

KV Cache Compression for Inference Efficiency in LLMs: A Review

Withtherapid advancement of large language models (LLMs), the context length for inference has been continuously increasing, leading to an exponential growth in the demand for Key-Value (KV) caching. This has resulted in a significant memory bottleneck, limiting the inference efficiency and scalability of the models. Therefore, optimizing the KV cache during inference is crucial for enhancing performance and efficiency. This review systematically examines current KV cache optimization techniques, including compression strategies such as selective token strategies, quantization, and attention compression. We evaluate the effectiveness, trade-offs, and application scenarios of these methods, providing a comprehensive analysis of their impact on memory usage and inference speed. We focus on identifying the limitations and challenges of existing methods, such as compatibility issues with different models and tasks. Additionally, this review highlights future research directions, including hybrid optimization techniques, adaptive dynamic strategies, and software-hardware co-design. These approaches aim to improve inference efficiency and promote the practical application of large language models.

cs.DC↗

arXiv2025

Universal band center model for the HER activity of non-metal site

In this work, the hydrogen evolution reaction activities of non-metal sites in the transition metal dichalcogenides with the stoichiometry of MX2 are investigated using the first principles calculations. The trained machine learning model demonstrates that the pz band center, bandgap, and period effect are the key factors influencing the HER activity of MX2. Furthermore, it also reveals that the observed non-scaling law between the pz band center and HER activity in the semiconductor-like materials originates from the bandgap-induced downshift of bonding and antibonding states. In addition to the bandgap, the intrinsic p orbital energy level of the non-metal atoms also contributes to the periodic variation of pz band center. Extended calculations indicate that the descriptor is equally applicable to the other catalysts, suggesting its universality in predicting the HER activity of non-metal sites.

cond-mat.mtrl-sci↗

arXiv2025

Numerical simulation of wormhole propagation with the mixed hybridized discontinuous Galerkin finite element method

The acid treatment of carbonate reservoirs is a widely employed technique for enhancing the productivity of oil and gas reservoirs. In this paper, we present a novel combined hybridized mixed discontinuous Galerkin (HMDG) finite element method to simulate the dissolution process near the wellbore, commonly referred to as the wormhole phenomenon. The primary contribution of this work lies in the application of hybridization techniques to both the pressure and concentration equations. Additionally, an upwind scheme is utilized to address convection-dominant scenarios, and a ``cut-off" operator is introduced to maintain the boundedness of porosity. Compared to traditional discontinuous Galerkin methods, the proposed approach results in a global system with fewer unknowns and sparser stencils, thereby significantly reducing computational costs. We analyze the existence and uniqueness of the new combined method and derive optimal error estimates using the developed technique. Numerical examples are provided to validate the theoretical analysis.

math.NA↗

arXiv2025

X Modality Assisting RGBT Object Tracking

Developing robust multi-modal feature representations is crucial for enhancing object tracking performance. In pursuit of this objective, a novel X Modality Assisting Network (X-Net) is introduced, which explores the impact of the fusion paradigm by decoupling visual object tracking into three distinct levels, thereby facilitating subsequent processing. Initially, to overcome the challenges associated with feature learning due to significant discrepancies between RGB and thermal modalities, a plug-and-play pixel-level generation module (PGM) based on knowledge distillation learning is proposed. This module effectively generates the X modality, bridging the gap between the two patterns while minimizing noise interference. Subsequently, to optimize sample feature representation and promote cross-modal interactions, a feature-level interaction module (FIM) is introduced, integrating a mixed feature interaction transformer and a spatial dimensional feature translation strategy. Finally, to address random drifting caused by missing instance features, a flexible online optimization strategy called the decision-level refinement module (DRM) is proposed, which incorporates optical flow and refinement mechanisms. The efficacy of X-Net is validated through experiments on three benchmarks, demonstrating its superiority over state-of-the-art trackers. Notably, X-Net achieves performance gains of 0.47%/1.2% in the average of precise rate and success rate, respectively. Additionally, the research content, data, and code are pledged to be made publicly accessible at https://github.com/DZSYUNNAN/XNet.

cs.CV↗

arXiv2024

Active Legibility in Multiagent Reinforcement Learning

A multiagent sequential decision problem has been seen in many critical applications including urban transportation, autonomous driving cars, military operations, etc. Its widely known solution, namely multiagent reinforcement learning, has evolved tremendously in recent years. Among them, the solution paradigm of modeling other agents attracts our interest, which is different from traditional value decomposition or communication mechanisms. It enables agents to understand and anticipate others' behaviors and facilitates their collaboration. Inspired by recent research on the legibility that allows agents to reveal their intentions through their behavior, we propose a multiagent active legibility framework to improve their performance. The legibility-oriented framework allows agents to conduct legible actions so as to help others optimise their behaviors. In addition, we design a series of problem domains that emulate a common scenario and best characterize the legibility in multiagent reinforcement learning. The experimental results demonstrate that the new framework is more efficient and costs less training time compared to several multiagent reinforcement learning algorithms.

