arXiv subjects

Mingrui Liu

Publications and source records attributed to Mingrui Liu.

At least 19 recordsLinked to original sources

arXiv2026

How Edge of Stability Hinders SCAFFOLD in Federated Optimization

In federated learning, it is well known that heterogeneous data can (in theory) slow down optimization, and much effort has been directed at designing optimization algorithms that are unaffected by data heterogeneity, such as the SCAFFOLD algorithm. Yet, despite strong theoretical guarantees, SCAFFOLD does not usually outperform the much simpler FedAvg in practice. In this work, we propose that this gap is due to the presence of Edge of Stability (EoS) and progressive sharpening in federated optimization, supported by extensive empirical probing. First, we find that EoS-like dynamics occur with both FedAvg and SCAFFOLD under a variety of architectures and hyperparameters. We observe that the equilibrium value of the sharpness is inversely proportional to the learning rate (as in GD), and interestingly, the degree of data heterogeneity (but not the number of local steps) also affects the equilibrium value. Most importantly, we observe that SCAFFOLD's ability to estimate the gradient of the global objective is severely degraded at the EoS, as measured by the correlation between sharpness and SCAFFOLD's error in estimating the global gradient along the optimization trajectory. This suggests a mechanism for SCAFFOLD's lackluster performance in deep learning: with high sharpness at the EoS, SCAFFOLD cannot reliably estimate the global gradient.

cs.LG↗

arXiv2026

ROEVO: Robust Organized Edge Feature-based Visual Odometry Using RGB-D Cameras

This work presents a visual odometry (VO) system that leverages image edge features. Edges are spatially expressive cues commonly present across diverse environments, offering rich textural and structural information. However, existing edge-based VO methods often fail to fully exploit this potential. To this end, we introduce a novel feature representation termed \textit{organized edges}, which transforms disjoint edge pixels into sequentialized clusters, enabling more effective retention and utilization of the underlying textural and structural information. Another nice property of this formulation is that organized edges can perform edge-level association across multiple frames, enabling the establishment of a co-visibility graph. To achieve precise and efficient pose estimation, we propose a range of particularly designed tracking and joint optimization methods based on the characteristics of organized edges. For tracking, we formulate edge-wise rather than pixel-wise residuals to achieve robust and accurate inter-frame registration. For joint optimization, we introduce a novel shape-preserving edge-fitting method and an organized edge-based Bundle Adjustment (BA) approach, which decomposes the traditional BA problem into fitting and registration to preserve the structural integrity. Based on these novel techniques, we develop a complete VO system that exclusively employs organized edge features, achieving efficient tracking and precise local mapping. Extensive experiments demonstrate its accuracy and robustness in indoor environments, outperforming or achieving comparable performance to state-of-the-art methods. The source code is publicly available at https://github.com/liumingrui814/ROEVO

cs.RO↗

arXiv2026

Varstrometry for Off-nucleus and Dual sub-Kpc AGN (VODKA): Radio Classification of High-Redshift Dual AGN Candidates with the Very Large Array

Dual active galactic nuclei (dual AGNs) are pairs of simultaneously accreting supermassive black holes in merging galaxies. We investigate dual AGNs to understand whether merger-induced accretion is a significant growth mechanism for supermassive black holes. Searching for such systems is favorable at close separations and high redshift (Cosmic Noon, $z \sim 2$) due to the expected combination of high galaxy merger rate and peak AGN activity which characterize this era of the Universe. The sample of nine dual AGN candidates is selected based on resolved optical dual detections with Gaia, whose angular separations are less than 1$' '$ and redshifts range between 1.5 and 2.8. Each pair is spatially coincident with a Sloan Digital Sky Survey quasar primary target. We aim to classify the secondary targets and other components in the radio regime using 2-band Very Large Array imaging (C and Ku-bands) to test for dual AGN presence. We identify two dual AGNs and three quadruply imaged gravitational lens AGNs, two out of which show evidence of radio flux anomaly. The two new dual AGNs add to the limited census of confirmed kpc-scale pairs at Cosmic Noon, while the radio flux anomalies in J0911+0550 and J1118+0745 provide independent evidence for substructure in their lensing potentials, consistent with microlensing on compact AGN emission regions. Besides one confirmed AGN-star pair, three candidates remain unclassified due to lack of radio detection for one or two components.

