arXiv subjects

Jiajun Li

Publications and source records attributed to Jiajun Li.

At least 55 records · Page 3Linked to original sources

arXiv2025

APRIL: Active Partial Rollouts in Reinforcement Learning to Tame Long-tail Generation

Reinforcement learning (RL) has become a cornerstone in advancing large-scale pre-trained language models (LLMs). Successive generations, including GPT-o series, DeepSeek-R1, Kimi-K1.5, Grok 4, and GLM-4.5, have relied on large-scale RL training to enhance reasoning and coding capabilities. To meet the community's growing RL needs, numerous RL frameworks have been proposed. However, RL training remains computationally expensive, with rollout generation accounting for more than 90% of total runtime. In addition, its efficiency is often constrained by the long-tail distribution of rollout response lengths, where a few lengthy responses stall entire batches, leaving GPUs idle and underutilized. As model and rollout sizes continue to grow, this bottleneck increasingly limits scalability. To address this challenge, we propose Active Partial Rollouts in Reinforcement Learning (APRIL), which mitigates long-tail inefficiency. In the rollout phase, APRIL over-provisions rollout requests, terminates once the target number of responses is reached, and recycles incomplete responses for continuation in future steps. This strategy ensures that no rollouts are discarded while substantially reducing GPU idle time. Experiments show that APRIL improves rollout throughput by 22.5% on average (at most 44%) across commonly used RL algorithms (GRPO, DAPO, GSPO), accelerates convergence, and achieves 2.1% on average(at most 8%) higher final accuracy across tasks. Moreover, APRIL is both framework and hardware agnostic, already integrated into the slime RL framework, and deployable on NVIDIA and AMD GPUs alike. Taken together, this work unifies system-level and algorithmic considerations in proposing APRIL, with the aim of advancing RL training efficiency and inspiring further optimizations in RL systems. Our codebase is available at https://github.com/RLsys-Foundation/APRIL

cs.LG↗

arXiv2025

Video-LevelGauge: Investigating Contextual Positional Bias in Large Video Language Models

Large video language models (LVLMs) have made notable progress in video understanding, spurring the development of corresponding evaluation benchmarks. However, existing benchmarks generally assess overall performance across entire video sequences, overlooking nuanced behaviors such as contextual positional bias, a critical yet under-explored aspect of LVLM performance. We present Video-LevelGauge, a dedicated benchmark designed to systematically assess positional bias in LVLMs. We employ standardized probes and customized contextual setups, allowing flexible control over context length, probe position, and contextual types to simulate diverse real-world scenarios. In addition, we introduce a comprehensive analysis method that combines statistical measures with morphological pattern recognition to characterize bias. Our benchmark comprises 438 manually curated videos spanning multiple types, yielding 1,177 high-quality multiple-choice questions and 120 open-ended questions, validated for their effectiveness in exposing positional bias. Based on these, we evaluate 27 state-of-the-art LVLMs, including both commercial and open-source models. Our findings reveal significant positional biases in many leading open-source models, typically exhibiting head or neighbor-content preferences. In contrast, commercial models such as Gemini2.5-Pro show impressive, consistent performance across entire video sequences. Further analyses on context length, context variation, and model scale provide actionable insights for mitigating bias and guiding model enhancement . https://github.com/Cola-any/Video-LevelGauge

