arXiv subjects

Jie Lou

Publications and source records attributed to Jie Lou.

At least 19 recordsLinked to original sources

arXiv2026

Studying line defect at Deconfined Quantum Criticality via fuzzy sphere regularization

The interplay between bulk critical fluctuations and nontrivial topology can enrich defect physics and give rise to novel defect universality classes. Understanding the fate of such defects therefore constitutes an important open problem. In this work, we studied a particularly simple setting: a (0+1)-dimensional pinning-field defect coupled to a (2+1)-dimensional deconfined quantum critical bulk. Using the fuzzy-sphere regularization, we numerically investigated the defect operator spectrum and extracted several universal quantities characterizing the defect conformal fixed point, including the scaling dimensions of defect-changing(creating) operators and the defect $g$-function. These results establish the first numerical characterization of line-defect conformal data at deconfined quantum criticality and may stimulate further investigations of defect critical phenomena in topological quantum critical matter.

cond-mat.str-el↗

arXiv2026

Winding-Sector Transitions and Thermodynamic Incommensurability in Helical Valence Bond Phase under Tilted Boundary Conditions

We investigate the ground states of the $S = 1/2$ staircase $J$-$Q_3$ model in the maximally anisotropic limit by employing projector quantum Monte Carlo simulations. To overcome boundary-induced finite-size ambiguities inherent in the study of spatially modulated structures, we implement a $45^{\circ}$ tilted periodic boundary condition that eliminates intermediate phases and provides direct access to winding-sector transitions of the system. By defining a domain wall density to quantify the spatial modulation of the helical valence bond phase, we perform thermodynamic extrapolations and demonstrate that both the domain wall density and the characteristic wavevector evolve continuously with the coupling ratio, exhibiting no commensurate lock-in behavior. Our results establish that the helical valence bond phase is a genuine two-dimensional incommensurate phase with long-range bond-bond order in the thermodynamic limit, clarifying that winding-sector transitions are finite-size effects enforced by boundary commensurability. Furthermore, we determine the phase transition point between columnar valence bond solid phase and helical valence bond phase to be $g_c = 0.046(2)$.

cond-mat.str-el↗

arXiv2026

Your Teacher Can't Help You Here: Combating Supervision Fidelity Decay in On-Policy Distillation

On-policy distillation transfers reasoning capabilities by training a student model on its own generated trajectories using token-level feedback from a teacher. However, we identify a critical bottleneck, \textbf{Supervision Fidelity Decay (SFD)}: as student-generated prefixes lengthen, the teacher's next-token distribution becomes less confident and less discriminative. Consequently, the teacher-dependent corrective signal in reverse-KL distillation weakens, causing student drift to compound across long reasoning chains. To mitigate SFD, we introduce \textbf{Lookahead Group Reward (\ours{})}. Building on the insight that next-step teacher confidence reflects the discriminative strength of future reverse-KL supervision, \ours{} evaluates the student's top-K candidate tokens by the teacher confidence they induce at the subsequent step and assigns a group-normalized reward. To maintain computational efficiency, we further design an entropy-triggered tree-attention mechanism. Across six math and code benchmarks, \ours{} improves mean@8 by \textbf{2.57} points over OPD for a 7B student, with gains increasing in longer-generation and reaching +\textbf{4.92} points on AIME-26 at 39k tokens.

cs.CL↗

arXiv2026

Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillation

Multimodal Large Language Models (MLLMs) still struggle with fine-grained visual understanding, where answers often depend on small but decisive evidence in the full image. We observe a regional-to-global perception gap: the same MLLM answers fine-grained questions more accurately when conditioned on evidence-centered crops than on the corresponding full images, suggesting that many failures stem from difficulty to focus on relevant evidence rather than insufficient local recognition ability. Motivated by this observation, we propose Vision-OPD (Vision On-Policy Distillation), a regional-to-global self-distillation framework that transfers the model's own privileged regional perception to its full-image policy. Vision-OPD instantiates two conditional policies from the same MLLM: a crop-conditioned teacher and a full-image-conditioned student. The student generates on-policy rollouts, and Vision-OPD minimizes token-level divergence between the teacher and student next-token distributions along these rollouts. This enables the model to internalize the benefit of visual zooming without external teacher models, ground-truth labels, reward verifiers, or inference-time tool use. Experiments on multiple fine-grained visual understanding benchmarks show that Vision-OPD models achieve competitive or superior performance against much larger open-source, closed-source, and "Thinking-with-Images" agentic models. The code is available at https://github.com/VisionOPD/Vision-OPD

cs.CV↗

arXiv2026

Learning from Failures: Correction-Oriented Policy Optimization with Verifiable Rewards

