As supervised fine-tuning (SFT) evolves from a lightweight post-training step into a compute-intensive phase rivaling mid-training in scale, data efficiency has become critical for aligning large language models (LLMs) under tight budgets. Existing data pruning methods suffer from a fragmented design: they operate either at the sample level or the token level in isolation, failing to jointly optimize both dimensions. This disconnect leads to significant inefficiencies--high-value samples may still contain redundant tokens, while token-level pruning often discards crucial instructional or corrective signals embedded in individual examples. To address this bottleneck, we introduce the Error-Uncertainty (EU) Plane, a diagnostic framework that jointly characterizes the heterogeneous utility of training data across samples and tokens. Guided by this insight, we propose Quadrant-based Tuning (Q-Tuning), a unified framework that strategically coordinates sample pruning and token pruning. Q-Tuning employs a two-stage strategy: first, it performs sample-level triage to retain examples rich in informative misconceptions or calibration signals; second, it applies an asymmetric token-pruning policy, using a context-aware scoring mechanism to trim less salient tokens exclusively from misconception samples while preserving calibration samples in their entirety. Our method sets a new state of the art across five diverse benchmarks. Remarkably, on SmolLM2-1.7B, Q-Tuning achieves a +38\% average improvement over the full-data SFT baseline using only 12.5\% of the original training data. As the first dynamic pruning approach to consistently outperform full-data training, Q-Tuning provides a practical and scalable blueprint for maximizing data utilization in budget-constrained LLM SFT.

While automatic music transcription is well-established in music information retrieval, most models are limited to transcribing pitch and timing information from audio, and thus omit crucial expressive and instrument-specific nuances. One example is playing technique on the violin, which affords its distinct palette of timbres for maximal emotional impact. Here, we propose VioPTT (Violin Playing Technique-aware Transcription), a lightweight cascade model that directly transcribes violin playing technique in addition to pitch onset and offset. Furthermore, we release MOSA-VPT, a novel, high-quality synthetic violin playing technique dataset to circumvent the need for manually labeled annotations. Leveraging this dataset, our model demonstrated strong generalization to real-world note-level violin technique recordings in addition to achieving state-of-the-art transcription performance. To our knowledge, VioPTT is the first to jointly combine violin transcription and playing technique prediction within a unified framework.

Diffusion large language models (dLLMs) enable any-order generation, but this flexibility enlarges the attack surface: harmful spans may appear at arbitrary positions, and template-based prefilling attacks such as DIJA bypass response-level refusals. We introduce A2D (Any-Order, Any-Step Defense), a token-level alignment method that aligns dLLMs to emit an [EOS] refusal signal whenever harmful content arises. By aligning safety directly at the token-level under randomized masking, A2D achieves robustness to both any-decoding-order and any-step prefilling attacks under various conditions. It also enables real-time monitoring: dLLMs may begin a response but automatically terminate if unsafe continuation emerges. On safety benchmarks, A2D consistently prevents the generation of harmful outputs, slashing DIJA success rates from over 80% to near-zero (1.3% on LLaDA-8B-Instruct, 0.0% on Dream-v0-Instruct-7B), and thresholded [EOS] probabilities allow early rejection, yielding up to 19.3x faster safe termination.

Standard LLM cascades improve efficiency by deferring difficult queries from weak to strong models. However, these systems are typically static: when faced with repeated or semantically similar queries, they redundantly consult the expensive model, failing to adapt during inference. To address this, we propose Inter-Cascade, an online, interactive framework that transforms the strong model from a temporary helper into a long-term teacher. In our approach, when the strong model resolves a deferred query, it generates a generalized, reusable problem-solving strategy. These strategies are stored in a dynamic repository and retrieved via similarity matching to augment the weak model's context for future queries. This enables the weak model to learn on the job without expensive parameter fine-tuning. We theoretically show that this mechanism improves the weak model's confidence calibration. Empirically, Inter-Cascade outperforms standard cascades on multiple benchmarks, improving weak model and overall system accuracy by up to 33.06 percent and 6.35 percent, while reducing strong model calls by up to 48.05 percent and saving fee by up to 49.63 percent. Inter-Cascade demonstrates effective in-context knowledge transfer between LLMs and provides a general, scalable framework applicable to both open-source and API-based LLMs.