astro-ph.GA↗

arXiv2026

Short-Acquisition Contrast-Free Super-Resolution Microvascular Imaging in Rabbit Kidney

Ultrasound localization microscopy (ULM) enables micrometer-scale microvascular imaging by localizing and tracking intravascular microbubbles, but its dependence on exogenous contrast agents and long acquisition times limits clinical translation. This study presents a high-frame-rate contrast-free super-resolution ultrasound microvascular imaging method based on high-frequency ultrafast ultrasound and nonlinear beamforming of backscatter signals from native blood flow. Using only 125 milliseconds of in vivo ultrafast data per image, the proposed method achieved an imaging frame rate of 8 frames/s in a rabbit kidney model. The reconstructed microvascular images resolved vessels with a global spatial resolution of 22.2 um over a field of view of 23.04 x 15.18 mm2, where the wavelength of ultrasound was 67.5 um. This corresponds to a three-fold improvement over conventional power Doppler imaging under the same acquisition duration. Compared with conventional flow imaging, the proposed method provided improved microvascular contrast and finer vessel delineation without microbubble injection. These results demonstrate a practical pathway toward high frame rate, contrast-free super-resolution ultrasound imaging for microvascular assessment.

eess.SP↗

arXiv2026

Generative Conversational Recommender System

Conversational recommender systems aim to provide personalized recommendations via natural language interactions. However, existing approaches either decouple recommendation from dialog generation or rely on retrieval-based pipelines, limiting the integration between recommendation and response generation and leading to suboptimal modeling of user intent. In this paper, we propose a fully generative conversational recommender system that unifies recommendation and dialog generation within a single autoregressive framework. Our approach represents items as discrete semantic IDs and integrates them directly into the generation process, enabling joint prediction of items and responses via next-token modeling. We further introduce a structured generation paradigm that factorizes conversational recommendation into a sequence of interdependent decisions, where the model first predicts the response intent and the recommendation target, and then generates the response conditioned on them. This design enables end-to-end optimization, enforces a more coherent dependency structure, and supports faithful item generation via constrained decoding. Extensive experiments demonstrate that our method consistently improves recommendation performance, achieving gains of up to 29% on Recall@1 over strong baselines, while maintaining competitive dialog quality.

cs.IR↗

arXiv2026

CURE-OOD: Benchmarking Out-of-Distribution Detection for Survival Prediction

``How long can I live and remain free of cancer?'' is often the first question a patient asks after receiving a cancer diagnosis and treatment. Accurate survival prediction helps alleviate psychological distress and supports risk stratification and personalized treatment planning. Recent survival prediction frameworks have shown strong performance using computed tomography (CT) images. However, variations in imaging acquisition introduce out-of-distribution (OOD) samples caused by covariate shifts that undermine model reliability. Despite this challenge, to our knowledge, no existing benchmark systematically studies OOD detection in cancer survival prediction. To address this gap, we introduce the Cancer sURvival bEnchmark for OOD Detection (CURE-OOD), the first benchmark for systematically evaluating OOD detection in survival prediction under controlled acquisition-induced distribution shifts. CURE-OOD defines scanner-parameter-based training, in-distribution (ID), and OOD test splits across four survival prediction tasks. Our experiments show that covariate shifts notably reduce survival prediction performance. It also shows that mainstream classification-oriented OOD detectors can fail in survival prediction. Finally, we include HazardDev as a simple survival-aware reference baseline for OOD detection. CURE-OOD enables systematic analysis of how distribution shifts affect both downstream survival performance and OOD detectability.

cs.CV↗

arXiv2026

Varstrometry for Off-nucleus and Dual Subkiloparsec AGN (VODKA): Three Quadruply Lensed Quasars at Cosmic Noon in HST and JWST

