arXiv subjects

Yiran Zhu

Publications and source records attributed to Yiran Zhu.

At least 19 recordsLinked to original sources

arXiv2026

ContextMaster: Interactive Multi-Shot Video Creation via Fixed-Budget Sparse Context Routing

Recent video models increasingly support generation, reference conditioning, and editing within a single model, yet typically expose them as separate operations over fixed inputs. Practical creation unfolds across multiple shots, requiring one model to generate from text, follow a reference, or edit source footage while maintaining shared history. We formalize this setting as interactive multi-shot video creation (IMVC) and introduce ContextMaster, a unified model with a role-aware context representation for these operations. An interactive model must retain access to an expanding history without allowing the context read cost at each denoising step to grow. ContextMaster combines reusable clean context states with fixed budget sparse context routing and uses ConstraintSink to keep task constraints visible. To address the dual challenges of sparse context access and inference with few denoising steps, we propose a two-stage privileged context distillation framework, which transfers full context behavior from a dense teacher through consistency distillation and then refines deployment rollouts with distribution matching. Experiments on the three primitive tasks demonstrate improved task fulfillment and consistency across shots over specialized baselines. User studies further validate flexibly composed workflows, while the model reaches 16 FPS on a single GPU.

cs.CV↗

arXiv2026

Have You Ever Seen Them? Entity-level Membership Inference through Interrogating Large Language Models

Large Language Models (LLMs) raise growing concerns about privacy leakage and copyright compliance. Membership inference is a key tool for assessing such risks, but existing studies mainly focus on whether specific samples or sample-based data units are used for training. We argue that LLMs exhibit a human-memory-like behavior: an LLM may not memorize a specific sample verbatim, yet it can accumulate and reveal knowledge about a real-world entity from scattered mentions. This analogy motivates us to examine whether an LLM can be interrogated like a human interviewee to reveal its exposure to entity-related information. Motivated by this question, we propose entity-level membership inference, which determines whether information related to a target entity is used in LLM training. We study this task in the practical label-only black-box setting, where only generated texts are observable. We formalize the task under clue, input, and model constraints, establish the necessary and sufficient conditions for its feasibility, and instantiate five interrogation strategies based on this formalization. The strategies use limited entity clues to construct prompts, elicit entity-related responses, and infer membership from semantic features among the generated texts. We construct entity-level datasets and adapt state-of-the-art sample-level label-only methods to the entity-level setting as baselines. Experiments on person entities show that our methods achieve AUC up to 0.97 and bring gains of 6.0%--17.5% in Balanced Accuracy over the best adapted baseline.

cs.CL↗

arXiv2026

Geometry-Aware Implicit Memory for Video World Models

Video world models aim to simulate controllable visual environments, but long-horizon rollouts depend on what the model remembers after observations leave its native context window. Explicit memories retain frames or online 3D reconstructions, which can suffer from heuristic retrieval errors, redundant appearance storage, or reconstruction artifacts. Implicit memories compress history into a compact state, but existing designs are not explicitly constrained to encode cross-view scene geometry. We propose GIM-World, a geometry-aware implicit memory framework for video world models. A lightweight transformer encoder compresses variable-length history into fixed-size memory tokens, a camera-queryable geometry head distills 3D scene structure from a frozen foundation model into the memory during training, and an information-guided pruning rule keeps encoding cost bounded as history grows. The geometry teacher is discarded at inference, leaving a lightweight memory module. Experiments on MIND show that GIM-World better preserves long-horizon geometric and visual consistency than both explicit- and implicit-memory baselines.

cs.CV↗

arXiv2026

A Dynamic Prognostic Prediction Method for Colorectal Cancer Liver Metastasis

Colorectal cancer liver metastasis (CRLM) exhibits high postoperative recurrence and pronounced prognostic heterogeneity, challenging individualized management. Existing prognostic approaches often rely on static representations from a single postoperative snapshot, and fail to jointly capture tumor spatial distribution, longitudinal disease dynamics, and multimodal clinical information, limiting predictive accuracy. We propose DyPro, a deep learning framework that infers postoperative latent trajectories via residual dynamic evolution. Starting from an initial patient representation, DyPro generates a 12-step sequence of trajectory snapshots through autoregressive residual updates and integrates them to predict recurrence and survival outcomes. On the MSKCC CRLM dataset, DyPro achieves strong discrimination under repeated stratified 5-fold cross-validation, reaching a C-index of 0.755 for OS and 0.714 for DFS, with OS AUC@1y of 0.920 and OS IBS of 0.143. DyPro provides quantitative risk cues to support adjuvant therapy planning and follow-up scheduling.