We study the capacity of the self-attention key-query channel: for a fixed budget, how many distinct token-token relations can a single layer reliably encode? We introduce Relational Graph Recognition, where the key-query channel encodes a directed graph and, given a context (a subset of the vertices), must recover the neighbors of each vertex in the context. We measure resources by the total key dimension $D_K = h\,d_k$. In a tractable multi-head model, we prove matching information-theoretic lower bounds and upper bounds via explicit constructions showing that recovering a graph with $m'$ relations in $d_{\text{model}}$-dimensional embeddings requires $D_K$ to grow essentially as $m'/d_{\text{model}}$ up to logarithmic factors, and we obtain corresponding guarantees for scaled-softmax attention. This analysis yields a new, capacity-based rationale for multi-head attention: even in permutation graphs, where all queries attend to a single target, splitting a fixed $D_K$ budget into multiple heads increases capacity by reducing interference from embedding superposition. Controlled experiments mirror the theory, revealing sharp phase transitions at the predicted capacity, and the multi-head advantage persists when adding softmax normalization, value routing, and a full Transformer block trained with frozen GPT-2 embeddings.

Large Vision-Language Models (LVLMs) have achieved remarkable success across a wide range of multimodal tasks, yet their robustness to spatial variations remains insufficiently understood. In this work, we conduct a systematic study of the spatial bias of LVLMs, examining how models respond when identical key visual information is placed at different locations within an image. Through controlled probing experiments, we observe that current LVLMs often produce inconsistent outputs under such spatial shifts, revealing a clear spatial bias in their semantic understanding. Further analysis indicates that this bias does not stem from the vision encoder, but rather from a mismatch in attention mechanisms between the vision encoder and the large language model, which disrupts the global information flow. Motivated by this insight, we propose Adaptive Global Context Injection (AGCI), a lightweight mechanism that dynamically injects shared global visual context into each image token. AGCI works without architectural modifications, mitigating spatial bias by enhancing the semantic accessibility of image tokens while preserving the model's intrinsic capabilities. Extensive experiments demonstrate that AGCI not only enhances the spatial robustness of LVLMs, but also achieves strong performance on various downstream tasks and hallucination benchmarks.

Retrieval-Augmented Generation (RAG) aims to mitigate hallucinations in large language models (LLMs) by grounding responses in retrieved documents. Yet, RAG-based LLMs still hallucinate even when provided with correct and sufficient context. A growing line of work suggests that this stems from an imbalance between how models use external context and their internal knowledge, and several approaches have attempted to quantify these signals for hallucination detection. However, existing methods require extensive hyperparameter tuning, limiting their generalizability. We propose LUMINA, a novel framework that detects hallucinations in RAG systems through context--knowledge signals: external context utilization is quantified via distributional distance, while internal knowledge utilization is measured by tracking how predicted tokens evolve across transformer layers. We further introduce a framework for statistically validating these measurements. Experiments on common RAG hallucination benchmarks and four open-source LLMs show that LUMINA achieves consistently high AUROC and AUPRC scores, outperforming prior utilization-based methods by up to +13% AUROC on HalluRAG. Moreover, LUMINA remains robust under relaxed assumptions about retrieval quality and model matching, offering both effectiveness and practicality. LUMINA: https://github.com/deeplearning-wisc/LUMINA

Magnetic Resonance Imaging is a critical imaging modality in clinical diagnosis and research, yet its complexity and heterogeneity hinder scalable, generalizable machine learning. Although foundation models have revolutionized language and vision tasks, their application to MRI remains constrained by data scarcity and narrow anatomical focus. We present Decipher-MR, a 3D MRI-specific vision-language foundation model trained on 200,000 MRI series from over 22,000 studies spanning diverse anatomical regions, sequences, and pathologies. Decipher-MR integrates self-supervised vision learning with report-guided text supervision to build robust representations for broad applications. To enable efficient use, Decipher-MR supports a modular design that enables tuning of lightweight, task-specific decoders attached to a frozen pretrained encoder. Following this setting, we evaluate Decipher-MR across disease classification, demographic prediction, anatomical localization, and cross-modal retrieval, demonstrating consistent improvements over existing foundation models and task-specific approaches. These results position Decipher-MR as a versatile foundation for MRI-based AI in clinical and research settings.

