arXiv:2601.21382v2 Announce Type: replace
Abstract: Nonlinear metasurfaces offer compact control over frequency conversion and wavefront shaping. However, existing approaches, often based on geometric phase, lack wavelength selectivity, resulting in static nonlinear responses. Here, we demonstrate a thin-film lithium niobate metasurface that enables spectrally selective shaping of second-harmonic generation through resonance-engineered phase control. The structure consists of two regions with distinct phase responses, realized via spectral tuning of Mie-type resonances. This design enables simultaneous frequency conversion and spatial mode shaping, transforming a Gaussian pump near 1100 nm into a first-order Hermite-Gaussian mode at 550 nm, while maintaining the pump profile. The demonstrated approach offers a pathway toward ultracompact and tunable components for nonlinear holography and related applications.
Science Journals
arXiv:2603.14798v2 Announce Type: replace-cross
Abstract: We propose a machine-learning algorithm for Bayesian inverse problems in the function-space regime. Based on one-step generative transport, the method learns an amortized neural operator whose pushforward of a Gaussian source approximates the posterior distribution conditioned on each new observation. We show that white-noise sources are incompatible with the function-space limit, and therefore adopt a prior-aligned GRF as the source. We justify this choice through the Lipschitz regularity of the resulting one-step conditional posterior transport and numerical experiments on linear inverse and PDE-based inverse problems. The method is not distilled from MCMC: it is trained only with prior samples and simulated partial noisy observations. Once trained, it generates a $64\times64$ posterior sample in $\sim 10^{-3}$s, avoiding repeated forward-model evaluations in MCMC and repeated network evaluations in multistep generative samplers while matching key posterior summaries.
arXiv:2601.21444v2 Announce Type: replace
Abstract: The efficiency of long-video inference remains a critical bottleneck, mainly due to the dense computation in the prefill stage of Large Multimodal Models (LMMs). Existing methods either compress visual embeddings or apply sparse attention on a single GPU, yielding limited acceleration or degraded performance and restricting LMMs from handling longer, more complex videos. To overcome these issues, we propose APB-V, a sequence-parallel framework with optimized attention that accelerates long-video inference across multiple GPUs. By distributing approximate attention, APB-V reduces computation and increases parallelism, enabling efficient processing of more visual embeddings without compression and thereby improving task performance. System-level optimizations, such as load balancing and fused forward passes, further unleash the potential of APB-V, delivering speedups of 12.72x, 1.70x, and 1.18x over FlashAttn, ZigZagRing, and APB, without notable performance loss. Code available at https://github.com/thunlp/APB
arXiv:2604.03399v2 Announce Type: replace-cross
Abstract: Impulse-to-peak response (I2P) analysis for state-space ordinary differential equation (ODE) systems is a well-studied classical problem. However, the techniques employed for I2P optimal control of ODEs have not been extended to partial differential equation (PDE) systems due to the lack of a universal transfer function and state-space representation. Recently, however, partial integral equation (PIE) representation was proposed as the desired state-space representation of a PDE, and Lyapunov stability theory was used to solve various control problems, such as stability and optimal ${H}_\infty$ control. In this work, we utilize this PIE framework, and associated Lyapunov techniques, to formulate the I2P response analysis problem as a solvable convex optimization and obtain provable bounds for the I2P-norm of linear PDEs. Moreover, by establishing strong duality between primal and dual formulations of the optimization problem, we develop a constructive method for I2P optimal state-feedback control of PDEs and demonstrate the effectiveness of the method on various examples.
