arXiv subjects

Ka Chun Cheung

Publications and source records attributed to Ka Chun Cheung.

At least 19 recordsLinked to original sources

arXiv2026

MEDIC-AD: Towards Medical Vision-Language Model's Clinical Intelligence

Lesion detection, symptom tracking, and visual explainability are central to real-world medical image analysis, yet current medical Vision-Language Models (VLMs) still lack mechanisms that translate their broad knowledge into clinically actionable outputs. To bridge this gap, we present MEDIC-AD, a clinically oriented VLM that strengthens these three capabilities through a stage-wise framework. First, learnable anomaly-aware tokens ( ) encourage the model to focus on abnormal regions and build more discriminative lesion centered representations. Second, inter image difference tokens ( ) explicitly encode temporal changes between studies, allowing the model to distinguish worsening, improvement, and stability in disease burden. Finally, a dedicated explainability stage trains the model to generate heatmaps that highlight lesion-related regions, offering clear visual evidence that is consistent with the model's reasoning. Through our staged design, MEDIC-AD steadily boosts performance across anomaly detection, symptom tracking, and anomaly segmentation, achieving state-of-the-art results compared with both closed source and medical-specialized baselines. Evaluations on real longitudinal clinical data collected from real hospital workflows further show that MEDIC-AD delivers stable predictions and clinically faithful explanations in practical patient-monitoring and decision-support workflows

cs.CV↗

arXiv2026

Compensation-based risk-sharing

This paper studies the mathematical problem of allocating payouts (compensations) in an endowment contingency fund using a risk-sharing rule that satisfies full allocation. Besides the participants, an administrator manages the fund by collecting ex-ante contributions to establish the fund and distributing ex-post payouts to members. Two types of administrators are considered. An 'active' administrator both invests in the fund and receives the payout of the fund when no participant receives a payout. A 'passive' administrator performs only administrative tasks and neither invests in nor receives a payout from the fund. We analyze the actuarial fairness of both compensation-based risk-sharing schemes and provide general conditions under which fairness is achieved. The results extend earlier work by Denuit and Robert (2025) and Dhaene and Milevsky (2024), who focused on payouts based on Bernoulli distributions, by allowing for general non-negative loss distributions.

q-fin.RM↗

arXiv2026

Reinforcing Dual-Path Reasoning in Spatial Vision Language Models

Spatial VLMs have made substantial progress in geometric perception, yet complex spatial reasoning requiring multi-step inference over depth, distance, and scene relations remains challenging. Moreover, different spatial queries call for fundamentally different strategies: some are best addressed through purely linguistic, step-by-step deduction, while others require explicit 3D grounding before quantitative inference. We present Dual-Path Spatial Reasoning via Reinforcement Learning for Spatial VLMs (SR-REAL), a unified framework that equips a spatial VLM with two complementary reasoning paths: Language-Only Reasoning (LOR), which performs step-by-step linguistic deduction, and Detect-Then-Reason (DTR), which detects 3D geometric cues (e.g., centers or bounding boxes) via region tokens before explicit geometric inference. SR-REAL begins with a cold-start supervised fine-tuning stage that constructs LOR and DTR chain-of-thought supervision and exposes a region-to-3D interface, followed by RL that optimizes the policy model with accuracy and format rewards; for DTR, a discrete center-based detection reward further refines geometric alignment. Across diverse spatial benchmarks, SR-REAL significantly outperforms spatial VLM baselines: (i) a single RL-trained model supports both reasoning paths, with DTR excelling in region-aware tasks through precise 3D localization and LOR enhancing general spatial reasoning; (ii) jointly training both paths fosters mutual reinforcement; (iii) high-quality, blended cold-start data is crucial for stable RL optimization; and (iv) the model generalizes across datasets and domains without per-task tuning, demonstrating positive transfer between LOR and DTR.