The elephant trunk exemplifies a natural gripper where structure, actuation, and sensing are seamlessly integrated. Inspired by the distal morphology of the African elephant trunk, we present MELEGROS, a Monolithic ELEphant-inspired GRipper with Optical Sensors, emphasizing sensing as an intrinsic, co-fabricated capability. Unlike multi-material or tendon-based approaches, MELEGROS directly integrates six optical waveguide sensors and five pneumatic chambers into a pneumatically actuated lattice structure (12.5 mm cell size) using a single soft resin and one continuous 3D print. This eliminates mechanical mismatches between sensors, actuators, and body, reducing model uncertainty and enabling simulation-guided sensor design and placement. Only four iterations were required to achieve the final prototype, which features a continuous structure capable of elongation, compression, and bending while decoupling tactile and proprioceptive signals. MELEGROS (132 g) lifts more than twice its weight, performs bioinspired actions such as pinching, scooping, and reaching, and delicately grasps fragile items like grapes. The integrated optical sensors provide distinct responses to touch, bending, and chamber deformation, enabling multifunctional perception. MELEGROS demonstrates a new paradigm for soft robotics where fully embedded sensing and continuous structures inherently support versatile, bioinspired manipulation.

Few-Shot Class-Incremental Learning (FSCIL) must contend with the dual challenge of learning new classes from scarce samples while preserving old class knowledge. Existing methods use the frozen feature extractor and class-averaged prototypes to mitigate against catastrophic forgetting and overfitting. However, new-class prototypes suffer significant estimation bias due to extreme data scarcity, whereas base-class prototypes benefit from sufficient data. In this work, we theoretically demonstrate that aligning the new-class priors with old-class statistics via Bayesian analysis reduces variance and improves prototype accuracy. Furthermore, we propose large-scale contrastive learning to enforce cross-category feature tightness. To further enrich feature diversity and inject prior information for new-class prototypes, we integrate momentum self-supervision and virtual categories into the Momentum Tightness and Contrast framework (MoTiC), constructing a feature space with rich representations and enhanced interclass cohesion. Experiments on three FSCIL benchmarks produce state-of-the-art performances, particularly on the fine-grained task CUB-200, validating our method's ability to reduce estimation bias and improve incremental learning robustness.

Accurate registration between LiDAR (Light Detection and Ranging) point clouds and semantic 3D city models is a fundamental topic in urban digital twinning and a prerequisite for downstream tasks, such as digital construction, change detection, and model refinement. However, achieving accurate LiDAR-to-Model registration at the individual building level remains challenging, particularly due to the generalization uncertainty in semantic 3D city models at the Level of Detail 2 (LoD2). This paper addresses this gap by proposing L2M-Reg, a plane-based fine registration method that explicitly accounts for model uncertainty. L2M-Reg consists of three key steps: establishing reliable plane correspondence, building a pseudo-plane-constrained Gauss-Helmert model, and adaptively estimating vertical translation. Overall, extensive experiments on five real-world datasets demonstrate that L2M-Reg is both more accurate and computationally efficient than current leading ICP-based and plane-based methods. Therefore, L2M-Reg provides a novel building-level solution regarding LiDAR-to-Model registration when model uncertainty is present. The datasets and code for L2M-Reg can be found: https://github.com/Ziyang-Geodesy/L2M-Reg.