We present results from imaging observations of three quadruply lensed quasars by Hubble Space Telescope (HST) and James Webb Space Telescope (JWST) at redshifts $z = 2.550$, 2.975, and 1.500. We model our targets assuming a singular isothermal ellipsoid mass profile and an elliptical S\~A\c{opyright}rsic profile for the lensing galaxies, and reconstruct the geometric configuration of each system with measured Einstein radii of 0.44$'$, 0.58$'$, and 0.49$'$. While no spectroscopic measurements are available for the lenses, we constrain the redshift of each lens to $0.5 < z < 1.2$, $1.0 < z < 1.5$, and $0.4 < z < 0.9$. For all three lenses, the best-fit light model yield a typical de Vaucouleurs $n_{\rm S\acute{e}rsic} \sim 4$ profile and an effective radius $R_e$ around $\sim 1.5 - 3.5$ kpc. We accordingly classify the three lenses as early-type galaxies at an intermediate to high redshift, a common type for strong lensing galaxies. Compared to other known quadruple lenses, the lensing galaxies in this work are at the lower end of the distribution of Einstein radii and upper end of the distribution of the lens redshifts. They represent an interesting quadrant of subarcsecond-separation lenses in the population of single-galaxy strong lensing which have been largely unexplored yet and will be great targets of interest in upcoming high-resolution lensing surveys.

astro-ph.GA↗

arXiv2026

Tight Bounds for Logistic Regression with Large Stepsize Gradient Descent in Low Dimension

We consider the optimization problem of minimizing the logistic loss with gradient descent to train a linear model for binary classification with separable data. With a budget of $T$ iterations, it was recently shown that an accelerated $1/T^2$ rate is possible by choosing a large stepsize $\eta = \Theta(\gamma^2 T)$ (where $\gamma$ is the dataset's margin) despite the resulting non-monotonicity of the loss. In this paper, we provide a tighter analysis of gradient descent for this problem when the data is two-dimensional: we show that GD with a sufficiently large learning rate $\eta$ finds a point with loss smaller than $\mathcal{O}(1/(\eta \gamma^2 T))$, as long as $T \geq \Omega(n/\gamma + 1/\gamma^2)$, where $n$ is the dataset size. Our improved rate comes from a tighter bound on the time $\tau$ that it takes for GD to transition from unstable (non-monotonic loss) to stable (monotonic loss), via a fine-grained analysis of the oscillatory dynamics of GD in the subspace orthogonal to the max-margin classifier. We also provide a lower bound of $\tau$ matching our upper bound up to logarithmic factors, showing that our analysis is tight.

cs.LG↗

arXiv2026

Bilevel Optimization with Lower-Level Uniform Convexity: Theory and Algorithm

Bilevel optimization is a hierarchical framework where an upper-level optimization problem is constrained by a lower-level problem, commonly used in machine learning applications such as hyperparameter optimization. Existing bilevel optimization methods typically assume strong convexity or Polyak-{\L}ojasiewicz (PL) conditions for the lower-level function to establish non-asymptotic convergence to a solution with small hypergradient. However, these assumptions may not hold in practice, and recent work~\citep{chen2024finding} has shown that bilevel optimization is inherently intractable for general convex lower-level functions with the goal of finding small hypergradients. In this paper, we identify a tractable class of bilevel optimization problems that interpolates between lower-level strong convexity and general convexity via \emph{lower-level uniform convexity}. For uniformly convex lower-level functions with exponent $p\geq 2$, we establish a novel implicit differentiation theorem characterizing the hyperobjective's smoothness property. Building on this, we design a new stochastic algorithm, termed UniBiO, with provable convergence guarantees, based on an oracle that provides stochastic gradient and Hessian-vector product information for the bilevel problems. Our algorithm achieves $\widetilde{O}(\epsilon^{-5p+6})$ oracle complexity bound for finding $\epsilon$-stationary points. Notably, our complexity bounds match the optimal rates in terms of the $\epsilon$ dependency for strongly convex lower-level functions ($p=2$), up to logarithmic factors. Our theoretical findings are validated through experiments on synthetic tasks and data hyper-cleaning, demonstrating the effectiveness of our proposed algorithm.