arXiv:2606.02551v1 Announce Type: new
Abstract: Affordance understanding bridges visual perception and physical action, serving as an explainable interface for robot manipulation in open and unstructured real-world environments. Yet, building an affordance foundation model that not only understands where and how the interaction should happen, but also generalizes across diverse environments, objects, and tasks, remains a long-standing research challenge. Existing methods typically address only part of this challenge, either localizing task-relevant regions without specifying executable motion, or predicting motion but with limited scalability. In this paper, we present ourmodel, a step towards an affordance foundation model for functionality understanding. From a single RGB-D observation and a language task description, ourmodel predicts a task-conditional functional mask (where to interact) and a 3D post-contact motion curve (how to interact). To support open-world generalization, we build a large-scale standardized data pipeline that converts heterogeneous robot, human, simulation, and real-world scan data into a shared affordance schema with language, masks, and object-centric 3D motion labels. We evaluate ourmodel from three aspects: for affordance segmentation, ourmodel outperforms all baselines by a large margin across 8 test sets from 4 benchmarks, improving mean gIoU/cIoU by +23.9/+26.3; for contact-point prediction, it predicts substantially more accurate points, with a 12.7--61.3% hit-rate gain over the best baseline; and for 3D motion, it achieves the best performance on all three test sets. ourmodel can be deployed for real-world robot manipulation without finetuning for robot embodiment or using task-specific heuristics, demonstrating the ability to adapt to open-world affordance tasks. Project page: https://www.zhaoningwang.com/AFUN
arXiv:2606.02037v1 Announce Type: new
Abstract: In this paper, we propose that the idea of devices respecting their end-users may serve as a strong design goal for highly personal and intimate smart devices. We ask what respect is, how it shapes interaction, and how good-faith simulation of respect might inform user-friendly smart device design. Respect is a natural and integral part of natural human relationships that is seen to shape work and personal relations. In a basic sense, this is the core purpose of smart things: we expect them to be ready and willing to help us. In this vein, we distil the characteristics of more complex respectful behaviours into 4 main types relevant to smart devices, drawing from philosophical analyses of the conceptual dimensions of respect: directive respect, obstacle respect, recognition respect, and care respect. We discuss the implications of each of these kinds of respect for the future of smart personal devices.
arXiv:2507.19702v1 Announce Type: cross
Abstract: Identifying influential nodes in complex networks is a critical task with a wide range of applications across different domains. However, existing approaches often face trade-offs between accuracy and computational efficiency. To address these challenges, we propose 1D-CGS, a lightweight and effective hybrid model that integrates the speed of one-dimensional convolutional neural networks (1D-CNN) with the topological representation power of GraphSAGE for efficient node ranking. The model uses a lightweight input representation built on two straightforward and significant topological features: node degree and average neighbor degree. These features are processed through 1D convolutions to extract local patterns, followed by GraphSAGE layers to aggregate neighborhood information. We formulate the node ranking task as a regression problem and use the Susceptible-Infected-Recovered (SIR) model to generate ground truth influence scores. 1D-CGS is initially trained on synthetic networks generated by the Barabasi-Albert model and then applied to real world networks for identifying influential nodes. Experimental evaluations on twelve real world networks demonstrate that 1D-CGS significantly outperforms traditional centrality measures and recent deep learning models in ranking accuracy, while operating in very fast runtime. The proposed model achieves an average improvement of 4.73% in Kendall's Tau correlation and 7.67% in Jaccard Similarity over the best performing deep learning baselines. It also achieves an average Monotonicity Index (MI) score 0.99 and produces near perfect rank distributions, indicating highly unique and discriminative rankings. Furthermore, all experiments confirm that 1D-CGS operates in a highly reasonable time, running significantly faster than existing deep learning methods, making it suitable for large scale applications.
arXiv:2602.06060v2 Announce Type: replace
Abstract: The paper examines and critiques the expression of entropy as the logarithm of the number of quantum states of a physical system. Boltzmann method of expressing entropy as the logarithm of the number of states of a gas with a given total energy is analyzed. We demonstrate that entropy is the product of subquantum processes and show that entropy is expressed as the ratio of the logarithm of the maximum number of realizations, over the observation period, of a macroscopic system's states with a given total energy, to the number of occurrences of its quantum states over this time.