Aligning AI systems with human values remains a fundamental challenge, but does our inability to create perfectly aligned models preclude obtaining the benefits of alignment? We study a strategic setting where a human user interacts with multiple differently misaligned AI agents, none of which are individually well-aligned. Our key insight is that when the users utility lies approximately within the convex hull of the agents utilities, a condition that becomes easier to satisfy as model diversity increases, strategic competition can yield outcomes comparable to interacting with a perfectly aligned model. We model this as a multi-leader Stackelberg game, extending Bayesian persuasion to multi-round conversations between differently informed parties, and prove three results: (1) when perfect alignment would allow the user to learn her Bayes-optimal action, she can also do so in all equilibria under the convex hull condition (2) under weaker assumptions requiring only approximate utility learning, a non-strategic user employing quantal response achieves near-optimal utility in all equilibria and (3) when the user selects the best single AI after an evaluation period, equilibrium guarantees remain near-optimal without further distributional assumptions. We complement the theory with two sets of experiments.

Graph Transformers (GTs) show considerable potential in graph representation learning. The architecture of GTs typically integrates Graph Neural Networks (GNNs) with global attention mechanisms either in parallel or as a precursor to attention mechanisms, yielding a local-and-global or local-to-global attention scheme. However, as the global attention mechanism primarily captures long-range dependencies between nodes, these integration schemes may suffer from information loss, where the local neighborhood information learned by GNN could be diluted by the attention mechanism. Therefore, we propose G2LFormer, featuring a novel global-to-local attention scheme where the shallow network layers use attention mechanisms to capture global information, while the deeper layers employ GNN modules to learn local structural information, thereby preventing nodes from ignoring their immediate neighbors. An effective cross-layer information fusion strategy is introduced to allow local layers to retain beneficial information from global layers and alleviate information loss, with acceptable trade-offs in scalability. To validate the feasibility of the global-to-local attention scheme, we compare G2LFormer with state-of-the-art linear GTs and GNNs on node-level and graph-level tasks. The results indicate that G2LFormer exhibits excellent performance while keeping linear complexity.

Accurate visual localization is crucial for autonomous driving, yet existing methods face a fundamental dilemma: While high-definition (HD) maps provide high-precision localization references, their costly construction and maintenance hinder scalability, which drives research toward standard-definition (SD) maps like OpenStreetMap. Current SD-map-based approaches primarily focus on Bird's-Eye View (BEV) matching between images and maps, overlooking a ubiquitous signal-noisy GPS. Although GPS is readily available, it suffers from multipath errors in urban environments. We propose DiffVL, the first framework to reformulate visual localization as a GPS denoising task using diffusion models. Our key insight is that noisy GPS trajectory, when conditioned on visual BEV features and SD maps, implicitly encode the true pose distribution, which can be recovered through iterative diffusion refinement. DiffVL, unlike prior BEV-matching methods (e.g., OrienterNet) or transformer-based registration approaches, learns to reverse GPS noise perturbations by jointly modeling GPS, SD map, and visual signals, achieving sub-meter accuracy without relying on HD maps. Experiments on multiple datasets demonstrate that our method achieves state-of-the-art accuracy compared to BEV-matching baselines. Crucially, our work proves that diffusion models can enable scalable localization by treating noisy GPS as a generative prior-making a paradigm shift from traditional matching-based methods.

The reliance on implicit point matching via attention has become a core bottleneck in drag-based editing, resulting in a fundamental compromise on weakened inversion strength and costly test-time optimization (TTO). This compromise severely limits the generative capabilities of diffusion models, suppressing high-fidelity inpainting and text-guided creation. In this paper, we introduce LazyDrag, the first drag-based image editing method for Multi-Modal Diffusion Transformers, which directly eliminates the reliance on implicit point matching. In concrete terms, our method generates an explicit correspondence map from user drag inputs as a reliable reference to boost the attention control. This reliable reference opens the potential for a stable full-strength inversion process, which is the first in the drag-based editing task. It obviates the necessity for TTO and unlocks the generative capability of models. Therefore, LazyDrag naturally unifies precise geometric control with text guidance, enabling complex edits that were previously out of reach: opening the mouth of a dog and inpainting its interior, generating new objects like a ``tennis ball'', or for ambiguous drags, making context-aware changes like moving a hand into a pocket. Additionally, LazyDrag supports multi-round workflows with simultaneous move and scale operations. Evaluated on the DragBench, our method outperforms baselines in drag accuracy and perceptual quality, as validated by VIEScore and human evaluation. LazyDrag not only establishes new state-of-the-art performance, but also paves a new way to editing paradigms.