cs.CV↗

arXiv2026

Context-Driven Incremental Compression for Multi-Turn Dialogue Generation

Modern conversational agents condition on an ever-growing dialogue history at each turn, incurring redundant attention and encoding costs that grow with conversation length. Naive truncation or summarization degrades fidelity, while existing context compressors lack cross-turn memory sharing or revision, causing information loss and compounding errors in long dialogues. We revisit the context compression under conversational dynamics and empirically present its fragility. To improve both efficiency and robustness, we introduce Context-Driven Incremental Compression (C-DIC), which treats a conversation as interleaved contextual threads and stores revisable per-thread compression states in a single, compact dialogue memory. At each turn, a lightweight retrieve, revise, and write-back loop shares information across turns and updates stale memories, stabilizing long-horizon behavior. In addition, we adapt truncated backpropagation-through-time (TBPTT) to our multi-turn setting, learning cross-turn dependencies without full-history backpropagation. Extensive experiments on long-form dialogue benchmarks demonstrate superior performance and efficiency of C-DIC; notably, C-DIC shows stable inference latency and perplexity over hundreds of dialogue turns, supporting a scalable path to high-quality dialogue modeling.

cs.CL↗

arXiv2026

DyaPlex: Full-Duplex Speech-Motion Model for Dyadic Interaction

We present DyaPlex, a streaming, full-duplex speech-and-motion model designed for dyadic interaction. To capture the continuous and reciprocal nature of human communication, this full-duplex capability empowers the agent to simultaneously perceive and generate both speech and physical motion in a streaming fashion. At its core, our method leverages the strong priors of a foundational full-duplex speech model and integrates a novel motion pathway, thereby achieving fully synchronized multi-modal interaction. Specifically, we design a dual-tower Transformer architecture that preserves the zero-shot conversational reasoning of a frozen base speech model while constructing a deeply coupled, streaming motion pathway. By introducing a unified dyadic token interleaving mechanism and guiding cross-attention via a time-aligned speech-motion RoPE, our model effectively aligns autoregressive motions with rich latent speech features. Trained on the 4,000-hour Seamless Interaction dataset, our model effectively captures cross-speaker dependencies and establishes new state-of-the-art performance across both monadic and dyadic human interaction benchmarks.

cs.CV↗

arXiv2026

Scaling Parallel Sequence Models to Foundation-Scale Vision Encoders

Vision foundation models are bottlenecked by the quadratic cost of self-attention, which limits usable resolution and increases the cost of large-scale pretraining. Subquadratic alternatives such as linear attention and state-space models reduce this cost, but often serialize images into 1D token streams and weaken the 2D spatial structure important for vision. Generalized Spatial Propagation Networks (GSPN) instead propagate context directly on the 2D grid through line-scan recurrences, achieving near-linear complexity without positional embeddings, but have seen little use as foundation-scale encoders. We present C-GSPN, a foundation-scale vision encoder based on 2D spatial propagation. C-GSPN makes the operator practical through three improvements: (1) a fast GSPN CUDA kernel that fuses per-step launches into a single warp-specialized implementation with shared-memory tiling, coalesced access, and a compact multi-channel propagation, reaching over 90% of peak memory bandwidth and running up to 40--52x faster than the original GSPN implementation; (2) a compressed latent-space propagation block with fused normalization, which turns kernel-level speed into block- and model-level efficiency; and (3) a two-stage cross-operator distillation recipe that trains the new architecture from an attention teacher without the cost of from-scratch foundation-scale training. Distilled with 600M image-text pairs, C-GSPN matches an isomorphic ViT baseline with 15% fewer parameters, improves ADE20K segmentation by +2.1%, transfers to high resolution with a fraction of the data needed from scratch, and delivers a 4x end-to-end block speedup at 2K with single-pass, tiling-free inference.

cs.CV↗

arXiv2026

Minimum-norm interpolation for unknown surface reconstruction

We study algorithms to estimate geometric properties of raw point cloud data through implicit surface representations. Given that any level-set function with a constant level set corresponding to the surface can be used for such estimations, numerical methods need not specify a unique target function for these domain-type interpolation problems. In this paper, we focus on kernel-based interpolation by radial basis functions (RBF) and reformulate the uniquely solvable interpolation problem into a constrained optimization model. This model minimizes some user-defined norm while enforcing all interpolation conditions. To enable nontrivial feasible solutions, we propose to enhance the trial space with 1D kernel basis functions inspired by Kolmogorov-Arnold Networks (KANs). Numerical experiments demonstrate that our proposed mixed-dimensional trial space significantly improves surface reconstruction from raw point clouds. This is particularly evident in the precise estimation of surface normals, outperforming traditional RBF trial spaces including the one for Hermite interpolation. This framework not only enhances processing of raw point cloud data but also shows potential for further contributions to computational geometry. We demonstrate this with a point cloud processing example.

math.NA↗

arXiv2026

LUMINA: Laplacian-Unifying Mechanism for Interpretable Neurodevelopmental Analysis via Quad-Stream GCN