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as an effective paradigm for improving the reasoning capabilities of large language models. However, RLVR training is often hindered by sparse binary rewards and weak credit assignment, resulting in ambiguous optimization signals and underutilization of the useful information embedded in failed trajectories. To address this challenge, we propose Correction-Oriented Policy Optimization (CIPO), a simple and effective extension to RLVR that converts on-policy failed trajectories into correction-oriented supervision, without relying on any external signals. By jointly optimizing correction samples derived from the model's own failed attempts together with the standard RLVR objective, CIPO improves learning effectiveness while explicitly enhancing the model's ability to correct its own errors. Extensive experiments across 11 benchmarks spanning mathematical reasoning and code generation demonstrate that CIPO consistently and significantly outperforms strong baselines in both reasoning and correction performance. Moreover, CIPO yields stronger pass@K gains, indicating that it improves the model's intrinsic reasoning capacity rather than merely redistributing probability mass over existing correct answers.

cs.CL↗

arXiv2026

Multimodal OCR: Parse Anything from Documents

We present Multimodal OCR (MOCR), a document parsing paradigm that jointly parses text and graphics into unified textual representations. Unlike conventional OCR systems that focus on text recognition and leave graphical regions as cropped pixels, our method, termed dots.mocr, treats visual elements such as charts, diagrams, tables, and icons as first-class parsing targets, enabling systems to parse documents while preserving semantic relationships across elements. It offers several advantages: (1) it reconstructs both text and graphics as structured outputs, enabling more faithful document reconstruction; (2) it supports end-to-end training over heterogeneous document elements, allowing models to exploit semantic relations between textual and visual components; and (3) it converts previously discarded graphics into reusable code-level supervision, unlocking multimodal supervision embedded in existing documents. To make this paradigm practical at scale, we build a comprehensive data engine from PDFs, rendered webpages, and native SVG assets, and train a compact 3B-parameter model through staged pretraining and supervised fine-tuning. We evaluate dots.mocr from two perspectives: document parsing and structured graphics parsing. On document parsing benchmarks, it ranks second only to Gemini 3 Pro on our OCR Arena Elo leaderboard, surpasses existing open-source document parsing systems, and sets a new state of the art of 83.9 on olmOCR Bench. On structured graphics parsing, our model achieves higher reconstruction quality than Gemini 3 Pro across image-to-SVG benchmarks, demonstrating strong performance on charts, UI layouts, scientific figures, and chemical diagrams. These results show a scalable path toward building large-scale image-to-code corpora for multimodal pretraining. Code and models are publicly available at https://github.com/rednote-hilab/dots.mocr.

cs.CV↗

arXiv2026

Tackling Length Inflation Without Trade-offs: Group Relative Reward Rescaling for Reinforcement Learning

Reinforcement learning significantly enhances LLM capabilities but suffers from a critical issue: length inflation, where models adopt verbosity or inefficient reasoning to maximize rewards. Prior approaches struggle to address this challenge in a general and lossless manner, primarily because additive penalties introduce a compensatory effect that creates optimization shortcuts, while heuristic gating strategies lack generality beyond binary feedback. To bridge this gap, we present Group Relative Reward Rescaling (GR$^3$), which reframes length control as a multiplicative rescaling paradigm, effectively establishing a generalized, continuous, and reward-dependent gating mechanism. To further ensure lossless optimization, we incorporate group-relative regularization and advantage-aware calibration, which dynamically adapt length budgets to instance difficulty and preserve the advantage signal of high-quality trajectories. Empirically, across both RLHF and RLVR settings, GR$^3$~maintains training dynamics and downstream performance comparable to standard GRPO while significantly mitigating length inflation, outperforming state-of-the-art length-regularized baselines.