Prompt engineering is crucial for fully leveraging large language models (LLMs), yet most existing optimization methods follow a single trajectory, resulting in limited adaptability, gradient conflicts, and high computational overhead. We propose MAPGD (Multi-Agent Prompt Gradient Descent), a novel framework that reconceptualizes prompt optimization as a collaborative process among specialized agents. Each agent focuses on a distinct refinement dimension, such as instruction clarity, example selection, format structure, or stylistic adaptation, and their contributions are coordinated through semantic gradient embedding, conflict detection, and fusion. To further enhance robustness and stability, MAPGD introduces two new mechanisms: Hypersphere Constrained Gradient Clustering (HCGC), which enforces angular margin constraints for compact and well-separated clusters, and Channel Adaptive Agent Weighting (CAAW), which dynamically reweights agent contributions based on validation performance. Experiments on classification and reasoning benchmarks show that MAPGD consistently surpasses single-agent and random baselines in both accuracy and efficiency. Ablation studies confirm the effectiveness of gradient fusion, agent specialization, and conflict resolution. Together, these components establish MAPGD as a unified, gradient-based, and interpretable framework for robust prompt optimization with theoretical convergence guarantees.

This paper presents a correspondence-free, function-based sim-to-real learning method for controlling deformable freeform surfaces. Unlike traditional sim-to-real transfer methods that strongly rely on marker points with full correspondences, our approach simultaneously learns a deformation function space and a confidence map -- both parameterized by a neural network -- to map simulated shapes to their real-world counterparts. As a result, the sim-to-real learning can be conducted by input from either a 3D scanner as point clouds (without correspondences) or a motion capture system as marker points (tolerating missed markers). The resultant sim-to-real transfer can be seamlessly integrated into a neural network-based computational pipeline for inverse kinematics and shape control. We demonstrate the versatility and adaptability of our method on both vision devices and across four pneumatically actuated soft robots: a deformable membrane, a robotic mannequin, and two soft manipulators.

Model reduction, which aims to learn a simpler model of the original mixed integer linear programming (MILP), can solve large-scale MILP problems much faster. Most existing model reduction methods are based on variable reduction, which predicts a solution value for a subset of variables. From a dual perspective, constraint reduction that transforms a subset of inequality constraints into equalities can also reduce the complexity of MILP, but has been largely ignored. Therefore, this paper proposes a novel constraint-based model reduction approach for the MILP. Constraint-based MILP reduction has two challenges: 1) which inequality constraints are critical such that reducing them can accelerate MILP solving while preserving feasibility, and 2) how to predict these critical constraints efficiently. To identify critical constraints, we first label these tight-constraints at the optimal solution as potential critical constraints and design a heuristic rule to select a subset of critical tight-constraints. To learn the critical tight-constraints, we propose a multi-modal representation technique that leverages information from both instance-level and abstract-level MILP formulations. The experimental results show that, compared to the state-of-the-art methods, our method improves the quality of the solution by over 50\% and reduces the computation time by 17.47\%.

