arXiv subjects

Yiming Dong

Publications and source records attributed to Yiming Dong.

At least 19 recordsLinked to original sources

arXiv2026

On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability

We describe the architecture and ablations of Qwen3.8-Flash-Next, a sparse mixture-of-experts model with 125B parameters, 6B activated per token, and additional 51B parameters of n-gram embedding tables held off the accelerator. On fourteen pre-training benchmarks the model leads the 397B-A17B predecessor on eight and trails it on the rest by at most 2.6 points, at 1/3 the activated parameters, 1/3 the training tokens, and roughly 1/9 the training FLOPs. Token mixing uses a layer-wise hybrid of Gated DeltaNet (GDN) and global attention, with one full-attention layer in every four; at continued-pretraining time those full-attention layers are replaced by Qwen Sparse Attention (QSA), which scores context at micro-block granularity with a compressed lightweight indexer. The residual stream is widened to four branches and read through an elementwise gate, a design we call the Gated Residual (GR). Capacity is added outside the backbone by a single n-gram embedding layer whose tables are prefetched from host memory. We evaluate every candidate change along three axes: loss together with downstream benchmarks; the cost of the change in training, prefill and decode; and its effect on the optimal hyperparameters and training stability. Loss and downstream accuracy do not always move together: enlarging the n-gram vocabulary lowers loss monotonically while downstream accuracy saturates. The architecture and the Muon optimizer together shift the optimal learning rate and batch size upwards, render batch-size warmup unnecessary, and substantially improve stability under stress tests. Loss, benchmarks, efficiency and stability form one design problem. Solved jointly, they yield a recipe that is simultaneously more efficient, more capable and more stable.

cs.CL↗

arXiv2026

Learning the Pareto Frontier of Predictive Models under Distribution Shift

Modern machine learning pipelines increasingly rely on reusing pretrained and foundation models across downstream tasks. These pretrained models can differ not only in performance but also in how they can be used: some only provide black-box predictions, while others may permit white-box access to internal representations that can be probed or fine-tuned. When deployed to the target domain in the presence of distribution shift, no single strategy, including zero-shot application, fine-tuning, or directly training a target-specific model, is uniformly the best. In this work, we propose Frontier Learning, a framework that treats a library of candidate models spanning different training histories and access regimes as complementary sources of information rather than mutually exclusive alternatives. Frontier Learning constructs a unified target-domain feature by concatenating internal representations from white-box candidates as well as prediction outputs from black-box candidates, then fits a lightweight, regularized supervised learner on this concatenated representation using labeled target data. Because the resulting hypothesis class contains predictors obtained by zero-shot reuse, fine-tuning, and direct training as special cases, empirical risk minimization over the frontier learner is guaranteed to be no worse, on the training sample, than any individual baseline. We evaluate the framework in simulations spanning varying degrees of source-target compatibility and in two real-world distribution-shift settings: visual domain adaptation on DomainNet/VisDA and clinical mortality prediction across intensive care unit domains using MIMIC-IV-Notes. Across all settings, Frontier Learning matches or outperforms the strongest individual reuse strategy, with the largest gains arising precisely when no single baseline is reliable across the range of shift considered.

cs.LG↗

arXiv2026

Enhancing Implicit Neural Representations with Image Feature Embedding for Unsupervised Cardiac Cine MRI Reconstruction

Cardiac cine Magnetic Resonance Imaging (MRI) is a critical diagnostic tool that provides dynamic insights for radiologists. To accelerate acquisition, under-sampled k-space data is often used, requiring reconstruction methods that combine coil sensitivity encoding with prior information to recover missing data. Deep learning approaches have gained more attention for leveraging data-adaptive priors. While supervised learning approaches are a common choice, they depend on fully sampled reference data, which is not always available. Unsupervised methods eliminate the need for fully sampled reference data, which can be advantageous in cardiac cine MRI reconstruction. Among them, implicit neural representations (INRs) have shown great potential due to their simple architecture and good quality reconstructions. In this work, we propose an image-domain dual-branch INR framework, termed I-FP-INR, which extends the original INR design by introducing an additional feature-processing branch. This design aims to extract complementary feature embeddings to enhance the overall representation, thereby benefiting reconstruction. Extensive evaluations on both public datasets and in-house data show consistent improvements over baseline methods in reconstruction quality, with strong robustness across varied scenarios.