math.OC↗

arXiv2026

RedVisor: Reasoning-Aware Prompt Injection Defense via Zero-Copy KV Cache Reuse

Large Language Models (LLMs) are increasingly vulnerable to Prompt Injection (PI) attacks, where adversarial instructions hidden within retrieved contexts hijack the model's execution flow. Current defenses typically face a critical trade-off: prevention-based fine-tuning often degrades general utility via the "alignment tax", while detection-based filtering incurs prohibitive latency and memory costs. To bridge this gap, we propose RedVisor, a unified framework that synthesizes the explainability of detection systems with the seamless integration of prevention strategies. To the best of our knowledge, RedVisor is the first approach to leverage fine-grained reasoning paths to simultaneously detect attacks and guide the model's safe response. We implement this via a lightweight, removable adapter positioned atop the frozen backbone. This adapter serves a dual function: it first generates an explainable analysis that precisely localizes the injection and articulates the threat, which then explicitly conditions the model to reject the malicious command. Uniquely, the adapter is active only during this reasoning phase and is effectively muted during the subsequent response generation. This architecture yields two distinct advantages: (1) it mathematically preserves the backbone's original utility on benign inputs; and (2) it enables a novel KV Cache Reuse strategy, eliminating the redundant prefill computation inherent to decoupled pipelines. We further pioneer the integration of this defense into the vLLM serving engine with custom kernels. Experiments demonstrate that RedVisor outperforms state-of-the-art defenses in detection accuracy and throughput while incurring negligible utility loss.

cs.CR↗

arXiv2026

Facet-Aware Multi-Head Mixture-of-Experts Model with Text-Enhanced Pre-training for Sequential Recommendation

Sequential recommendation (SR) systems excel at capturing users' dynamic preferences by leveraging their interaction histories. Most existing SR systems assign a single embedding vector to each item to represent its features, adopting various models to combine these embeddings into a sequence representation that captures user intent. However, we argue that this representation alone is insufficient to capture an item's multi-faceted nature (e.g., movie genres, starring actors). Furthermore, users often exhibit complex and varied preferences within these facets (e.g., liking both action and musical films within the genre facet), which are challenging to fully represent with static identifiers. To address these issues, we propose a novel architecture titled Facet-Aware Multi-Head Mixture-of-Experts Model for Sequential Recommendation (FAME). We leverage sub-embeddings from each head in the final multi-head attention layer to predict the next item separately, effectively capturing distinct item facets. A gating mechanism then integrates these predictions by dynamically determining their importance. Additionally, we introduce a Mixture-of-Experts (MoE) network within each attention head to disentangle varied user preferences within each facet, utilizing a learnable router network to aggregate expert outputs based on context. Complementing this architecture, we design a Text-Enhanced Facet-Aware Pre-training module to overcome the limitations of randomly initialized embeddings. By utilizing a pre-trained text encoder and employing an alternating supervised contrastive learning objective, we explicitly disentangle facet-specific features from textual metadata (e.g., descriptions) before sequential training begins. This ensures that the item embeddings are semantically robust and aligned with the downstream multi-facet framework.

cs.IR↗

arXiv2025

The Trojan Example: Jailbreaking LLMs through Template Filling and Unsafety Reasoning

As Large Language Models (LLMs) become integral to computing infrastructure, safety alignment serves as the primary security control preventing the generation of harmful payloads. However, this defense remains brittle. Existing jailbreak attacks typically bifurcate into white-box methods, which are inapplicable to commercial APIs due to lack of gradient access, and black-box optimization techniques, which often yield unnatural (e.g., syntactically rigid) or non-transferable (e.g., lacking cross-model generalization) prompts. In this work, we introduce TrojFill, a black-box exploitation framework that bypasses safety filters by targeting a fundamental logic flaw in current alignment paradigms: the decoupling of unsafety reasoning from content generation. TrojFill structurally reframes malicious instructions as a template-filling task required for safety analysis. By embedding obfuscated payloads (e.g., via placeholder substitution) into a "Trojan" structure, the attack induces the model to generate prohibited content as a "demonstrative example" ostensibly required for a subsequent sentence-by-sentence safety critique. This approach effectively masks the malicious intent from standard intent classifiers. We evaluate TrojFill against representative commercial systems, including GPT-4o, Gemini-2.5, DeepSeek-3.1, and Qwen-Max. Our results demonstrate that TrojFill achieves near-universal bypass rates: reaching 100% Attack Success Rate (ASR) on Gemini-flash-2.5 and DeepSeek-3.1, and 97% on GPT-4o, significantly outperforming existing black-box baselines. Furthermore, unlike optimization-based adversarial prompts, TrojFill generates highly interpretable and transferable attack vectors, exposing a systematic vulnerability inaligned LLMs.

