arXiv Science⌕ Search

arXiv subjects

Shaobo Chen

Publications and source records attributed to Shaobo Chen.

10 recordsLinked to original sources

Expanding the Decision Space of Pretrained Models through Structured Re-Decoding

Native decoders limit how pretrained models combine local predictions under new structural constraints. We introduce Structured Re-Decoding (SRD), a training-free framework that constructs local candidates from model scores, samples, or external specifications, jointly optimizes their selection, and verifies reconstructed outputs. We show that candidate recombination expands feasible support and can improve attainable reward under a fixed KL divergence budget. Across molecular generation, protein design, RNA structure prediction, and antibody modeling, SRD enables improved structured inference, including verified molecules absent from the original sample pool. External chemical templates further extend SRD to noncanonical amino acids, peptide cyclization, and grammar-guided molecular construction. Preliminary experiments on a physical coherent Ising machine demonstrate the feasibility of hardware-based optimization, suggesting broader opportunities for quantum and quantum-inspired approaches to structured inference and combinatorial optimization across scientific domains.

cs.AI↗

Rethinking Prediction Horizons: Decoupling Model Queries from State Updates in Few-Step MeanFlow

MeanFlow predicts interval-average velocities, but conventional few-step sampling ties the interval queried from the model to the duration of the state update. We separate these two quantities at inference, fixing the model, update grid, and number of queries while varying the prediction horizon. Our central finding is that the horizon response depends on both query position and the states entering that query. Across seven independently trained CIFAR-10 checkpoints at classifier-free guidance (CFG) scale 2, a longer horizon worsens FID at early steps but improves it at the midpoint. To probe whether query time alone determines this response, we fix the midpoint query and current/subsequent guidance while changing only earlier CFG. In held-out factorial outcomes on five previously studied checkpoints, the mean midpoint horizon benefit increases from 0.109 to 1.308 FID, indicating sensitivity to the incoming-state distribution. A frozen four-query rule reduces mean CIFAR-10 FID by 2.023 without retraining or extra queries, including gains on five checkpoints trained after rule selection. A milder rule also improves mean FID by 1.041 on a stronger frozen grid, with fixed-rule FID gains on the evaluated public ImageNet-256 and AFHQ checkpoints. However, matched-budget calibration favors query-start shifting in historical CIFAR tests, and the tested Horizon allocation adds no FID benefit beyond Grid+Start on two public ImageNet models. Thus, prediction horizon is a distinct, position- and state-sensitive inference control, but not a universally preferred one.

cs.LG↗

SplatQ: Interaction-Aware Pruning of 3D Gaussian Splatting via Quadratic Selection

Existing 3D Gaussian Splatting (3DGS) pruning methods primarily rely on global Top-K selection based on independent importance scores, lacking regional budget control and ignoring joint deletion effects. Consequently, they cannot ensure balanced reconstruction quality across a scene. We propose SplatQ, a combinatorial pruning algorithm that coordinates regional budget allocation and joint Gaussian selection under a fixed budget. SplatQ partitions Gaussians by spatial proximity and appearance similarity into size-limited blocks, decomposing global selection into subproblems of a specific size. It then smooths importance-based block retention quotas to prevent excessive budget concentration. Finally, it constructs a quadratic approximation of joint deletion effects using the responses of representative pixel colors to changes in Gaussian opacity. Individual responses, pairwise interactions, and quota penalties are incorporated into a quadratic unconstrained binary optimization (QUBO) objective, solved using simulated annealing (SA) or a coherent Ising machine (CIM) to jointly select Gaussians. Experiments on 5 benchmarks demonstrate that SplatQ outperforms state-of-the-art pruning methods in aggregate reconstruction quality, spatial quality balance, and rendering speed. Ablations confirm that budget smoothing improves local detail reconstruction, while interaction modeling improves Gaussian selection quality under a fixed budget.

cs.CV↗

Kaiwu-PyTorch-Plugin: Bridging Deep Learning and Photonic Quantum Computing for Energy-Based Models and Active Sample Selection