cs.CV↗

arXiv2026

An agentic framework for gravitational-wave counterpart association in the multi-messenger era

With the detection of gravitational waves (GWs), multi-messenger astronomy has opened a new window for advancing our understanding of astrophysics, dense matter, gravitation, and cosmology. The GW sources detected to date are from mergers of compact object binaries, which possess the potential to generate detectable electromagnetic (EM) counterparts. Searching for associations between GW signals and their EM counterparts is an essential step toward enabling subsequent multi-messenger studies. In the era of next-generation GW and EM detectors, the rapid increase in the number of events brings not only unprecedented scientific opportunities, but also substantial challenges to the existing data analysis paradigm. To help address these challenges, we develop GW-Eyes, an agentic framework powered by large language models (LLMs). For the first time, GW-Eyes integrates domain-specific tools and autonomously performs counterpart association tasks between GW and candidate EM events. It supports natural language interaction to assist human experts with auxiliary tasks such as catalog management, skymap visualization, and rapid verification. Our framework leverages the complex decision-making capabilities of LLMs and their traceable reasoning processes, offering a new perspective to the multi-messenger astronomy.

astro-ph.IM↗

arXiv2026

Cracking Gravitational Wave Multiple Ringdown Modes in Space

Ringdown signals from perturbed black holes (BHs) offer a clean window into BH spacetime, strong-field gravity, and fundamental physics. Presently the quasi-normal modes of stellar-mass BH ringdowns have been successfully extracted in the ground-based gravitational wave (GW) observations. Looking ahead, the future space-borne observatories will listen to the ringdowns from massive BH binary coalescences more loudly and resolve multiple modes to unprecedented precision, which calls for efficient approaches to mitigate the sharply increasing computational burden. We develop a practical ringdown analysis pipeline for space-borne detectors by implementing FIREFLY, a novel acceleration algorithm validated in ground-based detectors, and for the first time demonstrate its compatibility and effectiveness with the time-delay interferometry (TDI) observables. With high fidelity, we achieve a $\sim 200$-fold speedup for a simulated ringdown signal including six modes, providing a viable and scalable route for multi-mode ringdown analysis in the space context. This new approach has sound statistical interpretation and is extensible to other GW sources in band.

gr-qc↗

arXiv2026

Canzona: A Unified, Asynchronous, and Load-Balanced Framework for Distributed Matrix-based Optimizers

The scaling of Large Language Models (LLMs) drives interest in matrix-based optimizers (e.g., Shampoo, Muon, SOAP) for their convergence efficiency; yet their requirement for holistic updates conflicts with the tensor fragmentation in distributed frameworks like Megatron. Existing solutions are suboptimal: synchronous approaches suffer from computational redundancy, while layer-wise partitioning fails to reconcile this conflict without violating the geometric constraints of efficient communication primitives. To bridge this gap, we propose Canzona, a Unified, Asynchronous, and Load-Balanced framework that decouples logical optimizer assignment from physical parameter distribution. For Data Parallelism, we introduce an alpha-Balanced Static Partitioning strategy that respects atomicity while neutralizing the load imbalance. For Tensor Parallelism, we design an Asynchronous Compute pipeline utilizing Micro-Group Scheduling to batch fragmented updates and hide reconstruction overhead. Extensive evaluations on the Qwen3 model family (up to 32B parameters) on 256 GPUs demonstrate that our approach preserves the efficiency of established parallel architectures, achieving a 1.57x speedup in end-to-end iteration time and reducing optimizer step latency by 5.8x compared to the baseline.