eess.IV↗

arXiv2026

Identity-Preserving Image-to-Video Generation via Reward-Guided Optimization

Recent advances in image-to-video (I2V) generation have achieved remarkable progress in synthesizing high-quality, temporally coherent videos from static images. Among all the applications of I2V, human-centric video generation includes a large portion. However, existing I2V models encounter difficulties in maintaining identity consistency between the input human image and the generated video, especially when the person in the video exhibits significant expression changes and movements. This issue becomes critical when the human face occupies merely a small fraction of the image. Since humans are highly sensitive to identity variations, this poses a critical yet under-explored challenge in I2V generation. In this paper, we propose Identity-Preserving Reward-guided Optimization (IPRO), a novel video diffusion framework based on reinforcement learning to enhance identity preservation. Instead of introducing auxiliary modules or altering model architectures, our approach introduces a direct and effective tuning algorithm that optimizes diffusion models using a face identity scorer. To improve performance and accelerate convergence, our method backpropagates the reward signal through the last steps of the sampling chain, enabling richer gradient feedback. We also propose a novel facial scoring mechanism that treats faces in ground-truth videos as facial feature pools, providing multi-angle facial information to enhance generalization. A KL-divergence regularization is further incorporated to stabilize training and prevent overfitting to the reward signal. Extensive experiments on Wan 2.2 I2V model and our in-house I2V model demonstrate the effectiveness of our method. Our project and code are available at https://ipro-alimama.github.io/.

cs.CV↗

arXiv2026

DuoTeach: Dual Role Self-Teaching for Coarse-to-Fine Decision Coordination in Vision--Language Models

Coarse-to-fine path decision-making requires predicting a valid taxonomy path in which earlier decisions constrain later ones. However, existing benchmarks score each level independently, obscuring cross-level validity and consistency. To better align evaluation with this setting, we introduce a Joint Path Decision (JPD) protocol that requires predicting the full path in one call, together with Depth-Weighted Prefix Accuracy (DWPA), a metric family that measures path reliability with tunable emphasis on deeper levels. Under JPD, strong vision-language models (VLMs) frequently produce invalid parent-child pairs and brittle full-path predictions, suggesting that their failures stem not only from incomplete taxonomic knowledge but also from unstable cross-level decision coordination. To address this problem, we propose DuoTeach, a dual-role self-teaching distillation framework that requires no ground-truth labels and reuses the same pretrained VLM in two roles. Its Decision-Conditioned Rollout (DCR) generates more coherent teacher traces by conditioning each level on prior decisions, and distills this coordinated behavior into the student without additional test-time rollouts. Across multiple taxonomy-structured benchmarks and VLM base models, DuoTeach improves in-domain DWPA (alpha = 0.95) by up to 30.24 points and boosts zero-shot performance on unseen taxonomies from 17.17% to 43.66%. Further analyses attribute these gains to improved within-call multi-level decision coordination.

cs.MM↗

arXiv2026

Refold: Refining Protein Inverse Folding with Efficient Structural Matching and Fusion

Protein inverse folding aims to design an amino acid sequence that will fold into a given backbone structure, serving as a central task in protein design. Two main paradigms have been widely explored. Template-based methods exploit database-derived structural priors and can achieve high local precision when close structural neighbors are available, but their dependence on database coverage and match quality often degrades performance on out-of-distribution (OOD) targets. Deep learning approaches, in contrast, learn general structure-to-sequence regularities and usually generalize better to new backbones. However, they struggle to capture fine-grained local structure, which can cause uncertain residue predictions and missed local motifs in ambiguous regions. We introduce Refold, a novel framework that synergistically integrates the strengths of database-derived structural priors and deep learning prediction to enhance inverse folding. Refold obtains structural priors from matched neighbors and fuses them with model predictions to refine residue probabilities. In practice, low-quality neighbors can introduce noise, potentially degrading model performance. We address this issue with a Dynamic Utility Gate that controls prior injection and falls back to the base prediction when the priors are untrustworthy. Comprehensive evaluations on standard benchmarks demonstrate that Refold achieves state-of-the-art native sequence recovery of 0.63 on both CATH 4.2 and CATH 4.3. Also, analysis indicates that Refold delivers larger gains on high-uncertainty regions, reflecting the complementarity between structural priors and deep learning predictions.

cs.LG↗

arXiv2025

Video Echoed in Music: Semantic, Temporal, and Rhythmic Alignment for Video-to-Music Generation