cs.LG↗

arXiv2025

Coupled Variational Reinforcement Learning for Language Model General Reasoning

While reinforcement learning has achieved impressive progress in language model reasoning, it is constrained by the requirement for verifiable rewards. Recent verifier-free RL methods address this limitation by utilizing the probabilities that LLMs generate reference answers as reward signals. However, these approaches typically sample reasoning traces conditioned only on the question. This design decouples reasoning-trace sampling from answer information, leading to inefficient exploration and incoherence between traces and final answers. In this paper, we propose \textit{\b{Co}upled \b{V}ariational \b{R}einforcement \b{L}earning} (CoVRL), which bridges variational inference and reinforcement learning by coupling prior and posterior distributions through a hybrid sampling strategy. By constructing and optimizing a composite distribution that integrates these two distributions, CoVRL enables efficient exploration while preserving strong thought-answer coherence. Extensive experiments on mathematical and general reasoning benchmarks show that CoVRL improves performance by 12.4\% over the base model and achieves an additional 2.3\% improvement over state-of-the-art verifier-free RL baselines, providing a principled framework for enhancing the general reasoning capabilities of language models.

cs.CL↗

arXiv2025

Spectroscopic evidences for the spontaneous symmetry breaking at the $SO(5)$ deconfined critical point of $J$-$Q_3$ model

Recent numerical and theoretical studies on the two-dimensional $J$-$Q_3$ model suggests that the deconfined quantum critical point is actually a $SO(5)$-symmetry-enhanced first-order phase transition that is spontaneously broken to $O(4)$. However, this conclusion has mainly relied on finite-size scaling of the entanglement entropy, lacking direct evidence from physical observables.} Here, we investigate the dynamical spectra of spin and bond operators at the deconfined critical point of the $J$-$Q_3$ model using large-scale quantum Monte Carlo simulations, and contrasting them with the well-established $\mathrm{O(3)}$ Wilson-Fisher criticality in the $J_1$-$J_2$ Heisenberg model. Although both models exhibit two gapless magnon modes in the N\'eel phase, their critical behaviors diverge strikingly. At the $J_1$-$J_2$ critical point, the Higgs mode becomes gapless, yielding three gapless modes that reflect the full restoration of the $\mathrm{O(3)}$ symmetry. {In the $J$-$Q_3$ model, we instead observe four gapless transverse modes at the either side of the transition. This spectral feature, together with the entanglement entropy results, provides direct evidence for the weakly first-order scenario that the deconfined quantum critical point exhibits an emergent $\mathrm{SO(5)}$ symmetry that spontaneously breaks to $\mathrm{O(4)}$.

cond-mat.str-el↗

arXiv2025

PretrainZero: Reinforcement Active Pretraining

Mimicking human behavior to actively learning from general experience and achieve artificial general intelligence has always been a human dream. Recent reinforcement learning (RL) based large-thinking models demonstrate impressive expert-level abilities, i.e., software and math, but still rely heavily on verifiable rewards in specific domains, placing a significant bottleneck to extend the performance boundary of general reasoning capabilities. In this work, we propose PretrainZero, a reinforcement active learning framework built on the pretraining corpus to extend RL from domain-specific post-training to general pretraining. PretrainZero features the following characteristics: 1) Active pretraining: inspired by the active learning ability of humans, PretrainZero learns a unified reasoning policy to actively identify reasonable and informative contents from pretraining corpus, and reason to predict these contents by RL. 2) Self-supervised learning: without any verifiable labels, pretrained reward models, or supervised fine-tuning, we directly pretrain reasoners from 3 to 30B base models on the general Wikipedia corpus using RL, significantly breaking the verification data-wall for general reasoning. 3) Verification scaling: by tackling increasingly challenging masked spans, PretrainZero substantially enhances the general reasoning abilities of pretrained base models. In reinforcement pretraining, PretrainZero improves Qwen3-4B-Base for 8.43, 5.96 and 10.60 on MMLU-Pro, SuperGPQA and math average benchmarks. In post-training, the pretrained models can also serve as reasoning foundation models for downstream RLVR tasks.