Graph Neural Networks (GNNs) have demonstrated remarkable results in various real-world applications, including drug discovery, object detection, social media analysis, recommender systems, and text classification. In contrast to their vast potential, training them on large-scale graphs presents significant computational challenges due to the resources required for their storage and processing. Graph Condensation has emerged as a promising solution to reduce these demands by learning a synthetic compact graph that preserves the essential information of the original one while maintaining the GNN's predictive performance. Despite their efficacy, current graph condensation approaches frequently rely on a computationally intensive bi-level optimization. Moreover, they fail to maintain a mapping between synthetic and original nodes, limiting the interpretability of the model's decisions. In this sense, a wide range of decomposition techniques have been applied to learn linear or multi-linear functions from graph data, offering a more transparent and less resource-intensive alternative. However, their applicability to graph condensation remains unexplored. This paper addresses this gap and proposes a novel method called Multi-view Graph Condensation via Tensor Decomposition (GCTD) to investigate to what extent such techniques can synthesize an informative smaller graph and achieve comparable downstream task performance. Extensive experiments on six real-world datasets demonstrate that GCTD effectively reduces graph size while preserving GNN performance, achieving up to a 4.0\ improvement in accuracy on three out of six datasets and competitive performance on large graphs compared to existing approaches. Our code is available at https://anonymous.4open.science/r/gctd-345A.

High-precision controllable remote sensing image generation is both meaningful and challenging. Existing diffusion models often produce low-fidelity images due to their inability to adequately capture morphological details, which may affect the robustness and reliability of object detection models. To enhance the accuracy and fidelity of generated objects in remote sensing, this paper proposes Object Fidelity Diffusion (OF-Diff), which effectively improves the fidelity of generated objects. Specifically, we are the first to extract the prior shapes of objects based on the layout for diffusion models in remote sensing. Then, we introduce a dual-branch diffusion model with diffusion consistency loss, which can generate high-fidelity remote sensing images without providing real images during the sampling phase. Furthermore, we introduce DDPO to fine-tune the diffusion process, making the generated remote sensing images more diverse and semantically consistent. Comprehensive experiments demonstrate that OF-Diff outperforms state-of-the-art methods in the remote sensing across key quality metrics. Notably, the performance of several polymorphic and small object classes shows significant improvement. For instance, the mAP increases by 8.3%, 7.7%, and 4.0% for airplanes, ships, and vehicles, respectively.

The translation of natural language to formal constraint models requires expertise in the problem domain and modeling frameworks. To explore the effectiveness of agentic workflows, we propose CP-Agent, a Python coding agent that uses the ReAct framework with a persistent IPython kernel. We provide the relevant domain knowledge as a project prompt of under 50 lines. The algorithm works by iteratively executing code, observing the solver's feedback, and refining constraint models based on execution results. We evaluate CP-Agent on 101 constraint programming problems from CP-Bench. We made minor changes to the benchmark to address systematic ambiguities in the problem specifications and errors in the ground-truth models. On the clarified benchmark, CP-Agent achieves perfect accuracy on all 101 problems. Our experiments show that minimal guidance outperforms detailed procedural scaffolding. Our experiments also show that explicit task management tools can have both positive and negative effects on focused modeling tasks.

This paper presents DDTracking, a novel deep generative framework for diffusion MRI tractography that formulates streamline propagation as a conditional denoising diffusion process. In DDTracking, we introduce a dual-pathway encoding network that jointly models local spatial encoding (capturing fine-scale structural details at each streamline point) and global temporal dependencies (ensuring long-range consistency across the entire streamline). Furthermore, we design a conditional diffusion model module, which leverages the learned local and global embeddings to predict streamline propagation orientations for tractography in an end-to-end trainable manner. We conduct a comprehensive evaluation across diverse, independently acquired dMRI datasets, including both synthetic and clinical data. Experiments on two well-established benchmarks with ground truth (ISMRM Challenge and TractoInferno) demonstrate that DDTracking largely outperforms current state-of-the-art tractography methods. Furthermore, our results highlight DDTracking's strong generalizability across heterogeneous datasets, spanning varying health conditions, age groups, imaging protocols, and scanner types. Collectively, DDTracking offers anatomically plausible and robust tractography, presenting a scalable, adaptable, and end-to-end learnable solution for broad dMRI applications. Code is available at: https://github.com/yishengpoxiao/DDtracking.git