Functional Magnetic Resonance Imaging(fMRI) has now become a classic way for measuring brain activity, and recent trend is shifting toward utilizing fMRI brain data for AI-driven diagnosis. Given that the brain functions as not a discrete but interconnected whole, Graph-based architectures represented by Graph Convolutional Network(GCN) has emerged as a dominant framework for such task, since they are capable of treating ROIs as dynamically interconnected nodes and extracting relational architecture between them. Ironically, however, it is the very nature of GCN's architecture that acts as an obstacle to its performance. The mathematical foundation of GCN, effective for capturing global regularities, acts as a tradeoff; by smoothing features across the connected nodes repeatedly, traditional GCN tend to blur out the contrastive dynamics that might be crucial in identifying certain neurological disorders. In order to break through this structural bottleneck, we propose LUMINA, a Laplacian-Unifying Mechanism for Interpretable Neurodevelopmental Analysis. Our model is a Quad-Stream GCN that employs a bipolar RELU activation and a dual-spectrum graph Laplacian filtering mechanism, thereby capturing heterogeneous dynamics that were often blurred out in conventional GCN. By doing so, we can preserve the diverse range and characteristics of neural connections in each fMRI data. Through 5-fold cross validation on the ADHD200(N=144) and ABIDE(N=579) dataset, LUMINA demonstrates stable diagnostic performance in two of the most critical neurodevelopmental disorder in childhood, ADHD and ASD, outperforming existing models with an accuracy of 84.66% and 88.41% each.

cs.LG↗

arXiv2026

Optimal design of reinsurance contracts with a continuum of risk assessments

In this article, we employ a principal-agent model to analyze optimal contract design in a monopolistic reinsurance market under adverse selection with a continuum of insurer types. Instead of using the classical expected utility framework, we model each insurer's risk preference through their VaR at their chosen risk tolerance level. Under informational asymmetry, the reinsurer (principal) seeks to maximize expected profit by offering an optimal menu of reinsurance contracts to a continuum of insurers (agents) with hidden characteristics. To avoid the complexity of the traditional duality approach, which yields indirect multivariate utility functions, we introduce a change of variables that reduces the problem to a univariate one. We show that the optimal indirect utility for both stop-loss and quota-share reinsurance is in stop-loss form, implying that the reinsurer will classify agents into two risk groups-high and low-even in the continuum setting. Utilizing this new class of indirect utility functions, we fully solve the problem for three common reinsurance structures: stop-loss, quota-share, and change-loss. Numerical examples are also provided for illustrating the main findings.

q-fin.RM↗

arXiv2026

Stereo-GS: Multi-View Stereo Vision Model for Generalizable 3D Gaussian Splatting Reconstruction

Generalizable 3D Gaussian Splatting reconstruction showcases advanced Image-to-3D content creation but requires substantial computational resources and large datasets, posing challenges to training models from scratch. Current methods usually entangle the prediction of 3D Gaussian geometry and appearance, which rely heavily on data-driven priors and result in slow regression speeds. To address this, we propose \method, a disentangled framework for efficient 3D Gaussian prediction. Our method extracts features from local image pairs using a stereo vision backbone and fuses them via global attention blocks. Dedicated point and Gaussian prediction heads generate multi-view point-maps for geometry and Gaussian features for appearance, combined as GS-maps to represent the 3DGS object. A refinement network enhances these GS-maps for high-quality reconstruction. Unlike existing methods that depend on camera parameters, our approach achieves pose-free 3D reconstruction, improving robustness and practicality. By reducing resource demands while maintaining high-quality outputs, \method provides an efficient, scalable solution for real-world 3D content generation.

cs.CV↗

arXiv2025

GSPN-2: Efficient Parallel Sequence Modeling

Efficient vision transformer remains a bottleneck for high-resolution images and long-video related real-world applications. Generalized Spatial Propagation Network (GSPN) addresses this by replacing quadratic self-attention with a line-scan propagation scheme, bringing the cost close to linear in the number of rows or columns, while retaining accuracy. Despite this advancement, the existing GSPN implementation still suffers from (i) heavy overhead due to repeatedly launching GPU kernels, (ii) excessive data transfers from global GPU memory, and (iii) redundant computations caused by maintaining separate propagation weights for each channel. We introduce GSPN-2, a joint algorithm-system redesign. In particular, we eliminate thousands of micro-launches from the previous implementation into one single 2D kernel, explicitly pin one warp to each channel slice, and stage the previous column's activations in shared memory. On the model side, we introduce a compact channel propagation strategy that replaces per-channel matrices, trimming parameters, and align naturally with the affinity map used in transformer attention. Experiments demonstrate GSPN-2's effectiveness across image classification and text-to-image synthesis tasks, matching transformer-level accuracy with significantly lower computational cost. GSPN-2 establishes a new efficiency frontier for modeling global spatial context in vision applications through its unique combination of structured matrix transformations and GPU-optimized implementation. Project page: https://whj363636.github.io/GSPN2/

