Differentially private stochastic gradient descent (DP-SGD) injects noise into every updated coordinate, making the injected noise energy scale with the ambient parameter dimension \(d\). We ask when private training can update fewer coordinates without losing the signal needed for optimization. We propose \textsc{TP-TopK} (Two-Phase TopK DP-SGD), a two-phase method for coordinate-sparse private training without public data, in which a private warm-up phase identifies a coordinate support used to guide the main training phase. We give a criterion characterizing when coordinate restriction can be beneficial, show via a nonconvex stationarity bound that under this condition the relevant noise term scales with the active dimension \(k\) rather than the full parameter dimension \(d\), and provide a lower bound on the reliability of warm-up-based coordinate ranking. Experiments on MNIST, FMNIST, and CIFAR-10 show that learned coordinate supports can retain more gradient energy than size-matched random supports, with the largest gains when the active dimension is small and warm-up scores are informative.
Feed-forward 3D Gaussian Splatting methods reconstruct a scene from posed or pose-free images in a single forward pass, yet current approaches predict one Gaussian per input pixel, tying the representation budget to camera resolution rather than scene complexity. A flat wall and a richly textured object thus produce equally many Gaussians despite very different geometric needs. We propose ZipSplat, a token-based feed-forward model that decouples Gaussian placement from the pixel grid. A multi-view backbone extracts dense visual tokens, and k-means clustering compresses them into a compact set of scene tokens. Cross- and self-attention refine these tokens, and a lightweight MLP decodes each into a group of Gaussians with unconstrained 3D positions. Because clustering is applied at inference, a single trained model spans the quality-efficiency curve without retraining. ZipSplat operates without ground-truth poses or intrinsics, yet sets a new state of the art on DL3DV and RealEstate10K with ${\sim}6{\times}$ fewer Gaussians than pixel-aligned methods, surpassing the best pose-free baseline by 2.1dB and 1.2dB PSNR, respectively. It further generalizes zero-shot to Mip-NeRF360 and Scan
Gaussian splatting techniques have shown promising results in novel view synthesis, achieving high fidelity and efficiency. However, their high reconstruction quality comes at the cost of requiring a large number of primitives. We identify this issue as stemming from the entanglement of geometry and appearance in Gaussian Splatting. To address this, we introduce a neural shell texture, a global representation that encodes texture information around the surface. We use Gaussian primitives as both a geometric representation and texture field samplers, efficiently splatting texture features into image space. Our evaluation demonstrates that this disentanglement enables high parameter efficiency, fine texture detail reconstruction, and easy textured mesh extraction, all while using significantly fewer primitives.
This paper proposes a novel consistent δ+- Updated Lagrangian Particle Hydrodynamics (ULPH) model. Although the Smoothed Particle Hydrodynamics (SPH) model has gained recognized achievements, it is afflicted by excessive numerical dissipation when the neighboring particles are insufficient. The present proposed consistent δ+-ULPH model has advantages in overcoming this problem. To improve the accuracy, efficiency, stability, and energy conservation, several new techniques are introduced to the consistent δ+-ULPH model. A novel extended support domain technique is proposed to achieve higher accuracy with fewer neighboring particles. An optimal matrix for the velocity divergence is proposed to improve the free-surface stability. A consistent particle shifting technique for the ULPH scheme is proposed to maintain a uniform and regular particle distribution and obtain superior conservation. In addition, an acoustic damper term for the ULPH scheme is introduced to improve the pressure field stability. Five benchmark tests were carried out to validate the consistent δ+-ULPH model. The conventional ULPH and the consistent δ+-SPH results are presented for comparison. Results indicate that
This paper investigates the fundamental relationship between model capacity and the minimal number of visual tokens required to preserve image semantics. Inspired by the Minimum Description Length principle, we reinterpret image tokens as vectors in a visual semantic space and define the intrinsic semantic complexity of an image as the smallest set of basis vectors needed to span this space. Building on this perspective, we propose Orthogonal Filtering, a lightweight module that adaptively clusters redundant tokens into a compact set of orthogonal bases. Through extensive experiments across a range of ViT models, we reveal a consistent token, model scaling law: larger models require significantly fewer tokens to span visual semantic space. Besides, we also contribute a visual long-context dataset.