Few-shot fine-grained visual classification (FGVC) aims to leverage limited data to enable models to discriminate subtly distinct categories. Recent works mostly finetuned the pre-trained visual language models to achieve performance gain, yet suffering from overfitting and weak generalization. To deal with this, we introduce UniFGVC, a universal training-free framework that reformulates few-shot FGVC as multimodal retrieval. First, we propose the Category-Discriminative Visual Captioner (CDV-Captioner) to exploit the open-world knowledge of multimodal large language models (MLLMs) to generate a structured text description that captures the fine-grained attribute features distinguishing closely related classes. CDV-Captioner uses chain-of-thought prompting and visually similar reference images to reduce hallucination and enhance discrimination of generated captions. Using it we can convert each image into an image-description pair, enabling more comprehensive feature representation, and construct the multimodal category templates using few-shot samples for the subsequent retrieval pipeline. Then, off-the-shelf vision and text encoders embed query and template pairs, and FGVC is accomplished by retrieving the nearest template in the joint space. UniFGVC ensures broad compatibility with diverse MLLMs and encoders, offering reliable generalization and adaptability across few-shot FGVC scenarios. Extensive experiments on 12 FGVC benchmarks demonstrate its consistent superiority over prior few-shot CLIP-based methods and even several fully-supervised MLLMs-based approaches.

Lifelong Person Re-identification (LReID) aims to continuously match individuals across camera views from sequential data streams. Existing LReID methods often ignore domain-specific style awareness and unified knowledge consolidation, which are crucial for mitigating forgetting when adapting to new information. We propose DSKC, a novel rehearsal-free and distillation-free framework for LReID. DSKC designs a domain-style encoder (DSE) to dynamically model domain-specific styles, and a unified knowledge consolidation (UKC) mechanism to adaptively integrate instance-level representations with domain-specific style into a cross-domain unified representation. By leveraging unified representation as a bridge, DSKC explicitly models inter-domain associations at both instance and domain levels to enhance anti-forgetting and generalization. Experimental results demonstrate that our DSKC outperforms state-of-the-art methods in two training orders and enhances the model's strong performance. Our code is available at https://github.com/LiuShiBen/DKUA.

Recent advances in conditional generative modeling have introduced Continuous conditional Generative Adversarial Network (CcGAN) and Continuous Conditional Diffusion Model (CCDM) for estimating high-dimensional data distributions conditioned on scalar, continuous regression labels (e.g., angles, ages, or temperatures). However, these approaches face fundamental limitations: CcGAN suffers from data imbalance due to fixed-size vicinity constraints, while CCDM requires computationally expensive iterative sampling. To address these issues, we propose CcGAN-AVAR, an enhanced CcGAN framework featuring (1) two novel components for handling data imbalance - an adaptive vicinity mechanism that dynamically adjusts vicinity size and a multi-task discriminator that enhances generator training through auxiliary regression and density ratio estimation - and (2) the GAN framework's native one-step generator, enable 30x-2000x faster inference than CCDM. Extensive experiments on four benchmark datasets (64x64 to 256x256 resolution) across eleven challenging settings demonstrate that CcGAN-AVAR achieves state-of-the-art generation quality while maintaining sampling efficiency.

Few-shot classification and segmentation (FS-CS) focuses on jointly performing multi-label classification and multi-class segmentation using few annotated examples. Although the current state of the art (SOTA) achieves high accuracy in both tasks, it struggles with small objects. To overcome this, we propose the Efficient Masked Attention Transformer (EMAT), which improves classification and segmentation accuracy, especially for small objects. EMAT introduces three modifications: a novel memory-efficient masked attention mechanism, a learnable downscaling strategy, and parameter-efficiency enhancements. EMAT outperforms all FS-CS methods on the PASCAL-5$^i$ and COCO-20$^i$ datasets, using at least four times fewer trainable parameters. Moreover, as the current FS-CS evaluation setting discards available annotations, despite their costly collection, we introduce two novel evaluation settings that consider these annotations to better reflect practical scenarios.