cs.CV↗

arXiv2025

Cut2Next: Generating Next Shot via In-Context Tuning

Effective multi-shot generation demands purposeful, film-like transitions and strict cinematic continuity. Current methods, however, often prioritize basic visual consistency, neglecting crucial editing patterns (e.g., shot/reverse shot, cutaways) that drive narrative flow for compelling storytelling. This yields outputs that may be visually coherent but lack narrative sophistication and true cinematic integrity. To bridge this, we introduce Next Shot Generation (NSG): synthesizing a subsequent, high-quality shot that critically conforms to professional editing patterns while upholding rigorous cinematic continuity. Our framework, Cut2Next, leverages a Diffusion Transformer (DiT). It employs in-context tuning guided by a novel Hierarchical Multi-Prompting strategy. This strategy uses Relational Prompts to define overall context and inter-shot editing styles. Individual Prompts then specify per-shot content and cinematographic attributes. Together, these guide Cut2Next to generate cinematically appropriate next shots. Architectural innovations, Context-Aware Condition Injection (CACI) and Hierarchical Attention Mask (HAM), further integrate these diverse signals without introducing new parameters. We construct RawCuts (large-scale) and CuratedCuts (refined) datasets, both with hierarchical prompts, and introduce CutBench for evaluation. Experiments show Cut2Next excels in visual consistency and text fidelity. Crucially, user studies reveal a strong preference for Cut2Next, particularly for its adherence to intended editing patterns and overall cinematic continuity, validating its ability to generate high-quality, narratively expressive, and cinematically coherent subsequent shots.

cs.CV↗

arXiv2025

Efficient Input-level Backdoor Defense on Text-to-Image Synthesis via Neuron Activation Variation

In recent years, text-to-image (T2I) diffusion models have gained significant attention for their ability to generate high quality images reflecting text prompts. However, their growing popularity has also led to the emergence of backdoor threats, posing substantial risks. Currently, effective defense strategies against such threats are lacking due to the diversity of backdoor targets in T2I synthesis. In this paper, we propose NaviT2I, an efficient input-level backdoor defense framework against diverse T2I backdoors. Our approach is based on the new observation that trigger tokens tend to induce significant neuron activation variation in the early stage of the diffusion generation process, a phenomenon we term Early-step Activation Variation. Leveraging this insight, NaviT2I navigates T2I models to prevent malicious inputs by analyzing Neuron activation variations caused by input tokens. Extensive experiments show that NaviT2I significantly outperforms the baselines in both effectiveness and efficiency across diverse datasets, various T2I backdoors, and different model architectures including UNet and DiT. Furthermore, we show that our method remains effective under potential adaptive attacks.

cs.CR↗

arXiv2025

Context-Aware Probabilistic Modeling with LLM for Multimodal Time Series Forecasting

Time series forecasting is important for applications spanning energy markets, climate analysis, and traffic management. However, existing methods struggle to effectively integrate exogenous texts and align them with the probabilistic nature of large language models (LLMs). Current approaches either employ shallow text-time series fusion via basic prompts or rely on deterministic numerical decoding that conflict with LLMs' token-generation paradigm, which limits contextual awareness and distribution modeling. To address these limitations, we propose CAPTime, a context-aware probabilistic multimodal time series forecasting method that leverages text-informed abstraction and autoregressive LLM decoding. Our method first encodes temporal patterns using a pretrained time series encoder, then aligns them with textual contexts via learnable interactions to produce joint multimodal representations. By combining a mixture of distribution experts with frozen LLMs, we enable context-aware probabilistic forecasting while preserving LLMs' inherent distribution modeling capabilities. Experiments on diverse time series forecasting tasks demonstrate the superior accuracy and generalization of CAPTime, particularly in multimodal scenarios. Additional analysis highlights its robustness in data-scarce scenarios through hybrid probabilistic decoding.

cs.LG↗

arXiv2025

SkipVAR: Accelerating Visual Autoregressive Modeling via Adaptive Frequency-Aware Skipping

