Graph clustering is essential in graph analysis for revealing structural patterns and node communities. Despite recent advances in self-supervised contrastive learning that have improved clustering via structural and attribute signals, existing methods still struggle to flexibly capture high-order local structures and often overlook global semantics in complex graphs. These limitations lead to suboptimal node representations, especially in real-world graphs with fragmented structures and ambiguous cluster boundaries. To address these limitations, a contrastive graph clustering framework is proposed to jointly integrate multi-scale local structures with global semantics via attention mechanisms. At the local level, GNN-based topological signals extracted from multiple propagation depths are adaptively fused through attention-based weighting to capture multi-scale neighborhood features. At the global level, semantic prototypes derived from dynamically evolving cluster centers are adaptively aggregated through attention to guide node representations and enhance inter-cluster separability. The model is trained under a dual-view contrastive learning paradigm with a hybrid objective that
As large language model (LLM) agents are applied to longer tasks, they increasingly modify workspace state across multiple rounds of iteration. However, agents typically observe only tool outputs and log fragments, while the actual state changes occur in the file system. Without explicit workspace boundaries, state-changing operations such as file writes and temporary artifact generation may scatter changes across paths. Over time, these weakly constrained changes accumulate, making states such as modified files difficult to track. This paper presents LemonHarness, an integrated execution framework for long-horizon agents. LemonHarness establishes an explicit execution boundary by constraining state-changing operations within a clearly defined workspace and bringing model invocation, tool execution, and rule knowledge within a single controlled boundary. State-changing operations, including file writes, dependency installation, and temporary artifact creation, are executed through structured tool interfaces, with execution feedback recorded as observations available to subsequent model decisions. The system also introduces a reusable rule knowledge base, which turns recurring execu
In point clouds, noise directly perturbs point coordinates that encode both spatial location and geometry, making one-to-one correspondence construction more challenging than in images. Existing methods impose statistical mappings across noisy variants via noise or optimal transport, but suffer from correspondence ambiguity. In this work, we propose Self-Induced Mirror-Point Consistency (SIMPC) to learn deterministic correspondences between points and the underlying surface in an unsupervised manner. For each noisy point, SIMPC generates a mirror-point on the opposite side of the underlying surface, guided by geometric priors during the denoising process. By encouraging consistency between the denoising targets of the original point and its mirror counterpart, SIMPC effectively localizes the position of underlying surface. Extensive experiments on synthetic and real-world datasets demonstrate that SIMPC significantly outperforms state-of-the-art unsupervised methods and surpasses several strong supervised counterparts.
Recent progress in diffusion-based Singing Voice Synthesis (SVS) demonstrates strong expressiveness but remains limited by data scarcity and model scalability. We introduce a two-stage pipeline: a compact seed set of human-sung recordings is constructed by pairing fixed melodies with diverse LLM-generated lyrics, and melody-specific models are trained to synthesize over 500 hours of high-quality Chinese singing data. Building on this corpus, we propose DiTSinger, a Diffusion Transformer with RoPE and qk-norm, systematically scaled in depth, width, and resolution for enhanced fidelity. Furthermore, we design an implicit alignment mechanism that obviates phoneme-level duration labels by constraining phoneme-to-acoustic attention within character-level spans, thereby improving robustness under noisy or uncertain alignments. Extensive experiments validate that our approach enables scalable, alignment-free, and high-fidelity SVS.
The realization of ambient-pressure superconductivity in La$_3$Ni$_2$O$_7$ thin films raises a fundamental question: is the metallic ground state driven by lattice strain or interfacial charge reconstruction? Using fully self-consistent DFT+$U$ calculations on La$_3$Ni$_2$O$_7$/SrLaAlO$_4$ heterostructures, we identify that intrinsic hole doping via interfacial Sr interdiffusion is the decisive factor in stabilizing the metallic state. Our 1-unit-cell model accurately reproduces the ARPES-observed Fermi surface, particularly the critical Ni-$d_{z^2}$ derived $γ$ hole pocket, which originates exclusively from the interface-proximal bilayer. Furthermore, comparative tight-binding analysis suggests that the reduced superconducting transition temperature ($T_c$) in thin films stems from the synergistic suppression of the electronic density of states (DOS) and vertical superexchange coupling ($J \perp Z$). These findings highlight that interface engineering plays a critical role beyond simple strain imposition in modulating nickelate orbital physics.