cs.DC↗

arXiv2026

Probing RLVR training instability through the lens of objective-level hacking

Prolonged reinforcement learning with verifiable rewards (RLVR) has been shown to drive continuous improvements in the reasoning capabilities of large language models, but the training is often prone to instabilities, especially in Mixture-of-Experts (MoE) architectures. Training instability severely undermines model capability improvement, yet its underlying causes and mechanisms remain poorly understood. In this work, we introduce a principled framework for understanding RLVR instability through the lens of objective-level hacking. Unlike reward hacking, which arises from exploitable verifiers, objective-level hacking emerges from token-level credit misalignment and is manifested as system-level spurious signals in the optimization objective. Grounded in our framework, together with extensive experiments on a 30B MoE model, we trace the origin and formalize the mechanism behind a key pathological training dynamic in MoE models: the abnormal growth of the training-inference discrepancy, a phenomenon widely associated with instability but previously lacking a mechanistic explanation. These findings provide a concrete and causal account of the training dynamics underlying instabilities in MoE models, offering guidance for the design of stable RLVR algorithms.

cs.AI↗

arXiv2026

Convergence Rate Analysis of the AdamW-style Shampoo: Unifying One-Sided and Two-Sided Preconditioning

This paper studies AdamW-style Shampoo, an effective variant of the classical Shampoo that won the external tuning track of the AlgoPerf neural network training competition. Our analysis unifies one-sided and two-sided preconditioning. When the exponents of the two preconditioners sum to $1/2$, we establish the convergence rate $\frac{1}{K}\sum_{k=1}^KE\left[||\nabla f(X_k)||_*\right]\leq O(\frac{\sqrt{m+n}C}{K^{1/4}})$, where $K$ represents the number of iterations, $(m,n)$ denotes the dimensions of the matrix-valued parameters, and $C$ matches the constant appearing in the optimal convergence rate of SGD. Theoretically, the nuclear norm and Frobenius norm satisfy $||\nabla f(X)||_F\leq ||\nabla f(X)||_*\leq \sqrt{\min\{m,n\}}||\nabla f(X)||_F$, which suggests that our convergence rate is analogous to the optimal $\frac{1}{K}\sum_{k=1}^KE\left[||\nabla f(X_k)||_F\right]\leq O(\frac{C}{K^{1/4}})$ convergence rate of SGD in the ideal case where $||\nabla f(X)||_*= \Theta(\sqrt{\min\{m,n\}})||\nabla f(X)||_F$ and $m$ and $n$ are of comparable magnitude. Then, we extend our analysis to settings where the preconditioning exponents do not sum to 1/2, and establish convergence with an explicit but more involved rate.

math.OC↗

arXiv2025

Inferring neutron-star Love-Q relations from gravitational waves in the hierarchical Bayesian framework

Despite the large uncertainties in the equation of state for neutron stars (NSs), a tight universal ``Love-Q'' relation exists between their dimensionless tidal deformability, $\Lambda$, and the dimensionless quadrupole moment, $Q$. However, this relation has not yet been directly measured through observations. Gravitational waves (GWs) emitted from binary NS (BNS) coalescences provide an avenue for such a measurement. In this study, we adopt a hierarchical Bayesian framework and combine multiple simulated GW events to measure the Love-Q relation. We simulate 1000 GW sources and select 20 events with the highest signal-to-noise ratios and NS spins for the analysis. By inspecting four parameterization models of the Love-Q relation, we observe strong correlations between the model parameters. We verify that a linear relation between $\ln\Lambda$ and $\ln Q$ is practically sufficient to describe the Love-Q relation with the precision expected from next-generation GW detectors. Furthermore, we utilize the inferred Love-Q relation to test modified gravity. Taking the dynamical Chern-Simons gravity as an example, our results suggest that the characteristic length can be constrained to $10\, \mathrm{km}$ or less with future GW observations.

gr-qc↗

arXiv2025

Improving Model Representation and Reducing KV Cache via Skip Connections with First Value Heads