The radial positions of galaxies inferred from their measured redshift appear distorted due to their peculiar velocities. We argue that the contribution from stochastic velocities -- which gives rise to Fingers-of-God (FoG) anisotropy in the inferred maps -- does not lend itself to perturbative modelling already on scales targeted by current experiments. To get around this limitation, we propose to remove FoG using data-driven indicators of their abundance that are local in nature and thus avoid selection biases. In particular, we show that the scale where the measured power spectrum quadrupole changes sign is tightly anti-correlated with both the satellite fraction and the velocity dispersion, and can thus be used to select galaxy samples with fewer FoG. In addition, we show that excluding galaxies in haloes more massive than a given mass threshold can help to discard many of the most problematic galaxies. Such selection could be achieved in practice using maps of the thermal Sunyaev-Zeldovich distortion of the cosmic microwave background frequency spectrum. These techniques could potentially improve reconstructions of the large-scale velocity and displacement fields from the reds
In modern software ecosystems, dependency management plays a critical role in ensuring secure and maintainable applications. However, understanding the relationship between release practices and their impact on vulnerabilities and update cycles remains a challenge. In this study, we analyze the release histories of 10,000 Maven artifacts, covering over 203,000 releases and 1.7 million dependencies. We evaluate how release speed affects software security and lifecycle. Our results show an inverse relationship between release speed and dependency outdatedness. Artifacts with more frequent releases maintain significantly shorter outdated times. We also find that faster release cycles are linked to fewer CVEs in dependency chains, indicating a strong negative correlation. These findings emphasize the importance of accelerated release strategies in reducing security risks and ensuring timely updates. Our research provides valuable insights for software developers, maintainers, and ecosystem managers.
Masked diffusion language models (MDLMs) have recently emerged as a promising alternative to autoregressive (AR) language models, offering properties such as parallel decoding, flexible generation orders, and the potential for fewer inference steps. Despite these advantages, decoding strategies and reinforcement learning (RL) algorithms tailored for MDLMs remain underexplored. A naive approach is to directly transfer techniques well-established for AR models to MDLMs. However, this raises an immediate question: Is such a naive transfer truly optimal? For example, 1) Block-wise and semi-AR decoding strategies are not employed during the training of MDLMs, so why do they outperform full diffusion-style decoding during inference? 2) Applying RL algorithms designed for AR models directly to MDLMs exhibits a training-inference inconsistency, since MDLM decoding are non-causal (parallel). This results in inconsistencies between the rollout trajectory and the optimization trajectory. To address these challenges, we propose EOS Early Rejection (EOSER) and Ascending Step-Size (ASS) decoding scheduler, which unlock the potential of MDLMs to perform full diffusion-style decoding, achieving co
In 2008 Chen and Chvátal conjectured that any metric space on n points has at least n lines, unless all the points belong to one line. Chv\atal proved in 2014 that this is indeed the case for metric spaces with distances 0, 1 and 2. In this work, we prove that there exists a family of ten graphs such that a metric space defined by a graph of diameter two has fewer lines than points if and only if the associated graph belongs to that family.
We present Exact Causal Attention (ECA), a Strassen-style algorithm that computes exact Causal Attention using 10\% fewer operations. ECA improves a special class of matrix multiplications where either one operand or the output matrix is upper- or lower-triangular. This includes all matrix multiplication operations in the forward and backward pass of Causal Attention, such as masked product $\mathrm{Mask}(QK^{T})$. ECA is built upon algebraic identities discovered via machine learning and combinatorial search. We note that ECA cannot accelerate fused kernels such as FlashAttention on GPU. This is because ECA requires materialization of large intermediate expressions in the memory, while FlashAttention does not. However, it provides an alternative approach for compute-bound applications and can potentially be useful in scenarios with FLOPs considerations.