cs.CR↗

arXiv2025

Noise-Adaptive Layerwise Learning Rates: Accelerating Geometry-Aware Optimization for Deep Neural Network Training

Geometry-aware optimization algorithms, such as Muon, have achieved remarkable success in training deep neural networks (DNNs). These methods leverage the underlying geometry of DNNs by selecting appropriate norms for different layers and updating parameters via norm-constrained linear minimization oracles (LMOs). However, even within a group of layers associated with the same norm, the local curvature can be heterogeneous across layers and vary dynamically over the course of training. For example, recent work shows that sharpness varies substantially across transformer layers and throughout training, yet standard geometry-aware optimizers impose fixed learning rates to layers within the same group, which may be inefficient for DNN training. In this paper, we introduce a noise-adaptive layerwise learning rate scheme on top of geometry-aware optimization algorithms and substantially accelerate DNN training compared to methods that use fixed learning rates within each group. Our method estimates gradient variance in the dual norm induced by the chosen LMO on the fly, and uses it to assign time-varying noise-adaptive layerwise learning rates within each group. We provide a theoretical analysis showing that our algorithm achieves a sharp convergence rate. Empirical results on transformer architectures such as LLaMA and GPT demonstrate that our approach achieves faster convergence than state-of-the-art optimizers.

cs.LG↗

arXiv2025

An Exploration of Non-Euclidean Gradient Descent: Muon and its Many Variants

To define a steepest descent method over a neural network, we need to choose a norm for each layer, a way to aggregate these norms across layers, and whether to use normalization. We systematically explore different alternatives for aggregating norms across layers, both formalizing existing combinations of Adam and the recently proposed Muon as a type of non-Euclidean gradient descent, and deriving new variants of the Muon optimizer. Through a comprehensive experimental evaluation of the optimizers within our framework, we find that Muon is sensitive to the choice of learning rate, whereas a new variant we call MuonMax is significantly more robust. We then show how to combine any non-Euclidean gradient method with model based momentum (known as Momo). The new Momo variants of Muon are significantly more robust to hyperparameter tuning, and often achieve a better validation score. Thus for new tasks, where the optimal hyperparameters are not known, we advocate for using Momo in combination with MuonMax to save on costly hyperparameter tuning.

cs.LG↗

arXiv2025

BLISS: A Lightweight Bilevel Influence Scoring Method for Data Selection in Language Model Pretraining

Effective data selection is essential for pretraining large language models (LLMs), enhancing efficiency and improving generalization to downstream tasks. However, existing approaches often require leveraging external pretrained models, making it difficult to disentangle the effects of data selection from those of the external pretrained models. In addition, they often overlook the long-term impact of selected data if the model is trained to convergence, primarily due to the prohibitive cost of full-scale LLM pretraining. In this paper, we introduce BLISS (\textbf{B}ileve\textbf{L} \textbf{I}nfluence \textbf{S}coring method for data \textbf{S}election): a lightweight data selection method that operates entirely \emph{from scratch}, without relying on any external pretrained oracle models, while explicitly accounting for the long-term impact of selected data. BLISS leverages a small proxy model as a surrogate for the LLM and employs a score model to estimate the long-term influence of training samples if the proxy model is trained to convergence. We formulate data selection as a bilevel optimization problem, where the upper-level objective optimizes the score model to assign importance weights to training samples, ensuring that minimizing the lower-level objective (i.e., training the proxy model over the weighted training loss until convergence) leads to best validation performance. Once optimized, the trained score model predicts influence scores for the dataset, enabling efficient selection of high-quality samples for LLM pretraining. We validate BLISS by pretraining 410M/1B/2.8B Pythia and LLaMA-0.5B models on selected subsets of the C4 dataset. Notably, under the 1B model setting, BLISS achieves $1.7\times$ speedup in reaching the same performance as the state-of-the-art method, demonstrating superior performance across multiple downstream tasks.