cs.AI↗

arXiv2022

High-Throughput Screening of Transition Metal Single-Atom Catalysts for Nitrogen Reduction Reaction

The discovery of metals as catalytic centers for nitrogen reduction reactions has stimulated great enthusiasm for single-atom catalysts. However, the poor activity and low selectivity of available SACs are far away from the industrial requirement. Through the high throughout first principles calculations, the doping engineering can effectively regulate the NRR performance of b-Sb monolayer. Especially, the origin of activated N2 is revealed from the perspective of the electronic structure of the active center. Among the 24 transition metal dopants, Re@Sb and Tc@Sb showed the best NRR catalytic performance with a low limiting potential. The Re@Sb and Tc@Sb also could significantly inhibit HER and achieve a high theoretical Faradaic efficiency of 100%. Our findings not only accelerate discovery of catalysts for ammonia synthesis but also contribute to further elucidate the structure-performance correlations.

cond-mat.mtrl-sci↗

arXiv2020

Conductive Domain Walls in Non-Oxide Ferroelectrics Sn2P2S6

The conductive domain wall (CDW) is extensively investigated in ferroelectrics, which can be considered as a quasi-two-dimensional reconfigurable conducting channel embedded into an insulating material. Therefore, it is highly important for the application of ferroelectric nanoelectronics. Hitherto, most CDW investigations are restricted in oxides, and limited work has been reported in non-oxides to the contrary. Here, by successfully synthesizing the non-oxide ferroelectric Sn2P2S6 single crystal, we observed and confirmed the domain wall conductivity by using different scanning probe techniques which origins from the nature of inclined domain walls. Moreover, the domains separated by CDW also exhibit distinguishable electrical conductivity due to the interfacial polarization charge with opposite signs. The result provides a novel platform for understanding electrical conductivity behavior of the domains and domain walls in non-oxide ferroelectrics.

cond-mat.mtrl-sci↗

arXiv2020

Ferroelastic switching with van der Waals direction transformation in layered PdSe2 driven by uniaxial and shear strain

Uniaxial and biaxial strain approaches are usually implemented to switch the ferroelastic states, which play a key role in the application of the ferroics and shape memory materials. In this work, by using the first-principles calculations, we found not only uniaxial strain, but also shear strain can induce a novel ferroelastic switching, in which the van der Waals (vdW) layered direction rotates with the ferroelastic transition in layered bulk PdSe2. The shear strain induces ferroelastic switching with three times amplitude smaller than uniaxial strain. The novel three-states ferroelastic switching in layered PdSe2 also occurs under shear strain. Our result shows that the shear strain could be used as an effective approach for manipulating the functionalities of layered materials in potential device applications.

cond-mat.mtrl-sci↗

arXiv2019

Thickness-dependent in-plane polarization and structural phase transition in van der Waals Ferroelectric CuInP2S6

Van der Waals (vdW) layered materials have rather weaker interlayer bonding than the intra-layer bonding, therefore the exfoliation along the stacking direction enables the achievement of monolayer or few layers vdW materials with emerging novel physical properties and functionalities. The ferroelectricity in vdW materials recently attracts renewed interest for the potential use in high-density storage devices. As the thickness going thinner, the competition between the surface energy, depolarization field and interfacial chemical bonds may give rise to the modification of ferroelectricity and crystalline structure, which has limited investigations. In this work, combining the piezoresponse force microscope scanning, contact resonance imaging, we report the existence of the intrinsic in-plane polarization in vdW ferroelectrics CuInP2S6 (CIPS) single crystals, whereas below a critical thickness between 90-100 nm, the in-plane polarization disappears. The Young's modulus also shows an abrupt stiffness at the critical thickness. Based on the density functional theory calculations, we ascribe these behaviors to a structural phase transition from monoclinic to trigonal structure, which is further verified by transmission electron microscope technique. Taken together, these findings demonstrate the foundational importance of structural phase transition for enhancing the rich functionality and broad utility of vdW ferroelectrics.

cond-mat.mtrl-sci↗

arXiv2019

Spontaneous photo-generated carrier separation of SnO/BiOX (X=Cl, Br, I) bilayer under visible light irradiation for water splitting