Large Language Models (LLMs) excel on many task-specific benchmarks, yet the mechanisms that drive this success remain poorly understood. We move from asking what these systems can do to asking how they process information. Our contribution is a task-agnostic method that builds a quantitative Cognitive Profile for any model. The profile is built around the Entropy Decay Curve -- a plot of a model's normalised predictive uncertainty as context length grows. Across several state-of-the-art LLMs and diverse texts, the curves expose distinctive, stable profiles that depend on both model scale and text complexity. We also propose the Information Gain Span (IGS) as a single index that summarises the desirability of a decay pattern. Together, these tools offer a principled way to analyse and compare the internal dynamics of modern AI systems.

We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which improves upon Muon with a novel QK-clip technique to address training instability while enjoying the advanced token efficiency of Muon. Based on MuonClip, K2 was pre-trained on 15.5 trillion tokens with zero loss spike. During post-training, K2 undergoes a multi-stage post-training process, highlighted by a large-scale agentic data synthesis pipeline and a joint reinforcement learning (RL) stage, where the model improves its capabilities through interactions with real and synthetic environments. Kimi K2 achieves state-of-the-art performance among open-source non-thinking models, with strengths in agentic capabilities. Notably, K2 obtains 66.1 on Tau2-Bench, 76.5 on ACEBench (En), 65.8 on SWE-Bench Verified, and 47.3 on SWE-Bench Multilingual -- surpassing most open and closed-sourced baselines in non-thinking settings. It also exhibits strong capabilities in coding, mathematics, and reasoning tasks, with a score of 53.7 on LiveCodeBench v6, 49.5 on AIME 2025, 75.1 on GPQA-Diamond, and 27.1 on OJBench, all without extended thinking. These results position Kimi K2 as one of the most capable open-source large language models to date, particularly in software engineering and agentic tasks. We release our base and post-trained model checkpoints to facilitate future research and applications of agentic intelligence.

The exponential growth in demand for GPU computing resources has created an urgent need for automated CUDA optimization strategies. While recent advances in LLMs show promise for code generation, current SOTA models achieve low success rates in improving CUDA speed. In this paper, we introduce CUDA-L1, an automated reinforcement learning framework for CUDA optimization that employs a novel contrastive RL algorithm. CUDA-L1 achieves significant performance improvements on the CUDA optimization task: trained on A100, it delivers an average speedup of x3.12 with a median speedup of x1.42 against default baselines over across all 250 CUDA kernels of KernelBench, with peak speedups reaching x120. In addition to the default baseline provided by KernelBench, CUDA-L1 demonstrates x2.77 over Torch Compile, x2.88 over Torch Compile with reduce overhead, x2.81 over CUDA Graph implementations, and x7.72 over cuDNN libraries. Furthermore, the model also demonstrates portability across different GPU architectures. Beyond these benchmark results, CUDA-L1 demonstrates several properties: it 1) discovers a variety of CUDA optimization techniques and learns to combine them strategically to achieve optimal performance; 2) uncovers fundamental principles of CUDA optimization, such as the multiplicative nature of optimizations; 3) identifies non-obvious performance bottlenecks and rejects seemingly beneficial optimizations that actually harm performance. The capabilities demonstrate that, RL can transform an initially poor-performing LLM into an effective CUDA optimizer through speedup-based reward signals alone, without human expertise or domain knowledge. This paradigm opens possibilities for automated optimization of CUDA operations, and holds promise to substantially promote GPU efficiency and alleviate the rising pressure on GPU computing resources. Project: deepreinforce-ai.github.io/cudal1_blog

Stochastic gradients for deep neural networks exhibit strong correlations along the optimization trajectory, and are often aligned with a small set of Hessian eigenvectors associated with outlier eigenvalues. Recent work shows that projecting gradients away from this Hessian outlier subspace has little impact on optimization, despite capturing a large fraction of gradient variability. Since computing the Hessian is intractable in practice, we introduce a principled first-order characterization of the nuisance subspace based on the covariance of stochastic gradients, and propose an efficient method to estimate it online. We show that removing this subspace also has little impact on optimization, and yields practical benefits for applications sensitive to gradient signal-to-noise ratio such as gradient compression.