cs.LG↗

arXiv2025

Adaptive Algorithms with Sharp Convergence Rates for Stochastic Hierarchical Optimization

Hierarchical optimization refers to problems with interdependent decision variables and objectives, such as minimax and bilevel formulations. While various algorithms have been proposed, existing methods and analyses lack adaptivity in stochastic optimization settings: they cannot achieve optimal convergence rates across a wide spectrum of gradient noise levels without prior knowledge of the noise magnitude. In this paper, we propose novel adaptive algorithms for two important classes of stochastic hierarchical optimization problems: nonconvex-strongly-concave minimax optimization and nonconvex-strongly-convex bilevel optimization. Our algorithms achieve sharp convergence rates of $\widetilde{O}(1/\sqrt{T} + \sqrt{\bar{\sigma}}/T^{1/4})$ in $T$ iterations for the gradient norm, where $\bar{\sigma}$ is an upper bound on the stochastic gradient noise. Notably, these rates are obtained without prior knowledge of the noise level, thereby enabling automatic adaptivity in both low and high-noise regimes. To our knowledge, this work provides the first adaptive and sharp convergence guarantees for stochastic hierarchical optimization. Our algorithm design combines the momentum normalization technique with novel adaptive parameter choices. Extensive experiments on synthetic and deep learning tasks demonstrate the effectiveness of our proposed algorithms.

cs.LG↗

arXiv2025

Why and How Auxiliary Tasks Improve JEPA Representations

Joint-Embedding Predictive Architecture (JEPA) is increasingly used for visual representation learning and as a component in model-based RL, but its behavior remains poorly understood. We provide a theoretical characterization of a simple, practical JEPA variant that has an auxiliary regression head trained jointly with latent dynamics. We prove a No Unhealthy Representation Collapse theorem: in deterministic MDPs, if training drives both the latent-transition consistency loss and the auxiliary regression loss to zero, then any pair of non-equivalent observations, i.e., those that do not have the same transition dynamics or auxiliary value, must map to distinct latent representations. Thus, the auxiliary task anchors which distinctions the representation must preserve. Controlled ablations in a counting environment corroborate the theory and show that training the JEPA model jointly with the auxiliary head generates a richer representation than training them separately. Our work indicates a path to improve JEPA encoders: training them with an auxiliary function that, together with the transition dynamics, encodes the right equivalence relations.

cs.LG↗

arXiv2025

Wukong Framework for Not Safe For Work Detection in Text-to-Image systems

Text-to-Image (T2I) generation is a popular AI-generated content (AIGC) technology enabling diverse and creative image synthesis. However, some outputs may contain Not Safe For Work (NSFW) content (e.g., violence), violating community guidelines. Detecting NSFW content efficiently and accurately, known as external safeguarding, is essential. Existing external safeguards fall into two types: text filters, which analyze user prompts but overlook T2I model-specific variations and are prone to adversarial attacks; and image filters, which analyze final generated images but are computationally costly and introduce latency. Diffusion models, the foundation of modern T2I systems like Stable Diffusion, generate images through iterative denoising using a U-Net architecture with ResNet and Transformer blocks. We observe that: (1) early denoising steps define the semantic layout of the image, and (2) cross-attention layers in U-Net are crucial for aligning text and image regions. Based on these insights, we propose Wukong, a transformer-based NSFW detection framework that leverages intermediate outputs from early denoising steps and reuses U-Net's pre-trained cross-attention parameters. Wukong operates within the diffusion process, enabling early detection without waiting for full image generation. We also introduce a new dataset containing prompts, seeds, and image-specific NSFW labels, and evaluate Wukong on this and two public benchmarks. Results show that Wukong significantly outperforms text-based safeguards and achieves comparable accuracy of image filters, while offering much greater efficiency.

cs.CV↗