Amidst the advancements in image-based Large Vision-Language Models (image-LVLM), the transition to video-based models (video-LVLM) is hindered by the limited availability of quality video data. This paper addresses the challenge by leveraging the visual commonalities between images and videos to efficiently evolve image-LVLMs into video-LVLMs. We present a cost-effective video-LVLM that enhances model architecture, introduces innovative training strategies, and identifies the most effective types of video instruction data. Our innovative weighted token sampler significantly compresses the visual token numbers of each video frame, effectively cutting computational expenses. We also find that judiciously using just 10% of the video data, compared to prior video-LVLMs, yields impressive results during various training phases. Moreover, we delve into the influence of video instruction data in limited-resource settings, highlighting the significance of incorporating video training data that emphasizes temporal understanding to enhance model performance. The resulting Fewer Tokens and Fewer Videos LVLM (FTFV-LVLM) exhibits exceptional performance across video and image benchmarks, valid
This late-breaking work presents a large-scale analysis of explainable AI (XAI) literature to evaluate claims of human explainability. We collaborated with a professional librarian to identify 18,254 papers containing keywords related to explainability and interpretability. Of these, we find that only 253 papers included terms suggesting human involvement in evaluating an XAI technique, and just 128 of those conducted some form of a human study. In other words, fewer than 1% of XAI papers (0.7%) provide empirical evidence of human explainability when compared to the broader body of XAI literature. Our findings underscore a critical gap between claims of human explainability and evidence-based validation, raising concerns about the rigor of XAI research. We call for increased emphasis on human evaluations in XAI studies and provide our literature search methodology to enable both reproducibility and further investigation into this widespread issue.
XNet introduces a wavelet-based X-shaped unified architecture for fully- and semi-supervised biomedical segmentation. So far, however, XNet still faces the limitations, including performance degradation when images lack high-frequency (HF) information, underutilization of raw images and insufficient fusion. To address these issues, we propose XNet v2, a low- and high-frequency complementary model. XNet v2 performs wavelet-based image-level complementary fusion, using fusion results along with raw images inputs three different sub-networks to construct consistency loss. Furthermore, we introduce a feature-level fusion module to enhance the transfer of low-frequency (LF) information and HF information. XNet v2 achieves state-of-the-art in semi-supervised segmentation while maintaining competitve results in fully-supervised learning. More importantly, XNet v2 excels in scenarios where XNet fails. Compared to XNet, XNet v2 exhibits fewer limitations, better results and greater universality. Extensive experiments on three 2D and two 3D datasets demonstrate the effectiveness of XNet v2. Code is available at https://github.com/Yanfeng-Zhou/XNetv2 .
The adaptive variational quantum dynamics simulation (AVQDS) method performs real-time evolution of quantum states using automatically generated parameterized quantum circuits that often contain substantially fewer gates than Trotter circuits. Here we report an improved version of the method, which we call AVQDS(T), by porting the Tiling Efficient Trial Circuits with Rotations Implemented Simultaneously (TETRIS) technique. The algorithm adaptively adds layers of disjoint unitary gates to the ansatz circuit so as to keep the McLachlan distance, a measure of the accuracy of the variational dynamics, below a fixed threshold. We perform benchmark noiseless AVQDS(T) simulations of quench dynamics in local spin models demonstrating that the TETRIS technique significantly reduces the circuit depth and two-qubit gate count. We also show a method based on eigenvalue truncation to solve the linear equations of motion for the variational parameters with enhanced noise resilience. Finally, we propose a way to substantially alleviate the measurement overhead of AVQDS(T) while maintaining high accuracy by synergistically integrating quantum circuit calculations on quantum processing units with c
Neural speech codecs have gained great attention for their outstanding reconstruction with discrete token representations. It is a crucial component in generative tasks such as speech coding and large language models (LLM). However, most works based on residual vector quantization perform worse with fewer tokens due to low coding efficiency for modeling complex coupled information. In this paper, we propose a neural speech codec named FreeCodec which employs a more effective encoding framework by decomposing intrinsic properties of speech into different components: 1) a global vector is extracted as the timbre information, 2) a prosody encoder with a long stride level is used to model the prosody information, 3) the content information is from a content encoder. Using different training strategies, FreeCodec achieves state-of-the-art performance in reconstruction and disentanglement scenarios. Results from subjective and objective experiments demonstrate that our framework outperforms existing methods.