This paper introduces the Kaiwu-PyTorch-Plugin (KPP) to bridge Deep Learning and Photonic Quantum Computing across multiple dimensions. KPP integrates the Coherent Ising Machine into the PyTorch ecosystem, addressing classical inefficiencies in Energy-Based Models. The framework facilitates quantum integration in three key aspects: accelerating Boltzmann sampling, optimizing training data via Active Sampling, and constructing hybrid architectures like QBM-VAE and Q-Diffusion. Empirical results on single-cell and OpenWebText datasets demonstrate KPPs ability to achieve SOTA performance, validating a comprehensive quantum-classical paradigm.

quant-ph↗

A versatile coherent Ising computing platform

Coherent Ising Machines (CIMs) have emerged as a hybrid form of quantum computing devices designed to solve NP-complete problems, offering an exciting opportunity for discovering optimal solutions. Despite challenges such as susceptibility to noise-induced local minima, we achieved notable advantages in improving the computational accuracy and stability of CIMs. We conducted a successful experimental demonstration of CIM via femto-second laser pumping that integrates optimization strategies across optical and structural dimensions, resulting in significant performance enhancements. The results are particularly promising. An average success rate of 55% was achieved to identify optimal solutions within a Mobius Ladder graph comprising 100 vertices. Compared with other alternatives, the femto-second pulse results in significantly higher peak power, leading to more pronounced quantum effects and lower pump power in optical fiber based CIMs. In addition, we have maintained an impressive success rate for a continuous period of 8 hours, emphasizing the practical applicability of CIMs in real-world scenarios. Furthermore, our research extends to the application of these principles in practical applications such as molecular docking and credit scoring. The results presented substantiate the theoretical promise of CIMs, paving the way for their integration into large-scale practical applications.

quant-ph↗

Quantum-Boosted High-Fidelity Deep Learning

A fundamental limitation of probabilistic deep learning is its predominant reliance on Gaussian priors. This simplistic assumption prevents models from accurately capturing the complex, non-Gaussian landscapes of natural data, particularly in demanding domains like complex biological data, severely hindering the fidelity of the model for scientific discovery. The physically-grounded Boltzmann distribution offers a more expressive alternative, but it is computationally intractable on classical computers. To date, quantum approaches have been hampered by the insufficient qubit scale and operational stability required for the iterative demands of deep learning. Here, we bridge this gap by introducing the Quantum Boltzmann Machine-Variational Autoencoder (QBM-VAE), a large-scale and long-time stable hybrid quantum-classical architecture. Our framework leverages a quantum processor for efficient sampling from the Boltzmann distribution, enabling its use as a powerful prior within a deep generative model. Applied to million-scale single-cell datasets from multiple sources, the QBM-VAE generates a latent space that better preserves complex biological structures, consistently outperforming conventional Gaussian-based deep learning models like VAE and SCVI in essential tasks such as omics data integration, cell-type classification, and trajectory inference. It also provides a typical example of introducing a physics priori into deep learning to drive the model to acquire scientific discovery capabilities that breaks through data limitations. This work provides the demonstration of a practical quantum advantage in deep learning on a large-scale scientific problem and offers a transferable blueprint for developing hybrid quantum AI models.

cs.LG↗

Spin ordering-induced fully-compensated ferrimagnetism

Fully-compensated ferrimagnets exhibit zero net magnetic moment yet display non-relativistic global spin splitting, making them highly advantageous for constructing high-performance spintronic devices. The general strategy is to break the inversion symmetry of conventional antiferromagnets or the rotational/mirror symmetry of altermagnets to achieve fully-compensated ferrimagnets. Here, we propose to induce fully-compensated ferrimagnetism by engineering the spin ordering rather than modifying the lattice structure. Bilayer stacking engineering offers a convenient platform to verify our proposal and readily enables switching between two distinct electronic states by tuning the $\mathrm{N\acute{e}el}$ vector of one layer. By the first-principles calculations, a bilayer system is constructed with monolayer $\mathrm{Cr_2C_2S_6}$ as the elementary building block to corroborate our proposal. This strategy can also be extended to inducing altermagnetism via spin ordering engineering. Our work offers an alternative route to realize non-relativistic spin splitting in zero-net-magnetization magnets, paving the way for the advancement and construction of low-power spintronic device.