arXiv:2606.01738v1 Announce Type: new
Abstract: Multi-turn jailbreak attacks pose a growing threat to LLMs by exploiting conversational dynamics such as gradual escalation and cross-turn coordination. Existing defenses either rely on costly retraining -- often degrading model utility -- or apply single-turn analysis independently at each turn, failing to capture how risk accumulates along interaction trajectories. We observe that safety behavior in multi-turn interaction is trajectory-dependent: dialogue history continuously reshapes the model's conditioning context, making it insufficient to evaluate each turn in isolation. Motivated by this insight, we present THRD, the first training-free framework that explicitly models temporal risk accumulation for multi-turn jailbreak defense. THRD integrates four modules: a Turn-level Risk Assessor (TRA) for instantaneous risk estimation, a Historical Context Analyzer (HCA) for cross-turn intent escalation detection, a Response Evaluator (RE) for identifying facilitative outputs, and a Decision Module that combines these signals through a time-evolving scoring mechanism with attenuation-based modulation and trend-aware adjustment. Experiments against state-of-the-art multi-turn attacks -- including tree-search-based and multi-agent collaborative methods -- across two target models show that THRD reduces ASR to 0.2--4.0% while preserving model utility within 1.5% degradation on MMLU and GSM8K. Ablation studies confirm non-redundant module contributions and stable cross-architecture generalization. Analysis of first rejection triggers reveals that over 70% of multi-turn attacks require Turn~2 or later to detect, validating the necessity of explicit temporal aggregation.
arXiv:2606.01971v1 Announce Type: new
Abstract: The Compressed Baryonic Matter (CBM) experiment at the upcoming Facility for Antiproton and Ion Research (FAIR) is a high-rate fixed-target experiment designed to investigate nuclear matter at extreme baryon densities in relativistic nucleus-nucleus collisions. To enable high-statistics measurements of rare probes, CBM is designed to operate at event rates up to 10 MHz. This necessitates the development of fast and radiation-tolerant detectors, self-triggered front-end electronics, a free-streaming data acquisition architecture, and real-time event reconstruction capabilities. Prototype versions and pre-series productions of the CBM detector systems have been deployed in the mini-CBM demonstrator setup mCBM - an experimental precursor comprising sub-components of all major CBM systems, installed at the SIS18 facility of GSI/FAIR within the FAIR Phase-0 program. In 2024, Ni+Ni collisions at a kinetic beam energy of 1.93 AGeV and an average interaction rate of about 250 kHz were successfully recorded. This dataset enables a detailed evaluation of the operational performance of the detector systems as well as the complete CBM data chain, while the reconstruction of rare $\Lambda$ baryons serves as a natural benchmark. This paper presents the first results on $\Lambda$ signal reconstruction with the mCBM experiment, demonstrating the readiness of the detector technologies and the data chain for the upcoming full-scale CBM experiment.
arXiv:2606.01974v1 Announce Type: new
Abstract: Systems programming languages traditionally struggle with the tension between physical transparency and compile-time memory safety. C++ provides direct, zero-cost hardware access but lacks strict safety boundaries, whereas Rust guarantees safety at the cost of complex lifetime annotations and implicit dereferencing chains.
In this paper, we present Toka, a native systems programming language that establishes physical transparency in resource management via Explicit Resource Semantics. At the core of Toka's design is the Handle-Soul Duality (informally referred to as the Hat-Soul model), which cleanly dissociates pointer identities (Handles) from their underlying values (Souls) at the syntactic level. By enforcing that bare identifiers always represent values (Souls) and explicit sigils represent pointer handles, Toka eliminates the semantic ambiguity between rebind operations and value mutations. We detail Toka's resource morphology (supporting unique, shared, borrowed, and raw semantics), its lifetime checking mechanism, and its implementation of a prototype compiler. Our evaluation demonstrates that Toka achieves competitive runtime performance and minimal binary size while drastically reducing the cognitive overhead of lifetime annotations.