Superconducting qubits are a key contender for quantum computing elements, but they often face challenges like noise and decoherence from two-level systems (TLS). Tantalum (Ta) qubits are notable for their long T$_1$ coherence times nearing milliseconds, mainly due to fewer TLS, though the cause was unclear. Our research explored this by analyzing the air-oxide interface with density functional theory, particularly comparing Nb oxide (Nb$_2$O$_5$) and Ta oxide (Ta$_2$O$_5$). We discovered that Ta$_2$O$_5$ forms a smoother surface with fewer dangling O atoms and TLS than Nb$_2$O$_5$. The greater atomic mass of Ta also lowers the TLS tunnel splittings below the qubit's operating frequency. Furthermore, using external electric fields or SO$_2$ passivation can significantly reduce TLS on Nb surfaces, potentially improving their coherence times.
Prior work of Beverland et al. has shown that any exact Clifford+$T$ implementation of the $n$-qubit Toffoli gate must use at least $n$ $T$ gates. Here we show how to get away with exponentially fewer $T$ gates, at the cost of incurring a tiny $1/\mathrm{poly}(n)$ error that can be neglected in most practical situations. More precisely, the $n$-qubit Toffoli gate can be implemented to within error $ε$ in the diamond distance by a randomly chosen Clifford+$T$ circuit with at most $O(\log(1/ε))$ $T$ gates. We also give a matching $Ω(\log(1/ε))$ lower bound that establishes optimality, and we show that any purely unitary implementation achieving even constant error must use $Ω(n)$ $T$ gates. We also extend our sampling technique to implement other Boolean functions. Finally, we describe upper and lower bounds on the $T$-count of Boolean functions in terms of non-adaptive parity decision tree complexity and its randomized analogue.
When training data is scarce, it is common to make use of a feature extractor that has been pre-trained on a large base dataset, either by fine-tuning its parameters on the ``target'' dataset or by directly adopting its representation as features for a simple classifier. Fine-tuning is ineffective for few-shot learning, since the target dataset contains only a handful of examples. However, directly adopting the features without fine-tuning relies on the base and target distributions being similar enough that these features achieve separability and generalization. This paper investigates whether better features for the target dataset can be obtained by training on fewer base classes, seeking to identify a more useful base dataset for a given task.We consider cross-domain few-shot image classification in eight different domains from Meta-Dataset and entertain multiple real-world settings (domain-informed, task-informed and uninformed) where progressively less detail is known about the target task. To our knowledge, this is the first demonstration that fine-tuning on a subset of carefully selected base classes can significantly improve few-shot learning. Our contributions are simple a
In this paper, we mainly investigate the detection of quantum states containing fewer than $k$ unentangled particles in multipartite quantum systems. Based on calculations about operators, we derive two practical criteria for judging $N$-partite quantum states owning fewer than $k$ unentangled particles. In addition, we demonstrate the effectiveness of our frameworks through some concrete examples, and specifically point out the quantum states having fewer than $k$ unentangled particles that our methods can detect, while other criteria cannot recognize.
Score-based generative models (SGMs) are a popular family of deep generative models that achieve leading image generation quality. Early studies extend SGMs to tackle class-conditional generation by coupling an unconditional SGM with the guidance of a trained classifier. Nevertheless, such classifier-guided SGMs do not always achieve accurate conditional generation, especially when trained with fewer labeled data. We argue that the problem is rooted in the classifier's tendency to overfit without coordinating with the underlying unconditional distribution. To make the classifier respect the unconditional distribution, we propose improving classifier-guided SGMs by letting the classifier regularize itself. The key idea of our proposed method is to use principles from energy-based models to convert the classifier into another view of the unconditional SGM. Existing losses for unconditional SGMs can then be leveraged to achieve regularization by calibrating the classifier's internal unconditional scores. The regularization scheme can be applied to not only the labeled data but also unlabeled ones to further improve the classifier. Across various percentages of fewer labeled data, empi