Alloying in 2D materials plays a more and more important role due to wide range bandgap tunability and integrating the advantages of HER and OER. Here, the novel bilayers of SnO/BiOX (X= Cl, Br and I) bilayer are constructed to integrate the advantages of narrow bandgap and separating photo-generated carriers. The bandgap of the bilayers can be tuned from 1.09 to 1.84 eV, remarkably improving the utilization of solar energy. The large difference in effective masses and built-in electric field effectively hamper the fast recombination of photo-generated carries, which highly enhances the photocatalytic efficiency. Besides that, the type-II band alignment guarantee the two half reactions could occur at different surfaces. Moreover, the optical absorption (the strong transition between band edges and high joint density of states) and band-edge level further confirm the SnO/BiOX (X= Cl and Br) bilayer is a promising candidate for overall water-splitting.

cond-mat.mtrl-sci↗

arXiv2019

Lithium intercalation drives remarkable mechanical properties deterioration in bulk and single-layered phosphorus: A first-principles study

It is of critical importance to understand the mechanical properties change of electrode materials during lithium intercalation in the mechanical design of Li-ion batteries, for the purpose of the high reliability and safety in their applications. Here, we investigated the mechanical properties of both bulk and single layer phosphorus during the lithium intercalation process by using the first-principles calculations. Our results show that the Young's modulus of bulk and layered phosphorus strongly depends on the lithium intercalation. The mechanical bearing capacities, such as critical strain and stress, are significantly reduced by several times after lithium intercalation in both bulk and single layer phosphorus, which may reduce the reliability of Li-ion batteries. Our findings suggest that this remarkable mechanical properties deterioration during Li intercalation should be considered carefully in the mechanical design of Li-ion batteries, in order to keep they working reliably and safely in the charge-discharge process.

cond-mat.mtrl-sci↗

arXiv2019

The coexistence of magnetism and ferroelectricity in 3d transition metal doped SnTe monolayer

The realization of multiferroicity in 2D nanomaterials is crucially important for designing advanced nanoelectronic devices such as non-volatile multistate data storage. In this work, the coexistence of ferromagnetism and ferroelectricity is reported in monolayer SnTe system by transition metal (TM) doping. Based on first-principles calculations, the spontaneous spin polarization could be realized by TM doping in ferroelectric SnTe monolayer. In addition to in-plane ferroelectric polarization, the out-of-plane ferroelectric polarization emerges in Mn (Fe)-doped SnTe monolayer due to the internal displacement of TM from the surface. Interestingly, the crystalline field centered on TM and interaction between the dopant and Te gradually enhanced with the increment of atomic number of doping elements, which explains why the formation energy decreases. The realization of multiferroics in SnTe monolayer could provide theoretical guidance for experimental preparation of low-dimensional multiferroic materials.

cond-mat.mtrl-sci↗

arXiv2018

High Curie temperature half metallic 2D M2Se3 (M = Co, Ni, and Pd) monolayers with superior mechanical flexibility

Pursuing two-dimensional (2D) intrinsic ferromagnetism with high Curie temperature and great mechanical flexibility has attracted great interest in flexible spintronics. In the present work, we carried out a density functional theory (DFT) investigation on the 2D M2Se3 (M=Co, Ni and Pd) monolayers to understand their structural stabilities, electronic, magnetic and mechanical properties. Our results show that the Co2Se3 monolayer exhibits a fascinating half-metallic ferromagnetism with high Curie temperature (>700K). In addition, due to their unique buckling hinge-like structure, M2Se3 monolayers possess the large out-of-plane negative Poisson's ratio (NPR) and superior mechanical flexibility evidenced by their unusual critical strain two times greater than the well-known 2D materials. These findings imply that 2D M2Se3 family is the promising materials for the applications in the flexible and high-density spintronic nanodevices.

cond-mat.mtrl-sci↗

arXiv2018

Non-monotonic thickness dependence of Curie temperature and ferroelectricity in Two-dimensional SnTe film

Recently, the observation of atomic thin film SnTe with a Curie temperature (Tc) higher than that of the bulk (Chang et. al., Science 353, 274 (2016)) has boosted the research on two-dimensional (2D) ferroic materials tremendously. However, the origin of such phenomenon has yet been thoroughly investigated, which hinder the understanding and design of novel materials with ferroic orders at 2D limit. By using the density functional theory, we investigated the structural and ferroelectrical properties of 2D SnTe, to reveal the thickness dependence. The calculated results demonstrate that the 2D SnTe automatically transform into periodical bilayer structure, resulting from the surface effect. Moreover, based on the double-well potential and atomic distortion analysis, we found the Tc of the 2D SnTe is higher than the bulk counterpart, and more surprisingly, the Tc exhibits an unusual non-monotonous dependence of thickness, featuring a pronounced atomic distortion and Curie temperature maximum at 8 atomic-layers. In addition, this non-monotonous dependence is sensitive to the external strain and it can be tuned easily by the external compressive strain.

cond-mat.mtrl-sci↗