Video-to-Music generation seeks to generate musically appropriate background music that enhances audiovisual immersion for videos. However, current approaches suffer from two critical limitations: 1) incomplete representation of video details, leading to weak alignment, and 2) inadequate temporal and rhythmic correspondence, particularly in achieving precise beat synchronization. To address the challenges, we propose Video Echoed in Music (VeM), a latent music diffusion that generates high-quality soundtracks with semantic, temporal, and rhythmic alignment for input videos. To capture video details comprehensively, VeM employs a hierarchical video parsing that acts as a music conductor, orchestrating multi-level information across modalities. Modality-specific encoders, coupled with a storyboard-guided cross-attention mechanism (SG-CAtt), integrate semantic cues while maintaining temporal coherence through position and duration encoding. For rhythmic precision, the frame-level transition-beat aligner and adapter (TB-As) dynamically synchronize visual scene transitions with music beats. We further contribute a novel video-music paired dataset sourced from e-commerce advertisements and video-sharing platforms, which imposes stricter transition-beat synchronization requirements. Meanwhile, we introduce novel metrics tailored to the task. Experimental results demonstrate superiority, particularly in semantic relevance and rhythmic precision.

cs.SD↗

arXiv2025

NeuroGaze-Distill: Brain-informed Distillation and Depression-Inspired Geometric Priors for Robust Facial Emotion Recognition

Facial emotion recognition (FER) models trained only on pixels often fail to generalize across datasets because facial appearance is an indirect and biased proxy for underlying affect. We present NeuroGaze-Distill, a cross-modal distillation framework that transfers brain-informed priors into an image-only FER student via static Valence/Arousal (V/A) prototypes and a depression-inspired geometric prior (D-Geo). A teacher trained on EEG topographic maps from DREAMER (with MAHNOB-HCI as unlabeled support) produces a consolidated 5x5 V/A prototype grid that is frozen and reused; no EEG-face pairing and no non-visual signals at deployment are required. The student (ResNet-18/50) is trained on FERPlus with conventional CE/KD and two lightweight regularizers: (i) Proto-KD (cosine) aligns student features to the static prototypes; (ii) D-Geo softly shapes the embedding geometry in line with affective findings often reported in depression research (e.g., anhedonia-like contraction in high-valence regions). We evaluate both within-domain (FERPlus validation) and cross-dataset protocols (AffectNet-mini; optional CK+), reporting standard 8-way scores alongside present-only Macro-F1 and balanced accuracy to fairly handle label-set mismatch. Ablations attribute consistent gains to prototypes and D-Geo, and favor 5x5 over denser grids for stability. The method is simple, deployable, and improves robustness without architectural complexity.

cs.CV↗

arXiv2025

HER2 Expression Prediction with Flexible Multi-Modal Inputs via Dynamic Bidirectional Reconstruction

In breast cancer HER2 assessment, clinical evaluation relies on combined H&E and IHC images, yet acquiring both modalities is often hindered by clinical constraints and cost. We propose an adaptive bimodal prediction framework that flexibly supports single- or dual-modality inputs through two core innovations: a dynamic branch selector activating modality completion or joint inference based on input availability, and a cross-modal GAN (CM-GAN) enabling feature-space reconstruction of missing modalities. This design dramatically improves H&E-only accuracy from 71.44% to 94.25%, achieves 95.09% with full dual-modality inputs, and maintains 90.28% reliability under single-modality conditions. The "dual-modality preferred, single-modality compatible" architecture delivers near-dual-modality accuracy without mandatory synchronized acquisition, offering a cost-effective solution for resource-limited regions and significantly improving HER2 assessment accessibility.

cs.MM↗

arXiv2025

PupiNet: Seamless OCT-OCTA Interconversion Through Wavelet-Driven and Multi-Scale Attention Mechanisms

Optical Coherence Tomography (OCT) and Optical Coherence Tomography Angiography (OCTA) are key diagnostic tools for clinical evaluation and management of retinal diseases. Compared to traditional OCT, OCTA provides richer microvascular information, but its acquisition requires specialized sensors and high-cost equipment, creating significant challenges for the clinical deployment of hardware-dependent OCTA imaging methods. Given the technical complexity of OCTA image acquisition and potential mechanical artifacts, this study proposes a bidirectional image conversion framework called PupiNet, which accurately achieves bidirectional transformation between 3D OCT and 3D OCTA. The generator module of this framework innovatively integrates wavelet transformation and multi-scale attention mechanisms, significantly enhancing image conversion quality. Meanwhile, an Adaptive Discriminator Augmentation (ADA) module has been incorporated into the discriminator to optimize model training stability and convergence efficiency. To ensure clinical accuracy of vascular structures in the converted images, we designed a Vessel Structure Matcher (VSM) supervision module, achieving precise matching of vascular morphology between generated images and target images. Additionally, the Hierarchical Feature Calibration (HFC) module further guarantees high consistency of texture details between generated images and target images across different depth levels. To rigorously validate the clinical effectiveness of the proposed method, we conducted a comprehensive evaluation on a paired OCT-OCTA image dataset containing 300 eyes with various retinal pathologies. Experimental results demonstrate that PupiNet not only reliably achieves high-quality bidirectional transformation between the two modalities but also shows significant advantages in image fidelity, vessel structure preservation, and clinical usability.