Transformer models have driven breakthroughs across various language tasks by their strong capability to learn rich contextual representations. Scaling them to improve representation, however, often demands substantial memory and compute costs, such as the Key-Value (KV) cache used during auto-regressive decoding. Skip connections offer a promising way to improve representation without bloating resource usage, yet most prior works either improve expressivity while leaving KV costs unchanged, or reduce memory at the cost of weaker representation. In this work, we propose SkipV1Former, a Transformer variant that uses skip connections from the first layer's Value heads to strengthen model representation and reduce KV cache. Specifically, from the second block onward, each layer reuses half of its Value heads from the very first layer, while computing the other half as usual-cutting Value projections and V cache by nearly 50 \%. Theoretically, we show that routing uncompressed first-layer Values into deeper layers restores information lost to compression and accelerates the model's implicit mesa-optimization-a key pattern of Transformer in auto-regressive tasks. Empirically, across different model scales, SkipV1Former delivers consistent reductions of approximately 25 \% in KV cache while improving perplexity relative to standard Multi-Head Attention (MHA) Transformers and some advanced variants. Moreover, we propose a recipe for uptraining existing MHA Transformer checkpoints to SkipV1Former with only 10-15\% additional compute. Finally, SkipV1Former can seamlessly combine advanced methods like Group-Query Attention and Multi-Latent Attention to achieve further KV cache savings and performance improvement. When combined with YOCO, it cuts KV cache size by nearly 50 \% while still improving performance.

cs.LG↗

arXiv2025

Conda: Column-Normalized Adam for Training Large Language Models Faster

Large language models (LLMs) have demonstrated impressive generalization and emergent capabilities, yet their pre-training remains computationally expensive and sensitive to optimization dynamics. While Adam-based optimizers offer fast convergence by adapting learning rates coordinate-wise, recent studies reveal that their updates often suffer from poor spectral conditioning and low-rank structures, hindering efficiency. Muon addresses this issue via global spectral normalization but lacks the per-coordinate adaptivity of Adam. In this work, we propose Column-Normalized Adam (Conda), a novel optimizer that bridges the strengths of both approaches. Conda projects updates into an orthogonal subspace and applies column-wise second moment normalization based on the projected gradients, thereby achieving both improved spectral conditioning and maintaining coordinate-wise adaptivity. This design alleviates the spectral pathologies of Adam while preserving its fast convergence behavior. Extensive experiments on the LLaMA and GPT-2 series show that Conda consistently outperforms AdamW, Muon, and other baselines in pre-training. Remarkably, on the LLaMA series, Conda achieves 2-2.5 the convergence speed of AdamW, measured in both training steps and training time. Further ablations demonstrate its robustness under diverse training setups. These results collectively highlight Conda as an effective and broadly applicable optimizer for large-scale LLM training. The code is released on https://github.com/jie040109/Conda

cs.LG↗

arXiv2025

SADA: Safe and Adaptive Aggregation of Multiple Black-Box Predictions in Semi-Supervised Learning

Semi-supervised learning (SSL) arises in practice when labeled data are scarce or expensive to obtain, while large quantities of unlabeled data are readily available. With the growing adoption of machine learning techniques, it has become increasingly feasible to generate multiple predicted labels using a variety of models and algorithms, including deep learning, large language models, and generative AI. In this paper, we propose a novel approach that safely and adaptively aggregates multiple black-box predictions of uncertain quality for both inference and prediction tasks. Our method provides two key guarantees: (i) it never performs worse than using the labeled data alone, regardless of the quality of the predictions; and (ii) if any one of the predictions (without knowing which one) perfectly fits the ground truth, the algorithm adaptively exploits this to achieve either a faster convergence rate or the semiparametric efficiency bound. We demonstrate the effectiveness of the proposed algorithm through small-scale simulations and two real-data analyses with distinct scientific goals. A user-friendly R package, sada, is provided to facilitate practical implementation.

stat.ML↗

arXiv2025

Lightweight posterior construction for gravitational-wave catalogs with the Kolmogorov-Arnold network