Real-world environment-derived point clouds invariably exhibit noise across varying modalities and intensities. Hence, point cloud denoising (PCD) is essential as a preprocessing step to improve downstream task performance. Deep learning (DL)-based PCD models, known for their strong representation capabilities and flexible architectures, have surpassed traditional methods in denoising performance. To our best knowledge, despite recent advances in performance, no comprehensive survey systematically summarizes the developments of DL-based PCD. To fill the gap, this paper seeks to identify key challenges in DL-based PCD, summarizes the main contributions of existing methods, and proposes a taxonomy tailored to denoising tasks. To achieve this goal, we formulate PCD as a two-step process: outlier removal and surface noise restoration, encompassing most scenarios and requirements of PCD. Additionally, we compare methods in terms of similarities, differences, and respective advantages. Finally, we discuss research limitations and future directions, offering insights for further advancements in PCD.
The imperfect modeling of ternary complexes has limited the application of computer-aided drug discovery tools in PROTAC research and development. In this study, an AI-assisted approach for PROTAC molecule design pipeline named LM-PROTAC was developed, which stands for language model driven Proteolysis Targeting Chimera, by embedding a transformer-based generative model with dual constraints on structure and properties, referred to as the DCT. This study utilized the fragmentation representation of molecules and developed a language model driven pipeline. Firstly, a language model driven affinity model for protein compounds to screen molecular fragments with high affinity for the target protein. Secondly, structural and physicochemical properties of these fragments were constrained during the generation process to meet specific scenario requirements. Finally, a two-round screening of the preliminary generated molecules using a multidimensional property prediction model to generate a batch of PROTAC molecules capable of degrading disease-relevant target proteins for validation in vitro experiments, thus achieving a complete solution for AI-assisted PROTAC drug generation. Taking the
Recently, room-temperature superconductor has been claimed in a nitrogen-doped lutetium hydride at near-ambient pressure [Nature 615, 244 (2023)]. Using X-ray diffraction (XRD) and Raman spectra analysis, the authors believed that the superconducting properties can most probably be attributed to Fm-3m-LuH3δNε. Here, we systematic study the phase diagram of Lu-N-H at 1 GPa by using first-principle theory and find that there have no thermodynamically stable ternary compounds. Besides, we analyzed the dynamically stability and superconducting properties of N-doped Fm-3m-LuH3 using virtual crystal approximation (VCA) and supercell method. Our theoretical results show that the Tc of N-doped LuH3 cannot reach the level of room-temperature.
The stability of different stoichiometric H$_n$Br ($n$=1-7) compounds under pressure are extensively studied using density functional theory calculations. Five new energetically stable stoichiometries of H$_2$Br, H$_3$Br, H$_4$Br, H$_5$Br, and H$_7$Br were uncovered at high pressure. The results show that HBr is stable below 64 GPa, then decomposes into new compound H$_2$Br and Br$_2$ molecular crystal. For H$_2$Br and H$_3$Br compounds, they were found to become stable above 30 GPa and 8 GPa, respectively. In addition, we accidentally discovered the triangular H$_3^+$ species in H$_5$Br compounds at 100 GPa. Further electron-phonon coupling calculations predicted that hydrogen-rich H$_2$Br and H$_4$Br compounds are superconductors with critical temperature of superconductivity $T_c$ of 12.1 K and 2.4 K at 240 GPa, respectively.