cs.CL↗

arXiv2025

MemSearcher: Training LLMs to Reason, Search and Manage Memory via End-to-End Reinforcement Learning

LLM-based search agents often concatenate the full interaction history into the context, producing long and noisy inputs, and increasing compute cost and GPU memory overhead. To address this issue, we propose MemSearcher, an agent framework that maintains a compact memory during multi-turn interactions, retaining only question-relevant information and thereby keeping the context length stable across turns. Training MemSearcher is challenging because each trajectory spans multiple turns under different LLM contexts, making each turn an independent optimization target in reinforcement learning. We introduce multi-context GRPO, which propagates trajectory-level advantages to all turns for end-to-end optimization. Experiments demonstrate that MemSearcher outperforms strong history-concatenation (ReAct-style) baselines on a range of public datasets while maintaining nearly constant token counts across multi-turn interactions. The code and models will be publicly available at https://github.com/icip-cas/MemSearcher

cs.CL↗

arXiv2025

Hierarchical Trion Formation and Fractionalized Solitons in One Dimension

Stabilizing commensurate $2:1$ trions in one-dimensional quantum mixtures is typically hindered by phase separation. Within the conventional density-driven framework, the asymmetric locking mechanism generically couples to a softening density mode, creating a geometric constraint that suppresses the formation of a stable trionic liquid. We demonstrate that correlated kinetics can reorganize the low-energy structure to bypass this instability. Driven by microscopic fermion pair-hopping, the fermions first form an emergent pair liquid. At a $2:1$ filling, the density of the emergent pairs exactly matches the bosons. The low-energy theory is therefore reorganized into an effective symmetric $1:1$ pair-boson mixture. This symmetry decouples the locking mechanism from the softening mode, preempting phase separation and stabilizing a trionic liquid. The reconstructed phase exhibits a correlation hierarchy where only the composite trion retains quasi-long-range order, while the gapped relative sector supports parity-constrained topological kink excitations.

cond-mat.str-el↗

arXiv2025

dots.llm1 Technical Report

Mixture of Experts (MoE) models have emerged as a promising paradigm for scaling language models efficiently by activating only a subset of parameters for each input token. In this report, we present dots.llm1, a large-scale MoE model that activates 14B parameters out of a total of 142B parameters, delivering performance on par with state-of-the-art models while reducing training and inference costs. Leveraging our meticulously crafted and efficient data processing pipeline, dots.llm1 achieves performance comparable to Qwen2.5-72B after pretraining on 11.2T high-quality tokens and post-training to fully unlock its capabilities. Notably, no synthetic data is used during pretraining. To foster further research, we open-source intermediate training checkpoints at every one trillion tokens, providing valuable insights into the learning dynamics of large language models.

cs.CL↗

arXiv2025

Exploring Implicit Visual Misunderstandings in Multimodal Large Language Models through Attention Analysis

Recent advancements have enhanced the capability of Multimodal Large Language Models (MLLMs) to comprehend multi-image information. However, existing benchmarks primarily evaluate answer correctness, overlooking whether models genuinely comprehend the visual input. To address this, we define implicit visual misunderstanding (IVM), where MLLMs provide correct answers without fully comprehending the visual input. Through our analysis, we decouple the visual and textual modalities within the causal attention module, revealing that attention distribution increasingly converges on the image associated with the correct answer as the network layers deepen. This insight leads to the introduction of a scale-agnostic metric, \textit{attention accuracy}, and a novel benchmark for quantifying IVMs. Attention accuracy directly evaluates the model's visual understanding via internal mechanisms, remaining robust to positional biases for more reliable assessments. Furthermore, we extend our approach to finer granularities and demonstrate its effectiveness in unimodal scenarios, underscoring its versatility and generalizability.

cs.CV↗

arXiv2025

MLLM-Selector: Necessity and Diversity-driven High-Value Data Selection for Enhanced Visual Instruction Tuning