Recent studies on Visual Autoregressive (VAR) models have highlighted that high-frequency components, or later steps, in the generation process contribute disproportionately to inference latency. However, the underlying computational redundancy involved in these steps has yet to be thoroughly investigated. In this paper, we conduct an in-depth analysis of the VAR inference process and identify two primary sources of inefficiency: step redundancy and unconditional branch redundancy. To address step redundancy, we propose an automatic step-skipping strategy that selectively omits unnecessary generation steps to improve efficiency. For unconditional branch redundancy, we observe that the information gap between the conditional and unconditional branches is minimal. Leveraging this insight, we introduce unconditional branch replacement, a technique that bypasses the unconditional branch to reduce computational cost. Notably, we observe that the effectiveness of acceleration strategies varies significantly across different samples. Motivated by this, we propose SkipVAR, a sample-adaptive framework that leverages frequency information to dynamically select the most suitable acceleration strategy for each instance. To evaluate the role of high-frequency information, we introduce high-variation benchmark datasets that test model sensitivity to fine details. Extensive experiments show SkipVAR achieves over 0.88 average SSIM with up to 1.81x overall acceleration and 2.62x speedup on the GenEval benchmark, maintaining model quality. These results confirm the effectiveness of frequency-aware, training-free adaptive acceleration for scalable autoregressive image generation. Our code is available at https://github.com/fakerone-li/SkipVAR and has been publicly released.

cs.CV↗

arXiv2025

Synthesis of intrinsic magnetic topological insulator MnBi2nTe3n+1 family by chemical vapor transport method with feedback regulation

MnBi2nTe3n+1 (MBT) is a representative family of intrinsic magnetic topological insulators, in which numerous exotic phenomena such as the quantum anomalous Hall effect are expected. The high-quality crystal growth and magnetism manipulation are the most essential processes. Here we develop a modified chemical vapor transport method using a feedback-regulated strategy, which provides the closed-loop control of growth temperature within +/- 0.1 degree Celsius. Single crystals of MnBi2Te4, MnBi4Te7, and MnBi6Te10 are obtained under different temperature intervals respectively, and show variable tunability on magnetism by finely tuning the growth temperatures. Specifically, the cold-end temperatures not only vary the strength of antiferromagnetic coupling in MnBi2Te4, but also induce magnetic ground state transitions from antiferromagnetism to ferromagnetism in MnBi4Te7 and MnBi6Te10. In MnBi2Te4 with optimized magnetism, quantized transport with Chern insulator state is also realized at the low field of 3.7 T. Our results provide a systematic picture for the crystal growth and the rich magnetic tunability of MBT family, providing richer platforms for the related researches combining magnetism and topological physics.

cond-mat.mtrl-sci↗

arXiv2025

Host dependence of PL5 ensemble in 4H-SiC

Color center PL5 in 4H silicon carbide (4H-SiC) has drawn significant attention due to its room-temperature quantum coherence properties and promising potential of quantum sensing applications. The preparation of PL5 ensemble is a critical prerequisite for practical applications. In this work, we investigated the formation of PL5 ensembles in types of 4H-SiC wafers, focusing on their suitability as hosts for PL5 ensemble. Results demonstrate that PL5 signals are exclusively observed in high-purity semi-insulating (HPSI) substrates, whereas divacancies PL1-PL4 can be detected in both HPSI and epitaxial samples. The type of in-plane stress in HPSI and epitaxial hosts is compressive in the same order of magnitude. Defects like stacking faults and dislocations are not observed simultaneously in the PL5 ensemble. Notably, the PL5 ensemble exhibits a relatively uniform distribution in the HPSI host, highlighting its readiness for integration into quantum sensing platforms. Furthermore, signal of PL5 can always be detected in the HPSI samples with different doses of electron irradiation, which suggests that HPSI wafers are more suitable hosts for the production of PL5 ensemble. This work provides critical insights into the material-specific requirements for PL5 ensemble formation and advances the development of 4H-SiC-based quantum technologies.