cs.LG↗

arXiv2025

ImageSentinel: Protecting Visual Datasets from Unauthorized Retrieval-Augmented Image Generation

The widespread adoption of Retrieval-Augmented Image Generation (RAIG) has raised significant concerns about the unauthorized use of private image datasets. While these systems have shown remarkable capabilities in enhancing generation quality through reference images, protecting visual datasets from unauthorized use in such systems remains a challenging problem. Traditional digital watermarking approaches face limitations in RAIG systems, as the complex feature extraction and recombination processes fail to preserve watermark signals during generation. To address these challenges, we propose ImageSentinel, a novel framework for protecting visual datasets in RAIG. Our framework synthesizes sentinel images that maintain visual consistency with the original dataset. These sentinels enable protection verification through randomly generated character sequences that serve as retrieval keys. To ensure seamless integration, we leverage vision-language models to generate the sentinel images. Experimental results demonstrate that ImageSentinel effectively detects unauthorized dataset usage while preserving generation quality for authorized applications. Code is available at https://github.com/luo-ziyuan/ImageSentinel.

cs.CV↗

arXiv2025

QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs

We propose QeRL, a Quantization-enhanced Reinforcement Learning framework for large language models (LLMs). While RL is essential for LLMs' reasoning capabilities, it is resource-intensive, requiring substantial GPU memory and long rollout durations. QeRL addresses these issues by combining NVFP4 quantization with Low-Rank Adaptation (LoRA), accelerating rollout phase of RL while reducing memory overhead. Beyond efficiency, our findings show that quantization noise increases policy entropy, enhancing exploration, and enabling the discovery of better strategies during RL. To further optimize exploration, QeRL introduces an Adaptive Quantization Noise (AQN) mechanism, which dynamically adjusts noise during training. Experiments demonstrate that QeRL delivers over 1.5 times speedup in the rollout phase. Moreover, this is the first framework to enable RL training of a 32B LLM on a single H100 80GB GPU, while delivering overall speedups for RL training. It also achieves faster reward growth and higher final accuracy than 16-bit LoRA and QLoRA, while matching the performance of full-parameter fine-tuning on mathematical benchmarks such as GSM8K (90.8%) and MATH 500 (77.4%) in the 7B model. These results establish QeRL as an efficient and effective framework for RL training in LLMs.

cs.LG↗

arXiv2025

Biological Pathway Informed Models with Graph Attention Networks (GATs)

Biological pathways map gene-gene interactions that govern all human processes. Despite their importance, most ML models treat genes as unstructured tokens, discarding known pathway structure. The latest pathway-informed models capture pathway-pathway interactions, but still treat each pathway as a "bag of genes" via MLPs, discarding its topology and gene-gene interactions. We propose a Graph Attention Network (GAT) framework that models pathways at the gene level. We show that GATs generalize much better than MLPs, achieving an 81% reduction in MSE when predicting pathway dynamics under unseen treatment conditions. We further validate the correctness of our biological prior by encoding drug mechanisms via edge interventions, boosting model robustness. Finally, we show that our GAT model is able to correctly rediscover all five gene-gene interactions in the canonical TP53-MDM2-MDM4 feedback loop from raw time-series mRNA data, demonstrating potential to generate novel biological hypotheses directly from experimental data.