Neural density estimation has seen widespread applications in the gravitational-wave (GW) data analysis, which enables real-time parameter estimation for compact binary coalescences and enhances rapid inference for subsequent analysis such as population inference. In this work, we explore the application of using the Kolmogorov-Arnold network (KAN) to construct efficient and interpretable neural density estimators for lightweight posterior construction of GW catalogs. By replacing conventional activation functions with learnable splines, KAN achieves superior interpretability, higher accuracy, and greater parameter efficiency on related scientific tasks. Leveraging this feature, we propose a KAN-based neural density estimator, which ingests megabyte-scale GW posterior samples and compresses them into model weights of tens of kilobytes. Subsequently, analytic expressions requiring only several kilobytes can be further distilled from these neural network weights with minimal accuracy trade-off. In practice, GW posterior samples with fidelity can be regenerated rapidly using the model weights or analytic expressions for subsequent analysis. Our lightweight posterior construction strategy is expected to facilitate user-level data storage and transmission, paving a path for efficient analysis of numerous GW events in the next-generation GW detectors.

gr-qc↗

arXiv2025

P/D-Device: Disaggregated Large Language Model between Cloud and Devices

Serving disaggregated large language models has been widely adopted in industrial practice for enhanced performance. However, too many tokens generated in decoding phase, i.e., occupying the resources for a long time, essentially hamper the cloud from achieving a higher throughput. Meanwhile, due to limited on-device resources, the time to first token (TTFT), i.e., the latency of prefill phase, increases dramatically with the growth on prompt length. In order to concur with such a bottleneck on resources, i.e., long occupation in cloud and limited on-device computing capacity, we propose to separate large language model between cloud and devices. That is, the cloud helps a portion of the content for each device, only in its prefill phase. Specifically, after receiving the first token from the cloud, decoupling with its own prefill, the device responds to the user immediately for a lower TTFT. Then, the following tokens from cloud are presented via a speed controller for smoothed TPOT (the time per output token), until the device catches up with the progress. On-device prefill is then amortized using received tokens while the resource usage in cloud is controlled. Moreover, during cloud prefill, the prompt can be refined, using those intermediate data already generated, to further speed up on-device inference. We implement such a scheme P/D-Device, and confirm its superiority over other alternatives. We further propose an algorithm to decide the best settings. Real-trace experiments show that TTFT decreases at least 60%, maximum TPOT is about tens of milliseconds, and cloud throughput increases by up to 15x.

cs.DC↗

arXiv2025

Stepsize anything: A unified learning rate schedule for budgeted-iteration training

The expanding computational costs and limited resources underscore the critical need for budgeted-iteration training, which aims to achieve optimal learning within predetermined iteration budgets. While learning rate schedules fundamentally govern the performance of different networks and tasks, particularly in budgeted-iteration scenarios, their design remains largely heuristic, lacking theoretical foundations. In addition, the optimal learning rate schedule requires extensive trial-and-error selection, making the training process inefficient. In this work, we propose the Unified Budget-Aware (UBA) schedule, a theoretically grounded learning rate schedule that consistently outperforms commonly-used schedules among diverse architectures and tasks under different constrained training budgets. First, we bridge the gap by constructing a novel training budget-aware optimization framework, which explicitly accounts for the robustness to landscape curvature variations. From this framework, we derive the UBA schedule, controlled by a single hyper-parameter \varphi that provides a trade-off between flexibility and simplicity, eliminating the need for per-network numerical optimization. Moreover, we establish a theoretical connection between \varphi and the condition number, adding interpretation and justification to our approach. Besides, we prove the convergence for different values of \varphi. We offer practical guidelines for its selection via theoretical analysis and empirical results. Extensive experimental results show that UBA consistently surpasses the commonly-used schedules across diverse vision and language tasks, spanning network architectures (e.g., ResNet, OLMo) and scales, under different training-iteration budgets.