cond-mat.mtrl-sci↗

Anomalous valley Hall effect in electric-potential-difference antiferromagnetic $\mathrm{Cr_2CHCl}$ monolayer

The antiferromagnetic (AFM) valleytronics can be intrinsically more energy-saving and fast-operating in device applications. In general, the lacking spontaneous spin-splitting hinders the implementation and detection of anomalous valley Hall effect (AVHE). Here, we propose to implement AVHE in electric-potential-difference antiferromagnetic $\mathrm{Cr_2CHCl}$ monolayer with excellent stability, where the spontaneous spin-splitting can be induced due to layer-dependent electrostatic potential caused by out-of-plane built-in electric field. From a symmetry perspective, the introduction of Janus structure breaks the combined symmetry ($PT$ symmetry) of spatial inversion ($P$) and time reversal ($T$), which gives rise to spin-splitting. Both unstarined and strained monolayer $\mathrm{Cr_2CHCl}$ possess valley splitting of larger than 51 meV, which is higher than the thermal energy of room temperature (25 meV). The layer-locked Berry curvature gives rise to layer-locked AVHE. Our work reveals a route to achieve AVHE in AFM monolayer with spontaneous spin-splitting.

cond-mat.mtrl-sci↗

Intrinsic persistent spin helix in 2D T-XY (X$\neq$Y=P, As, Sb and Bi)

The persistent spin helix (PSH) is robust against spin-independent scattering and renders an extremely long spin lifetime, which can improve the performance of potential spintronic devices. To achieve the PSH, a unidirectional spin configuration is required in the momentum space. Here, T-XY (X$\neq$Y=P, As, Sb and Bi) monolayers with dynamical, mechanical and thermal stabilities are predicted to intrinsically possess PSH. Due to the $C_{2\upsilon}$ point-group symmetry, a unidirectional spin configuration is preserved in the out-of-plane direction for both conduction and valence bands around the high-symmetry $Γ$ point. That is, the expectation value of the spin $S$ only has the out-of-plane component $S_z$. The application of an out-of-plane external electric field can induce in-plane components $S_x$ and $S_y$, thus offering a promising platform for the on-off logical functionality of spin devices. Our work reveals a new family of T-phase two-dimensional (2D) materials, which could provide promising applications in spintronic devices.

cond-mat.mtrl-sci↗

Proposal for valleytronic materials: ferrovalley metal and valley gapless semiconductor

Valleytronic materials can provide new degrees of freedom to future electronic devices. In this work, the concepts of the ferrovalley metal (FVM) and valley gapless semiconductor (VGS) are proposed, which can be achieved in valleytronic bilayer systems by electric-field tuning, where the interaction between out-of-plane ferroelectricity and A-type antiferromagnetism can induce layer-polarized anomalous valley Hall (LP-AVH) effect. The K and -K valleys of FVM are both metallic, and electron and hole carriers simultaneously exist. In the extreme case, the FVM can become VGS by analogizing spin gapless semiconductor (SGS). Moreover, it is proposed that the valley splitting enhancement and valley polarization reversal can be achieved by electric field in valleytronic bilayer systems. Taking the bilayer $\mathrm{RuBr_2}$ as an example, our proposal is confirmed by the first-principle calculations. The FVM and VGS can be achieved in bilayer $\mathrm{RuBr_2}$ by applying electric field. With appropriate electric field range, increasing electric field can enhance valley splitting, and the valley polarization can be reversed by flipping electric field direction. To effectively tune valley properties by electric field in bilayer systems, the parent monolayer should possess out-of-plane magnetization, and have large valley splitting. Our results shed light on the possible role of electric field in tuning valleytronic bilayer systems, and provide a way to design the ferrovalley-related material by electric field.

cond-mat.mtrl-sci↗