cs.LG↗

arXiv2025

Align 3D Representation and Text Embedding for 3D Content Personalization

Recent advances in NeRF and 3DGS have significantly enhanced the efficiency and quality of 3D content synthesis. However, efficient personalization of generated 3D content remains a critical challenge. Current 3D personalization approaches predominantly rely on knowledge distillation-based methods, which require computationally expensive retraining procedures. To address this challenge, we propose \textbf{Invert3D}, a novel framework for convenient 3D content personalization. Nowadays, vision-language models such as CLIP enable direct image personalization through aligned vision-text embedding spaces. However, the inherent structural differences between 3D content and 2D images preclude direct application of these techniques to 3D personalization. Our approach bridges this gap by establishing alignment between 3D representations and text embedding spaces. Specifically, we develop a camera-conditioned 3D-to-text inverse mechanism that projects 3D contents into a 3D embedding aligned with text embeddings. This alignment enables efficient manipulation and personalization of 3D content through natural language prompts, eliminating the need for computationally retraining procedures. Extensive experiments demonstrate that Invert3D achieves effective personalization of 3D content. Our work is available at: https://github.com/qsong2001/Invert3D.

cs.CV↗

arXiv2025

One-Minute Video Generation with Test-Time Training

Transformers today still struggle to generate one-minute videos because self-attention layers are inefficient for long context. Alternatives such as Mamba layers struggle with complex multi-scene stories because their hidden states are less expressive. We experiment with Test-Time Training (TTT) layers, whose hidden states themselves can be neural networks, therefore more expressive. Adding TTT layers into a pre-trained Transformer enables it to generate one-minute videos from text storyboards. For proof of concept, we curate a dataset based on Tom and Jerry cartoons. Compared to baselines such as Mamba~2, Gated DeltaNet, and sliding-window attention layers, TTT layers generate much more coherent videos that tell complex stories, leading by 34 Elo points in a human evaluation of 100 videos per method. Although promising, results still contain artifacts, likely due to the limited capability of the pre-trained 5B model. The efficiency of our implementation can also be improved. We have only experimented with one-minute videos due to resource constraints, but the approach can be extended to longer videos and more complex stories. Sample videos, code and annotations are available at: https://test-time-training.github.io/video-dit

cs.CV↗

arXiv2025

M3Net: Multimodal Multi-task Learning for 3D Detection, Segmentation, and Occupancy Prediction in Autonomous Driving

The perception system for autonomous driving generally requires to handle multiple diverse sub-tasks. However, current algorithms typically tackle individual sub-tasks separately, which leads to low efficiency when aiming at obtaining full-perception results. Some multi-task learning methods try to unify multiple tasks with one model, but do not solve the conflicts in multi-task learning. In this paper, we introduce M3Net, a novel multimodal and multi-task network that simultaneously tackles detection, segmentation, and 3D occupancy prediction for autonomous driving and achieves superior performance than single task model. M3Net takes multimodal data as input and multiple tasks via query-token interactions. To enhance the integration of multi-modal features for multi-task learning, we first propose the Modality-Adaptive Feature Integration (MAFI) module, which enables single-modality features to predict channel-wise attention weights for their high-performing tasks, respectively. Based on integrated features, we then develop task-specific query initialization strategies to accommodate the needs of detection/segmentation and 3D occupancy prediction. Leveraging the properly initialized queries, a shared decoder transforms queries and BEV features layer-wise, facilitating multi-task learning. Furthermore, we propose a Task-oriented Channel Scaling (TCS) module in the decoder to mitigate conflicts between optimizing for different tasks. Additionally, our proposed multi-task querying and TCS module support both Transformer-based decoder and Mamba-based decoder, demonstrating its flexibility to different architectures. M3Net achieves state-of-the-art multi-task learning performance on the nuScenes benchmarks.

cs.CV↗

arXiv2025

Parallel Sequence Modeling via Generalized Spatial Propagation Network

We present the Generalized Spatial Propagation Network (GSPN), a new attention mechanism optimized for vision tasks that inherently captures 2D spatial structures. Existing attention models, including transformers, linear attention, and state-space models like Mamba, process multi-dimensional data as 1D sequences, compromising spatial coherence and efficiency. GSPN overcomes these limitations by directly operating on spatially coherent image data and forming dense pairwise connections through a line-scan approach. Central to GSPN is the Stability-Context Condition, which ensures stable, context-aware propagation across 2D sequences and reduces the effective sequence length to $\sqrt{N}$ for a square map with N elements, significantly enhancing computational efficiency. With learnable, input-dependent weights and no reliance on positional embeddings, GSPN achieves superior spatial fidelity and state-of-the-art performance in vision tasks, including ImageNet classification, class-guided image generation, and text-to-image generation. Notably, GSPN accelerates SD-XL with softmax-attention by over $84\times$ when generating 16K images.

cs.CV↗