cond-mat.mtrl-sci↗

arXiv2025

Aquatic-GS: A Hybrid 3D Representation for Underwater Scenes

Representing underwater 3D scenes is a valuable yet complex task, as attenuation and scattering effects during underwater imaging significantly couple the information of the objects and the water. This coupling presents a significant challenge for existing methods in effectively representing both the objects and the water medium simultaneously. To address this challenge, we propose Aquatic-GS, a hybrid 3D representation approach for underwater scenes that effectively represents both the objects and the water medium. Specifically, we construct a Neural Water Field (NWF) to implicitly model the water parameters, while extending the latest 3D Gaussian Splatting (3DGS) to model the objects explicitly. Both components are integrated through a physics-based underwater image formation model to represent complex underwater scenes. Moreover, to construct more precise scene geometry and details, we design a Depth-Guided Optimization (DGO) mechanism that uses a pseudo-depth map as auxiliary guidance. After optimization, Aquatic-GS enables the rendering of novel underwater viewpoints and supports restoring the true appearance of underwater scenes, as if the water medium were absent. Extensive experiments on both simulated and real-world datasets demonstrate that Aquatic-GS surpasses state-of-the-art underwater 3D representation methods, achieving better rendering quality and real-time rendering performance with a 410x increase in speed. Furthermore, regarding underwater image restoration, Aquatic-GS outperforms representative dewatering methods in color correction, detail recovery, and stability. Our models, code, and datasets can be accessed at https://aquaticgs.github.io.

cs.CV↗

arXiv2025

LazyMAR: Accelerating Masked Autoregressive Models via Feature Caching

Masked Autoregressive (MAR) models have emerged as a promising approach in image generation, expected to surpass traditional autoregressive models in computational efficiency by leveraging the capability of parallel decoding. However, their dependence on bidirectional self-attention inherently conflicts with conventional KV caching mechanisms, creating unexpected computational bottlenecks that undermine their expected efficiency. To address this problem, this paper studies the caching mechanism for MAR by leveraging two types of redundancy: Token Redundancy indicates that a large portion of tokens have very similar representations in the adjacent decoding steps, which allows us to first cache them in previous steps and then reuse them in the later steps. Condition Redundancy indicates that the difference between conditional and unconditional output in classifier-free guidance exhibits very similar values in adjacent steps. Based on these two redundancies, we propose LazyMAR, which introduces two caching mechanisms to handle them one by one. LazyMAR is training-free and plug-and-play for all MAR models. Experimental results demonstrate that our method achieves 2.83 times acceleration with almost no drop in generation quality. Our codes will be released in https://github.com/feihongyan1/LazyMAR.

cs.CV↗

arXiv2025

Single Image Rolling Shutter Removal with Diffusion Models

We present RS-Diffusion, the first Diffusion Models-based method for single-frame Rolling Shutter (RS) correction. RS artifacts compromise visual quality of frames due to the row-wise exposure of CMOS sensors. Most previous methods have focused on multi-frame approaches, using temporal information from consecutive frames for the motion rectification. However, few approaches address the more challenging but important single frame RS correction. In this work, we present an ``image-to-motion" framework via diffusion techniques, with a designed patch-attention module. In addition, we present the RS-Real dataset, comprised of captured RS frames alongside their corresponding Global Shutter (GS) ground-truth pairs. The GS frames are corrected from the RS ones, guided by the corresponding Inertial Measurement Unit (IMU) gyroscope data acquired during capture. Experiments show that RS-Diffusion surpasses previous single-frame RS methods, demonstrates the potential of diffusion-based approaches, and provides a valuable dataset for further research.

cs.CV↗

arXiv2025

ProReflow: Progressive Reflow with Decomposed Velocity

Diffusion models have achieved significant progress in both image and video generation while still suffering from huge computation costs. As an effective solution, flow matching aims to reflow the diffusion process of diffusion models into a straight line for a few-step and even one-step generation. However, in this paper, we suggest that the original training pipeline of flow matching is not optimal and introduce two techniques to improve it. Firstly, we introduce progressive reflow, which progressively reflows the diffusion models in local timesteps until the whole diffusion progresses, reducing the difficulty of flow matching. Second, we introduce aligned v-prediction, which highlights the importance of direction matching in flow matching over magnitude matching. Experimental results on SDv1.5 and SDXL demonstrate the effectiveness of our method, for example, conducting on SDv1.5 achieves an FID of 10.70 on MSCOCO2014 validation set with only 4 sampling steps, close to our teacher model (32 DDIM steps, FID = 10.05).

cs.GR↗

arXiv2024

Fast and Interpretable Mixed-Integer Linear Program Solving by Learning Model Reduction