arXiv:2606.02038v1 Announce Type: new
Abstract: Reconstructing spatially continuous daily temperature fields from sparse observations is important for urban climate monitoring and heat-risk analysis, but practical deployments are limited by sensor budgets and spacing constraints. This study proposes an uncertainty-aware graph neural network (GNN) framework for reconstructing daily maximum temperature fields from sparse sensors while supporting distance-constrained sensor placement and probabilistic exceedance mapping. The model predicts both the temperature field and a spatially varying predictive uncertainty field using a graph-attention-based mean-residual architecture trained with a Gaussian negative log-likelihood. Sensor placement is addressed using a Proper Orthogonal Decomposition with QR factorization (POD-QR) strategy with a 4 km minimum inter-sensor distance constraint and is compared with random feasible placement and farthest-point sampling. The framework is evaluated over a Montreal-area polygon using Daymet v4.1 daily temperature data (1 km resolution) under a strict temporal hold-out protocol (training: 2020-2023; testing: 2024). Across sensor budgets (10-40 sensors), the proposed GNN consistently outperforms inverse distance weighting and ordinary kriging in RMSE and MAE on unobserved nodes. Sensor-placement effects are most pronounced at low budgets and diminish at higher budgets, with a practical saturation regime emerging around 30 sensors under the imposed spacing constraint. Probabilistic evaluation further shows improved uncertainty calibration with increasing sensor density and a better sharpness-calibration trade-off than kriging. These results support the proposed framework as an effective tool for uncertainty-aware temperature field reconstruction and decision-oriented heat-risk mapping.
arXiv:2606.02042v1 Announce Type: new
Abstract: Continual industrial anomaly detection with diffusion models suffers from historical normality prior drift and catastrophic forgetting. Existing continual diffusion methods preserve previous knowledge through replay or constrained optimization, but they lack an explicit mechanism for isolating and protecting category-specific normality priors during sequential adaptation. Although low-rank adaptation provides modular residual updates, standard LoRA neither freezes historical normality subspaces nor prevents new adapters from interfering with previous ones. To address this issue, we propose a normality-preserving continual anomaly detection framework based on two modules: History Frozen Orthogonal LoRA Bank (HF-OLB) and Hierarchical Novelty Adaptive Bank Growth module (HNABG). HF-OLB freezes both the pre-trained U-Net backbone and the learned LoRA banks, and constrains new task-specific normality residuals to the orthogonal complement of historical LoRA subspaces. HNABG further allocates layer-dependent residual capacity and expands the bank only when the residual normality novelty exceeds the expressive capacity of existing banks. Extensive experiments on MVTec and VisA demonstrate the effectiveness of the proposed method. On the challenging VisA 2x6 setting, our method achieves 83.6/91.8 image and pixel level A-AUROC with 3.8/3.9 FM, improving pixel level A-AUROC over the state of the art by 3.2 points while reducing pixel level FM by 1.3. These results show that our method effectively preserves historical normality priors in long horizon continual category sequences.
arXiv:2606.01663v1 Announce Type: new
Abstract: The coordination of heterogeneous autonomous agents in dynamic, adversarial environments requires simultaneous satisfaction of geometric constraints, logical consistency, temporal reasoning, and strategic optimization. Existing sheaf- and topos-theoretic frameworks provide powerful tools for geometric consensus, knowledge alignment, and causal planning, but lack explicit models for value, reward, and strategic choice. This report presents a unified categorical framework that integrates event calculus, SCEL-like ensemble formation, and game-theoretic reward structures into a single Grothendieck topos of time-space histories. We introduce the notion of a \emph{game sheaf} whose stalks contain utility functions and policy distributions, and restriction maps encode both parallel transport and best-response dynamics. We prove that Nash equilibria correspond to global sections of a derived best-response correspondence sheaf, while cohomological obstructions classify failures of strategic consistency. A detailed case study of an immunological ``bastion defense'' scenario -- heterogeneous agents forming attack/defense ensembles under resource constraints -- demonstrates the framework's expressiveness. This synthesis provides a rigorous foundation for verifiable, autonomic, and economically rational multi-agent systems.