eess.IV↗

arXiv2025

4D-ACFNet: A 4D Attention Mechanism-Based Prognostic Framework for Colorectal Cancer Liver Metastasis Integrating Multimodal Spatiotemporal Features

Postoperative prognostic prediction for colorectal cancer liver metastasis (CRLM) remains challenging due to tumor heterogeneity, dynamic evolution of the hepatic microenvironment, and insufficient multimodal data fusion. To address these issues, we propose 4D-ACFNet, the first framework that synergistically integrates lightweight spatiotemporal modeling, cross-modal dynamic calibration, and personalized temporal prediction within a unified architecture. Specifically, it incorporates a novel 4D spatiotemporal attention mechanism, which employs spatiotemporal separable convolution (reducing parameter count by 41%) and virtual timestamp encoding to model the interannual evolution patterns of postoperative dynamic processes, such as liver regeneration and steatosis. For cross-modal feature alignment, Transformer layers are integrated to jointly optimize modality alignment loss and disentanglement loss, effectively suppressing scale mismatch and redundant interference in clinical-imaging data. Additionally, we design a dynamic prognostic decision module that generates personalized interannual recurrence risk heatmaps through temporal upsampling and a gated classification head, overcoming the limitations of traditional methods in temporal dynamic modeling and cross-modal alignment. Experiments on 197 CRLM patients demonstrate that the model achieves 100% temporal adjacency accuracy (TAA), with performance significantly surpassing existing approaches. This study establishes the first spatiotemporal modeling paradigm for postoperative dynamic monitoring of CRLM. The proposed framework can be extended to prognostic analysis of multi-cancer metastases, advancing precision surgery from "spatial resection" to "spatiotemporal cure."

eess.IV↗

arXiv2025

RURANET++: An Unsupervised Learning Method for Diabetic Macular Edema Based on SCSE Attention Mechanisms and Dynamic Multi-Projection Head Clustering

Diabetic Macular Edema (DME), a prevalent complication among diabetic patients, constitutes a major cause of visual impairment and blindness. Although deep learning has achieved remarkable progress in medical image analysis, traditional DME diagnosis still relies on extensive annotated data and subjective ophthalmologist assessments, limiting practical applications. To address this, we present RURANET++, an unsupervised learning-based automated DME diagnostic system. This framework incorporates an optimized U-Net architecture with embedded Spatial and Channel Squeeze & Excitation (SCSE) attention mechanisms to enhance lesion feature extraction. During feature processing, a pre-trained GoogLeNet model extracts deep features from retinal images, followed by PCA-based dimensionality reduction to 50 dimensions for computational efficiency. Notably, we introduce a novel clustering algorithm employing multi-projection heads to explicitly control cluster diversity while dynamically adjusting similarity thresholds, thereby optimizing intra-class consistency and inter-class discrimination. Experimental results demonstrate superior performance across multiple metrics, achieving maximum accuracy (0.8411), precision (0.8593), recall (0.8411), and F1-score (0.8390), with exceptional clustering quality. This work provides an efficient unsupervised solution for DME diagnosis with significant clinical implications.

eess.IV↗

arXiv2024

An electro-optically tunable arrayed waveguide grating fabricated on thin film lithium niobate

We design and fabricate an 8-channel thin film lithium niobate (TFLN) arrayed-waveguide grating (AWG) and demonstrate the electro-optical tunability of the device. The monolithically integrated microelectrodes are designed for waveguides phase modulation and wavelength tunning. Experiments show that the fabricated electro-optically controlled TFLN AWG has a channel spacing of 200 GHz and a wavelength tuning efficiency of 10 pm/V.