Visual instruction tuning (VIT) has emerged as a crucial technique for enabling multi-modal large language models (MLLMs) to follow user instructions adeptly. Yet, a significant gap persists in understanding the attributes of high-quality instruction tuning data and frameworks for its automated selection. To address this, we introduce MLLM-Selector, an automated approach that identifies valuable data for VIT by weighing necessity and diversity. Our process starts by randomly sampling a subset from the VIT data pool to fine-tune a pretrained model, thus creating a seed model with an initial ability to follow instructions. Then, leveraging the seed model, we calculate necessity scores for each sample in the VIT data pool to identify samples pivotal for enhancing model performance. Our findings underscore the importance of mixing necessity and diversity in data choice, leading to the creation of MLLM-Selector, our methodology that fuses necessity scoring with strategic sampling for superior data refinement. Empirical results indicate that within identical experimental conditions, MLLM-Selector surpasses LLaVA-1.5 in some benchmarks with less than 1% of the data and consistently exceeds performance across all validated benchmarks when using less than 50%.

cs.CV↗

arXiv2025

EDEA: Efficient Dual-Engine Accelerator for Depthwise Separable Convolution with Direct Data Transfer

Depthwise separable convolution (DSC) has emerged as a crucial technique, especially for resource-constrained devices. In this paper, we propose a dual-engine for the DSC hardware accelerator, which enables the full utilization of depthwise convolution (DWC) and pointwise convolution (PWC) processing elements (PEs) in all DSC layers. To determine the optimal dataflow, data reuse, and configuration of the target architecture, we conduct a design space exploration using MobileNetV1 with the CIFAR10 dataset. In the architecture, we introduce an additional non-convolutional unit, which merges the dequantization, batch normalization (BN), ReLU, and quantization between DWC and PWC into a simple fixed-point multiplication and addition operation. This also reduces the intermediate data access between the DWC and PWC, enabling streaming operation and reducing latency. The proposed DSC dual-engine accelerator is implemented using the 22nm FDSOI technology from GlobalFoundries, occupying an area of 0.58 $mm^2$. After signoff, it can operate at 1 GHz at TT corner, achieving a peak energy efficiency of 13.43 TOPS/W with a throughput of 973.55 GOPS with 8-bit precision. The average energy efficiency of all DSC layers on MobileNetV1 is 11.13 TOPS/W, demonstrating substantial hardware efficiency improvements for DSC-based applications.

cs.AR↗

arXiv2025

The Devil Is in the Details: Tackling Unimodal Spurious Correlations for Generalizable Multimodal Reward Models

Multimodal Reward Models (MM-RMs) are crucial for aligning Large Language Models (LLMs) with human preferences, particularly as LLMs increasingly interact with multimodal data. However, we find that MM-RMs trained on existing datasets often struggle to generalize to out-of-distribution data due to their reliance on unimodal spurious correlations, primarily text-only shortcuts within the training distribution, which prevents them from leveraging true multimodal reward functions. To address this, we introduce a Shortcut-aware MM-RM learning algorithm that mitigates this issue by dynamically reweighting training samples, shifting the distribution toward better multimodal understanding, and reducing dependence on unimodal spurious correlations. Our experiments demonstrate significant improvements in generalization, downstream task performance, and scalability, establishing a more robust framework for multimodal reward modeling.

cs.CL↗

arXiv2025

Ground State of $\mathrm{SU}\left(3\right)$ spin model on the checkerboard lattice

Geometric frustration in quantum spin systems can lead to exotic ground states. In this study, we investigate the $\mathrm{SU}(3)$ spin model on the checkerboard lattice to explore the effects of frustration arising from its point-connected $(N+1)$-site local structure. We employ density matrix renormalization group (DMRG) and exact diagonalization (ED) techniques to determine the ground state properties. Our results reveal the absence of both 3-sublattice antiferromagnetic order and valence cluster solid order. Instead, we identify ground states with bond stripe patterns sensitive to boundary conditions and system size, comprising staggered singlet arrays and uniform flat stripes. Notably, these stripes are relatively decoupled, and similar patterns can be reconstructed in quasi-one-dimensional ladders. These findings suggest that geometric frustration drives the system toward a mixed phase, combining characteristics of spin-liquid and valence cluster solid states, providing new insights into the behavior of frustrated quantum spin systems.

cond-mat.str-el↗