We have systematically searched for the ground state structures of bismuth hydrides based on evolutionary algorithm method and particle swarm optimization algorithm method. Given only rich-hydrogen region, except for BiH$_{3}$, other hydrides (BiH, BiH$_{2}$, BiH$_{4}$, BiH$_{5}$, BiH$_{6}$) have been predicted to be stable with pressurization. With the increase of hydrogen content, hydrogen exists in bismuth hydrides with the different forms and presents the characteristics of ionicity. Under high pressure, the remarkable structural feature is the emergence of H$_{2}$ units in BiH$_{2}$, BiH$_{4}$ and BiH$_{6}$, and BiH$_{6}$ adopts a startling layered structure intercalated by H$_{2}$ and the linear H$_{3}$ units. Further calculations show these energetically stable hydrides are good metal and their metallic pressures are lower than that of pure solid hydrogen because of the doping impurities. The $T_{c}$ in the range of 20-119 K has been calculated by the Allen-Dynes modified McMillan equation, which indicates all these stable hydrides are potential high-temperature superconductors. Remarkably, it is the H-Bi-H and Bi atoms vibrations rather than the high-frequency H$_{2}$ or H$
The semimetal antimony (Sb) element doped into hydrogen has been performed theoretically to explored high-pressure crystal structure and superconductivity of antimony hydrides. The unexpected stoichiometry $\textrm{SbH}_\textrm{4}$ with $P6_3/mmc$ symmetry is found to have most negative enthalpy and embody the coexistence of covalent and ionic bonds. It is a metallic phase and stable in the pressure ranges of 127-300 GPa. Furthermore, a superconducting critical temperature ($T_c$) of 106 K is obtained at 150 GPa by employing the Allen-Dynes modified McMillan equation. In addition, an extrusive distinguishing feature is the presence of soft phonon modes, which is primary contribution to the strength of electron-phonon coupling.
High pressure structures, phase diagram and superconductivity of polonium hydrides have been systematically investigated through the first-principles calculations based on the density functional theory. With the increasing pressure, several stoichiometries (PoH, $\textrm{PoH}_\textrm{2}$, $\textrm{PoH}_4$ and $\textrm{PoH}_6$) are predicted to stabilize in the excess hydrogen environment. All of the reported hydrides, exception of PoH, exhibit intriguing structural character with the appearing $\textrm{H}_2$ units. Moreover, our electronic band structure and the projected density of states (PDOS) demonstrate that these energetically stable phases are metallic. The application of the Allen-Dynes modified McMillan equation with the calculated electron-phonon coupling parameter reveals that $\textrm{PoH}_4$ is a superconductor with a critical temperature $T_c$ of 41.2-47.2 K at 300 GPa.
Solid hydrogen sulfide is well known as a typical molecular crystal but its stability under pressure is still under debate. Particularly, Eremets et al. found the high pressure superconductivity with $T_{c}\approx$ 190 K in a H$_{2}$S sample [arXiv: 1412.0460 (2014)] which is associates with the elemental decomposition into H$_{3}$S [Sci. Rep. 4, 6968 (2014)]. Therefore, on what pressure H$_{2}$S can decompose and which kind of the products of decomposition urgent need to be solved. In this paper, we have performed an extensive structural study on different stoichiometries H$_{n}$S with ${n> 1}$ under high pressure using $ab$ $initio$ calculations. Our results show that H$_{2}$S is stable below 50 GPa and decomposes into H$_3$S and sulfur at high pressure, while H$_{3}$S is stable at least up to 300 GPa. The other hydrogen-rich H$_{4}$S, H$_{5}$S, and H$_{6}$S are unstable in the pressure range from 20 to 300 GPa.
In this paper, we propose a novel top-down instance segmentation framework based on explicit shape encoding, named \textbf{ESE-Seg}. It largely reduces the computational consumption of the instance segmentation by explicitly decoding the multiple object shapes with tensor operations, thus performs the instance segmentation at almost the same speed as the object detection. ESE-Seg is based on a novel shape signature Inner-center Radius (IR), Chebyshev polynomial fitting and the strong modern object detectors. ESE-Seg with YOLOv3 outperforms the Mask R-CNN on Pascal VOC 2012 at mAP$^r$@0.5 while 7 times faster.