physics.optics↗

arXiv2024

An integrated electro-optically tunable multi-channel interference cavity laser

We demonstrated a continuously tunable laser system by butt coupling a reflective semiconductor optical amplifier (RSOA) chip with a thin-film lithium niobate (TFLN) based multi-channel interference (MCI) cavity chip. This hybrid integrated lasers allows for fine-tuning of the laser wavelength from 1538 nm to 1560 nm with a resolution of 0.014 nm and a side-mode suppression ratio (SMSR) exceeding 30 dB. The MCI cavity chip is fabricated using the photolithography assisted chemo-mechanical etching (PLACE) technique. The developed laser has an output power of approximately 10 μW, which can be further amplified to 70 mW using a commercial erbium-doped fiber amplifier (EDFA) without significant broadening of the laser linewidth.

physics.optics↗

arXiv2024

Artificial General Intelligence (AGI) for the oil and gas industry: a review

Artificial General Intelligence (AGI) is set to profoundly impact the oil and gas industry by introducing unprecedented efficiencies and innovations. This paper explores AGI's foundational principles and its transformative applications, particularly focusing on the advancements brought about by large language models (LLMs) and extensive computer vision systems in the upstream sectors of the industry. The integration of Artificial Intelligence (AI) has already begun reshaping the oil and gas landscape, offering enhancements in production optimization, downtime reduction, safety improvements, and advancements in exploration and drilling techniques. These technologies streamline logistics, minimize maintenance costs, automate monotonous tasks, refine decision-making processes, foster team collaboration, and amplify profitability through error reduction and actionable insights extraction. Despite these advancements, the deployment of AI technologies faces challenges, including the necessity for skilled professionals for implementation and the limitations of model training on constrained datasets, which affects the models' adaptability across different contexts. The advent of generative AI, exemplified by innovations like ChatGPT and the Segment Anything Model (SAM), heralds a new era of high-density innovation. These developments highlight a shift towards natural language interfaces and domain-knowledge-driven AI, promising more accessible and tailored solutions for the oil and gas industry. This review articulates the vast potential AGI holds for tackling complex operational challenges within the upstream oil and gas industry, requiring near-human levels of intelligence. We discussed the promising applications, the hurdles of large-scale AGI model deployment, and the necessity for domain-specific knowledge in maximizing the benefits of these technologies.

cs.IT↗

arXiv2024

Large independent sets in recursive Markov random graphs

Computing the maximum size of an independent set in a graph is a famously hard combinatorial problem that has been well-studied for various classes of graphs. When it comes to random graphs, only the classical Erdős-Rényi-Gilbert random graph $G_{n,p}$ has been analysed and shown to have largest independent sets of size $Θ(\log{n})$ w.h.p. This classical model does not capture any dependency structure between edges that can appear in real-world networks. We initiate study in this direction by defining random graphs $G^{r}_{n,p}$ whose existence of edges is determined by a Markov process that is also governed by a decay parameter $r\in(0,1]$. We prove that w.h.p. $G^{r}_{n,p}$ has independent sets of size $(\frac{1-r}{2+ε}) \frac{n}{\log{n}}$ for arbitrary $ε> 0$, which implies an asymptotic lower bound of $Ω(π(n))$ where $π(n)$ is the prime-counting function. This is derived using bounds on the terms of a harmonic series, Turán bound on stability number, and a concentration analysis for a certain sequence of dependent Bernoulli variables that may also be of independent interest. Since $G^{r}_{n,p}$ collapses to $G_{n,p}$ when there is no decay, it follows that having even the slightest bit of dependency (any $r < 1$) in the random graph construction leads to the presence of large independent sets and thus our random model has a phase transition at its boundary value of $r=1$. For the maximal independent set output by a greedy algorithm, we deduce that it has a performance ratio of at most $1 + \frac{\log{n}}{(1-r)}$ w.h.p. when the lowest degree vertex is picked at each iteration, and also show that under any other permutation of vertices the algorithm outputs a set of size $Ω(n^{1/1+τ})$, where $τ=1/(1-r)$, and hence has a performance ratio of $O(n^{\frac{1}{2-r}})$.

math.CO↗

arXiv2024

AtomoVideo: High Fidelity Image-to-Video Generation

Recently, video generation has achieved significant rapid development based on superior text-to-image generation techniques. In this work, we propose a high fidelity framework for image-to-video generation, named AtomoVideo. Based on multi-granularity image injection, we achieve higher fidelity of the generated video to the given image. In addition, thanks to high quality datasets and training strategies, we achieve greater motion intensity while maintaining superior temporal consistency and stability. Our architecture extends flexibly to the video frame prediction task, enabling long sequence prediction through iterative generation. Furthermore, due to the design of adapter training, our approach can be well combined with existing personalized models and controllable modules. By quantitatively and qualitatively evaluation, AtomoVideo achieves superior results compared to popular methods, more examples can be found on our project website: https://atomo-video.github.io/.

cs.CV↗