By exploiting the correlation between the structure and the solution of Mixed-Integer Linear Programming (MILP), Machine Learning (ML) has become a promising method for solving large-scale MILP problems. Existing ML-based MILP solvers mainly focus on end-to-end solution learning, which suffers from the scalability issue due to the high dimensionality of the solution space. Instead of directly learning the optimal solution, this paper aims to learn a reduced and equivalent model of the original MILP as an intermediate step. The reduced model often corresponds to interpretable operations and is much simpler, enabling us to solve large-scale MILP problems much faster than existing commercial solvers. However, current approaches rely only on the optimal reduced model, overlooking the significant preference information of all reduced models. To address this issue, this paper proposes a preference-based model reduction learning method, which considers the relative performance (i.e., objective cost and constraint feasibility) of all reduced models on each MILP instance as preferences. We also introduce an attention mechanism to capture and represent preference information, which helps improve the performance of model reduction learning tasks. Moreover, we propose a SetCover based pruning method to control the number of reduced models (i.e., labels), thereby simplifying the learning process. Evaluation on real-world MILP problems shows that 1) compared to the state-of-the-art model reduction ML methods, our method obtains nearly 20% improvement on solution accuracy, and 2) compared to the commercial solver Gurobi, two to four orders of magnitude speedups are achieved.

cs.LG↗

arXiv2024

Multi-Calorimetry in Light-based Neutrino Detectors

Neutrino detectors are among the largest photon detection instruments, built to capture scarce photons upon energy deposition. Many discoveries in neutrino physics, including the neutrino itself, are inseparable from the advances in photon detection technology, particularly in photo-sensors and readout electronics, to yield ever higher precision and richer detection information. The measurement of the energy of neutrinos, referred to as calorimetry, can be achieved in two distinct approaches: photon-counting, where single-photon can be counted digitally, and photon-integration, where multi-photons are aggregated and estimated via analogue signals. The energy is pursued today to reach permille level systematics control precision in ever-vast volumes, exemplified by experiments like JUNO. The unprecedented precision brings to the foreground the systematics due to calorimetric response entanglements in energy, position and time that were negligible in the past, thus driving further innovation in calorimetry. This publication describes a novel articulation that detectors can be endowed with multiple photon detection systems. This multi-calorimetry approach opens the notion of dual-calorimetry detector, consisting of both photon-counting and photon-integration systems, as a cost-effective evolution from the single calorimetry setups used over several decades for most experiments so far. The dual-calorimetry design exploits unique response synergies between photon-counting and photon-integration systems, including correlations and cancellations in calorimetric responses, to maximise the mitigation of response entanglements, thereby yielding permille-level high-precision calorimetry.

hep-ex↗

arXiv2024

Mozart's Touch: A Lightweight Multi-modal Music Generation Framework Based on Pre-Trained Large Models

In recent years, AI-Generated Content (AIGC) has witnessed rapid advancements, facilitating the creation of music, images, and other artistic forms across a wide range of industries. However, current models for image- and video-to-music synthesis struggle to capture the nuanced emotions and atmosphere conveyed by visual content. To fill this gap, we propose Mozart's Touch, a multi-modal music generation framework capable of generating music aligned with cross-modal inputs such as images, videos, and text. The framework consists of three key components: Multi-modal Captioning Module, Large Language Model (LLM) understanding \& Bridging Module, and Music Generation Module. Unlike traditional end-to-end methods, Mozart's Touch uses LLMs to accurately interpret visual elements without requiring the training or fine-tuning of music generation models, providing efficiency and transparency through clear, interpretable prompts. We also introduce the "LLM-Bridge" method to resolve the heterogeneous representation challenges between descriptive texts from different modalities. Through a series of objective and subjective evaluations, we demonstrate that Mozart's Touch outperforms current state-of-the-art models. Our code and examples are available at https://github.com/TiffanyBlews/MozartsTouch.