cs.LG↗

arXiv2025

From Macro to Micro: Probing Dataset Diversity in Language Model Fine-Tuning

Dataset diversity plays a pivotal role for the successful training of many machine learning models, particularly in the supervised fine-tuning (SFT) stage of large language model (LLM) development. Despite increasing recognition of its importance, systematic analyses of dataset diversity still remain underexplored. To address this gap, this work presents a systematic taxonomy of existing diversity-control strategies, which primarily focus on the instruction component, operating at either macroscopic (entire instruction semantics) or mesoscopic levels (instruction units), and furthermore introduces a novel analysis of microscopic diversity within the response component, specifically analyzing the statistical distribution of tokens in SFT training samples. In the experimental evaluation, we construct fixed-size datasets (e.g., 10,000 samples each) from a corpus of 117,000 open-source SFT samples, incorporating six distinct diversity-control strategies spanning macro-, meso-, and microscopic levels applied to both instructions and responses. We then fine-tune LLMs on these datasets to assess the six diversity-control strategies. Results reveal that while macroscopic and mesoscopic strategies lead to higher performance with increasing diversity, the microscopic strategy in responses exhibits both a stronger correlation between model performance and the degree of diversity and superior performance with maximum diversity across all strategies. These findings offer actionable insights for constructing high-performance SFT datasets.

cs.CL↗

arXiv2025

On the $O(\frac{\sqrt{d}}{K^{1/4}})$ Convergence Rate of AdamW Measured by $\ell_1$ Norm

As the default optimizer for training large language models, AdamW has achieved remarkable success in deep learning. However, its convergence behavior is not theoretically well-understood. This paper establishes the convergence rate $\frac{1}{K}\sum_{k=1}^KE\left[||\nabla f(x^k)||_1\right]\leq O(\frac{\sqrt{d}C}{K^{1/4}})$ for AdamW measured by $\ell_1$ norm, where $K$ represents the iteration number, $d$ denotes the model dimension, and $C$ matches the constant in the optimal convergence rate of SGD. Theoretically, we have $||\nabla f(x)||_2\ll ||\nabla f(x)||_1\leq \sqrt{d}||\nabla f(x)||_2$ for any high-dimensional vector $x$ and $E\left[||\nabla f(x)||_1\right]\geq\sqrt{\frac{2d}{\pi}}E\left[||\nabla f(x)||_2\right]$ when each element of $\nabla f(x)$ is generated from Gaussian distribution $\mathcal N(0,1)$. Empirically, our experimental results on real-world deep learning tasks reveal $||\nabla f(x)||_1=\varTheta(\sqrt{d})||\nabla f(x)||_2$. Both support that our convergence rate can be considered to be analogous to the optimal $\frac{1}{K}\sum_{k=1}^KE\left[||\nabla f(x^k)||_2\right]\leq O(\frac{C}{K^{1/4}})$ convergence rate of SGD in the ideal case. We also extend our result to NAdamW, an AdamW variant that employs a double-momentum mechanism, and demonstrate that it maintains the same convergence rate.

cs.LG↗

arXiv2025

A practical Bayesian method for gravitational-wave ringdown analysis with multiple modes

Gravitational-wave (GW) ringdown signals from black holes (BHs) encode crucial information about the gravitational dynamics in the strong-field regime, which offers unique insights into BH properties. In the future, the improving sensitivity of GW detectors is to enable the extraction of multiple quasi-normal modes (QNMs) from ringdown signals. However, incorporating multiple modes drastically enlarges the parameter space, posing computational challenges to data analysis. Inspired by the $F$-statistic method in the continuous GW searches, we develope an algorithm, dubbed as FIREFLY, for accelerating the ringdown signal analysis. FIREFLY analytically marginalizes the amplitude and phase parameters of QNMs to reduce the computational cost and speed up the full-parameter inference from hours to minutes, while achieving consistent posterior and evidence. The acceleration becomes more significant when more QNMs are considered. Rigorously based on the principle of Bayesian inference and importance sampling, our method is statistically interpretable, flexible in prior choice, and compatible with various advanced sampling techniques, providing a new perspective for accelerating future GW data analysis.

gr-qc↗