arXiv:2606.01670v1 Announce Type: new
Abstract: Recently, Generative Recommenders (GRs) have emerged as a transformative recommendation paradigm by replacing traditional item IDs with semantic indices (SIDs). Owing to the exceptional generative capabilities of diffusion models, a few pioneering works explore developing GRs with diffusion architectures as the backbone. However, a fatal limitation of existing diffusion-based GRs is that the diffusion process applies uniformly to all items within the historical interactions. In contrast, the user preference is shaped by multifaceted time-evolving factors and thus exhibits a non-stationary distribution in the temporal aspect. To bridge this gap, this study proposes a novel GR framework, named TDPM, by designing the time-aware diffusion on SID tokens. Specifically, TDPM explicitly integrates the impact of time-evolving user preferences into the diffusion process. In detail, the user preference is disentangled into (i) the period preference, which remains consistent over a long time-span, and (ii) the point preference, which is triggered by recent focal events. Extensive experiments on three public real-world datasets demonstrate the significant superiority of TDPM over the state-of-the-art baselines. TDPM achieves average improvements of up to 29.21% and 25.45% in terms of HR@20 and NDCG@20, respectively. The ablation study further underscores the necessity of time-aware token diffusion in diffusion-based GRs.
arXiv:2601.22651v2 Announce Type: replace
Abstract: Training-data attribution for vision generative models aims to identify which training data influenced a given output. While most methods score individual examples, practitioners often need group-level answers (e.g., artistic styles or object classes). Group-wise attribution is counterfactual: how would a model's behavior on a generated sample change if a group were absent from training? A natural realization of this counterfactual is Leave-One-Group-Out (LOGO) retraining, which retrains the model with each group removed; however, it becomes computationally prohibitive as the number of groups grows. We propose GUDA (Group Unlearning-based Data Attribution) for diffusion models, which approximates each counterfactual model by applying machine unlearning to a shared full-data model instead of training from scratch. GUDA quantifies group influence using differences in a likelihood-based scoring rule (ELBO) between the full model and each unlearned counterfactual. Experiments on CIFAR-10 and artistic style attribution with Stable Diffusion show that GUDA identifies primary contributing groups more reliably than semantic similarity, gradient-based attribution, and instance-level unlearning approaches, while achieving ~100x speedup on CIFAR-10 over LOGO retraining.
arXiv:2604.03567v2 Announce Type: replace-cross
Abstract: For a graph $H$ and an integer $k\ge 1$, the \emph{Token Sliding reconfiguration graph} $\mathsf{TS}_k(H)$ and the \emph{Token Jumping reconfiguration graph} $\mathsf{TJ}_k(H)$ have as vertices the $k$-cliques of $H$, with two vertices adjacent when one clique is obtained from the other by replacing one vertex with an adjacent non-member, and respectively by an arbitrary non-member. For a target graph $G$, we study the feasibility sets $\mathcal{K}^{\mathsf{TS}}(G)$ and $\mathcal{K}^{\mathsf{TJ}}(G)$, consisting of all integers $k$ for which $G$ is isomorphic to $\mathsf{TS}_k(H)$ and $\mathsf{TJ}_k(H)$, respectively, for some graph $H$. We determine the exact feasibility sets for complete graphs, paths, cycles, complete bipartite graphs, book graphs, friendship graphs, and their complements, and give complete classifications for all Johnson graphs.
arXiv:2508.10312v2 Announce Type: replace
Abstract: Recommender systems in concert with Large Language Models (LLMs) present promising avenues for generating semantically-informed recommendations. However, LLM-based recommenders exhibit a tendency to overemphasize semantic correlations within users' interaction history. When taking pretrained collaborative ID embeddings as input, LLM-based recommenders progressively weaken the inherent collaborative signals as the embeddings propagate through LLM backbones layer by layer, as opposed to traditional Transformer-based sequential models in which collaborative signals are typically preserved or even enhanced for state-of-the-art performance. To address this limitation, we introduce FreLLM4Rec, an approach designed to balance semantic and collaborative information from a spectral perspective. Item embeddings that incorporate both semantic and collaborative information are first purified using a Global Graph Low-Pass Filter (G-LPF) to preliminarily remove irrelevant high-frequency noise. Temporal Frequency Modulation (TFM) then actively preserves collaborative signal layer by layer. Note that the collaborative preservation capability of TFM is theoretically guaranteed by establishing a connection between the optimal but hard-to-implement local graph fourier filters and the suboptimal yet computationally efficient frequency-domain filters. Extensive experiments on four benchmark datasets demonstrate that FreLLM4Rec successfully mitigates collaborative signal attenuation and achieves competitive performance, with improvements of up to 8.00\% in NDCG@10 over the best baseline. Our findings provide insights into how LLMs process collaborative information and offer a principled approach for improving LLM-based recommendation systems.