cs.SD↗

arXiv2024

Optimal Matrix Sketching over Sliding Windows

Matrix sketching, aimed at approximating a matrix $\boldsymbol{A} \in \mathbb{R}^{N\times d}$ consisting of vector streams of length $N$ with a smaller sketching matrix $\boldsymbol{B} \in \mathbb{R}^{\ell\times d}, \ell \ll N$, has garnered increasing attention in fields such as large-scale data analytics and machine learning. A well-known deterministic matrix sketching method is the Frequent Directions algorithm, which achieves the optimal $O\left(\frac{d}{\varepsilon}\right)$ space bound and provides a covariance error guarantee of $\varepsilon = \lVert \boldsymbol{A}^\top \boldsymbol{A} - \boldsymbol{B}^\top \boldsymbol{B} \rVert_2/\lVert \boldsymbol{A} \rVert_F^2$. The matrix sketching problem becomes particularly interesting in the context of sliding windows, where the goal is to approximate the matrix $\boldsymbol{A}_W$, formed by input vectors over the most recent $N$ time units. However, despite recent efforts, whether achieving the optimal $O\left(\frac{d}{\varepsilon}\right)$ space bound on sliding windows is possible has remained an open question. In this paper, we introduce the DS-FD algorithm, which achieves the optimal $O\left(\frac{d}{\varepsilon}\right)$ space bound for matrix sketching over row-normalized, sequence-based sliding windows. We also present matching upper and lower space bounds for time-based and unnormalized sliding windows, demonstrating the generality and optimality of \dsfd across various sliding window models. This conclusively answers the open question regarding the optimal space bound for matrix sketching over sliding windows. Furthermore, we conduct extensive experiments with both synthetic and real-world datasets, validating our theoretical claims and thus confirming the correctness and effectiveness of our algorithm, both theoretically and empirically.

cs.DB↗

arXiv2024

Membership Inference on Text-to-Image Diffusion Models via Conditional Likelihood Discrepancy

Text-to-image diffusion models have achieved tremendous success in the field of controllable image generation, while also coming along with issues of privacy leakage and data copyrights. Membership inference arises in these contexts as a potential auditing method for detecting unauthorized data usage. While some efforts have been made on diffusion models, they are not applicable to text-to-image diffusion models due to the high computation overhead and enhanced generalization capabilities. In this paper, we first identify a conditional overfitting phenomenon in text-to-image diffusion models, indicating that these models tend to overfit the conditional distribution of images given the corresponding text rather than the marginal distribution of images only. Based on this observation, we derive an analytical indicator, namely Conditional Likelihood Discrepancy (CLiD), to perform membership inference, which reduces the stochasticity in estimating memorization of individual samples. Experimental results demonstrate that our method significantly outperforms previous methods across various data distributions and dataset scales. Additionally, our method shows superior resistance to overfitting mitigation strategies, such as early stopping and data augmentation.

cs.CR↗

arXiv2024

SAFETY-J: Evaluating Safety with Critique

The deployment of Large Language Models (LLMs) in content generation raises significant safety concerns, particularly regarding the transparency and interpretability of content evaluations. Current methods, primarily focused on binary safety classifications, lack mechanisms for detailed critique, limiting their utility for model improvement and user trust. To address these limitations, we introduce SAFETY-J, a bilingual generative safety evaluator for English and Chinese with critique-based judgment. SAFETY-J utilizes a robust training dataset that includes diverse dialogues and augmented query-response pairs to assess safety across various scenarios comprehensively. We establish an automated meta-evaluation benchmark that objectively assesses the quality of critiques with minimal human intervention, facilitating scalable and continuous improvement. Additionally, SAFETY-J employs an iterative preference learning technique to dynamically refine safety assessments based on meta-evaluations and critiques. Our evaluations demonstrate that SAFETY-J provides more nuanced and accurate safety evaluations, thereby enhancing both critique quality and predictive reliability in complex content scenarios. To facilitate further research and application, we open-source SAFETY-J's training protocols, datasets, and code at https://github.com/GAIR-NLP/Safety-J.

cs.CL↗