arXiv:2604.03989v3 Announce Type: replace-cross
Abstract: This paper develops a Finsler-based LMI for robust $\mathcal{H}_\infty$ observer design with integral quadratic constraints (IQCs) and block-structured uncertainty. By introducing a slack variable that relaxes the coupling between the Lyapunov matrix, the observer gain, and the IQC multiplier, the formulation addresses two limitations of the standard block-diagonal approach: the LMI requirement $\mathrm{He}(PA) \prec 0$ (which fails for marginally stable dynamics), and a multiplier--Lyapunov trade-off that causes infeasibility for wide uncertainty ranges. For marginally stable dynamics, artificial damping in the design model balances certified versus actual performance. The framework is demonstrated on quaternion attitude estimation with angular velocity uncertainty and mass-spring-damper state estimation with uncertain physical parameters.
arXiv:2604.04958v3 Announce Type: replace-cross
Abstract: Recent work suggests that large-scale, multi-animal modeling can significantly improve neural recording analysis. However, for functional calcium traces, existing approaches remain task-specific, limiting transfer across common neuroscience objectives. To address this challenge, we propose \textbf{CalM}, a self-supervised neural foundation model trained solely on neuronal calcium traces and adaptable to multiple downstream tasks, including forecasting and decoding. Our key contribution is a pretraining framework, composed of a high-performance tokenizer mapping single-neuron traces into a shared discrete vocabulary, and a dual-axis autoregressive transformer modeling dependencies along both the neural and the temporal axis. We evaluate CalM on a large-scale, multi-animal, multi-session dataset. On the neural population dynamics forecasting task, CalM achieves competitive performance against strong specialized baselines after pretraining. With a task-specific head, CalM further adapts to the behavior decoding task and achieves superior results compared with supervised decoding models. Moreover, linear analyses of CalM representations reveal interpretable functional structures beyond predictive accuracy. Taken together, we propose a novel and effective self-supervised pretraining paradigm for foundation models based on calcium traces, paving the way for scalable pretraining and broad applications in functional neural analysis. Code is released at https://github.com/TSuXinH/CalM.
arXiv:2606.01737v1 Announce Type: new
Abstract: Traffic accident liability analysis is a critical yet challenging task in intelligent transportation and legal assistance. Existing methods often suffer from low efficiency, subjective judgment, and inconsistent analysis results. Meanwhile, large language models are constrained by noisy video inputs and insufficient legal domain knowledge. To address these issues, this work presents TrafficRAG, a multimodal retrieval-augmented framework for automated traffic accident analysis and report generation. Specifically, the proposed framework first adopts a vision-language model to produce structured textual descriptions of accident scenarios, which serve as accurate retrieval queries. Based on these textual queries, a hybrid retrieval strategy integrating BM25 sparse retrieval and dense embedding retrieval is employed to fetch relevant traffic regulations and similar historical cases. Finally, the large language model incorporates retrieved legal knowledge and multimodal accident evidence for comprehensive reasoning, and generates standardized, legally grounded liability analysis reports. Extensive experiments show that TrafficRAG consistently outperforms baseline methods, achieving 77.32% Legal Norm Adaptation Accuracy, 81.71% Factual Faithfulness, and a Liability Ratio MAE of 5.48%. The results validate that integrating multimodal factual evidence with legal clauses via retrieval augmentation can effectively improve the reliability and accuracy of traffic accident liability determination.
arXiv:2606.01824v1 Announce Type: new
Abstract: We present \emph{DisFlow}, a novel framework for online scene flow estimation from distance field that enables \emph{6DoF dynamic object pose estimation}, \emph{motion tracking}, and \emph{surface reconstruction}. The scene is represented by Gaussian Process Implicit Surfaces (GPIS), with surface normals serving as derivative constraints, enabling accurate signed distance computations near the surface and gradient queries with uncertainty. With this representation as a foundation, we compute a scene flow from the distance field that describes how surface points are transported over time in consecutive frames. Through our flow, we can estimate an object's pose and motion by incrementally registering a new observed point cloud via an elegant closed-form optimisation. Unlike prior methods that operate in the camera or world frame, our approach performs probabilistic fusion directly in the \emph{object frame}, where the object remains geometrically consistent over time. The tight coupling of the DisFlow method in space and time yields dense geometry, surface normals, object pose trajectories, velocities, and uncertainty, all at real-time rates. We evaluate DisFlow on dynamic object sequences and demonstrate that it achieves accurate pose and motion tracking while simultaneously reconstructing high-quality object surfaces. Code publicly available at \href{https://github.com/LanWu076/disflow_ros2}{https://github.com/LanWu076/disflow\_ros2}
arXiv:2601.22813v2 Announce Type: replace
Abstract: The NVFP4 lower-precision format, supported in hardware by NVIDIA Blackwell GPUs, promises to allow, for the first time, end-to-end fully-quantized pre-training of massive models such as LLMs. Yet, existing quantized training methods still sacrifice some of the representation capacity of this format in favor of more accurate unbiased quantized gradient estimation by stochastic rounding (SR), losing noticeable accuracy relative to standard FP16 and FP8 training. In this paper, improve the state of the art for quantized training in NVFP4 via a novel unbiased quantization routine for micro-scaled formats, called MS-EDEN, that has more than 2x lower quantization error than SR. We integrate it into a novel fully-NVFP4 quantization scheme for linear layers, called Quartet II. We show analytically that Quartet II achieves consistently better gradient estimation across all major matrix multiplications, both on the forward and on the backward passes. In addition, our proposal synergizes well with recent training improvements aimed specifically at NVFP4. We further validate Quartet II on end-to-end LLM training with up to 1.9B parameters on 38B tokens. We provide kernels for execution on NVIDIA Blackwell GPUs with up to 4.2x speedup over BF16. Our code is available at https://github.com/IST-DASLab/Quartet-II .
arXiv:2601.22900v2 Announce Type: replace
Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning across domains, but outcome-only scalar rewards are often sparse and uninformative. This limitation is especially severe for failed samples, where scalar rewards indicate only that a solution is incorrect without explaining why the reasoning breaks down. In this paper, we leverage richer verbal feedback to guide RLVR on failed samples and convert feedback-induced progress into trainable learning signals. We propose MulFeRL (Multi-turn Feedback-guided Reinforcement Learning), a multi-turn, event-triggered RLVR framework that combines progress induction for feedback-guided regeneration of failed samples, progress credit assignment for learning from verifier-confirmed progress, and structured feedback injection for integrating feedback into the model's reasoning process. Trained on sampled OpenR1-Math, MulFeRL outperforms supervised, self-distillation-based, and RLVR baselines in-domain, while also showing strong out-of-domain generalization.
arXiv:2604.09808v2 Announce Type: replace-cross
Abstract: We present a complete formalization, in the Lean interactive theorem prover with the Mathlib library, of the Ramanujan--Nagell theorem: the only integer solutions to the Diophantine equation $x^2 + 7 = 2^n$ are $(n,x) \in \{(3,\pm1),(4,\pm3),(5,\pm5),(7,\pm11),(15,\pm181)\}$. The formalization includes all dependencies, notably the computation of the ring of integers of the quadratic field $\mathbb{Q}(\sqrt{-7})$, its class number, and unit group. We describe the proof strategy, the architecture of the formalization, and the challenges encountered in bridging the gap between textbook proofs and their machine-checked counterparts, with particular attention to the algebraic number theory infrastructure required.