Shiyu Shen

Shiyu Shen, Bin Pan, Tianyang Shi et al.

Domain invariant learning aims to learn models that extract invariant features over various training domains, resulting in better generalization to unseen target domains. Recently, Bayesian Neural Networks have achieved promising results in domain invariant learning, but most works concentrate on aligning features distributions rather than parameter distributions. Inspired by the principle of Bayesian Neural Network, we attempt to directly learn the domain invariant posterior distribution of network parameters. We first propose a theorem to show that the invariant posterior of parameters can be implicitly inferred by aggregating posteriors on different training domains. Our assumption is more relaxed and allows us to extract more domain invariant information. We also propose a simple yet effective method, named PosTerior Generalization (PTG), that can be used to estimate the invariant parameter distribution. PTG fully exploits variational inference to approximate parameter distributions, including the invariant posterior and the posteriors on training domains. Furthermore, we develop a lite version of PTG for widespread applications. PTG shows competitive performance on various domain generalization benchmarks on DomainBed. Additionally, PTG can use any existing domain generalization methods as its prior, and combined with previous state-of-the-art method the performance can be further improved. Code will be made public.

Keshu Wu, Chenchen Kuai, Zihao Li et al.

Retrieval-augmented generation (RAG) enhances large language models by grounding outputs in retrieved knowledge. However, existing RAG methods including graph- and hypergraph-based approaches treat retrieved evidence as an unordered set, implicitly assuming permutation invariance. This assumption is misaligned with many real-world reasoning tasks, where outcomes depend not only on which interactions occur, but also on the order in which they unfold. We propose Order-Aware Knowledge Hypergraph RAG (OKH-RAG), which treats order as a first-class structural property. OKH-RAG represents knowledge as higher-order interactions within a hypergraph augmented with precedence structure, and reformulates retrieval as sequence inference over hyperedges. Instead of selecting independent facts, it recovers coherent interaction trajectories that reflect underlying reasoning processes. A learned transition model infers precedence directly from data without requiring explicit temporal supervision. We evaluate OKH-RAG on order-sensitive question answering and explanation tasks, including tropical cyclone and port operation scenarios. OKH-RAG consistently outperforms permutation-invariant baselines, and ablations show that these gains arise specifically from modeling interaction order. These results highlight a key limitation of set-based retrieval: effective reasoning requires not only retrieving relevant evidence, but organizing it into structured sequences.

Shiyu Shen, Bin Pan, Tianyang Shi et al.

Bayesian Neural Networks (BNNs) have become one of the promising approaches for uncertainty estimation due to the solid theorical foundations. However, the performance of BNNs is affected by the ability of catching uncertainty. Instead of only seeking the distribution of neural network weights by in-distribution (ID) data, in this paper, we propose a new Bayesian Neural Network with an Attached structure (ABNN) to catch more uncertainty from out-of-distribution (OOD) data. We first construct a mathematical description for the uncertainty of OOD data according to the prior distribution, and then develop an attached Bayesian structure to integrate the uncertainty of OOD data into the backbone network. ABNN is composed of an expectation module and several distribution modules. The expectation module is a backbone deep network which focuses on the original task, and the distribution modules are mini Bayesian structures which serve as attachments of the backbone. In particular, the distribution modules aim at extracting the uncertainty from both ID and OOD data. We further provide theoretical analysis for the convergence of ABNN, and experimentally validate its superiority by comparing with some state-of-the-art uncertainty estimation methods Code will be made available.

Zhe Gao, Shiyu Shen, Taifeng Chai et al.

Existing Multimodal Large Language Models (MLLMs) often suffer from hallucinations in long video understanding (LVU), primarily due to the imbalance between textual and visual tokens. Observing that MLLMs handle short visual inputs well, recent LVU works alleviate hallucinations by automatically parsing the vast visual data into manageable segments that can be effectively processed by MLLMs. SFT-based tool-calling methods can serve this purpose, but they typically require vast amounts of fine-grained, high-quality data and suffer from constrained tool-calling trajectories. We propose a novel VideoTIR that leverages Reinforcement Learning (RL) to encourage proper usage of comprehensive multi-level toolkits for efficient long video understanding. VideoTIR explores both Zero-RL and SFT cold-starting to enable MLLMs to retrieve and focus on meaningful video segments/images/regions, enhancing long video understanding both accurately and efficiently. To reduce redundant tool-calling, we propose Toolkit Action Grouped Policy Optimization (TAGPO), which enhances the efficiency of the calling process through stepwise reward assignment and reuse of failed rollouts. Additionally, we develop a sandbox-based trajectory synthesis framework to generate high-quality trajectories data. Extensive experiments on three long-video QA benchmarks demonstrate the effectiveness and efficiency of our method.

Shiyu Shen, Zhe Gao, Taifeng Chai et al.

Deep learning has revolutionized solar image analysis, yet most approaches train task-specific encoders from scratch or rely on natural-image pretraining that ignores the unique characteristics of Solar Dynamics Observatory (SDO) data. We introduce SolarCHIP, a family of contrastively pretrained visual backbones tailored to multi-instrument SDO observations. SolarCHIP addresses three key challenges in solar imaging: multimodal sensing across AIA and HMI instruments, weak inter-class separability due to slow temporal evolution, and strong intra-class variability with sparse activity signals. Our pretraining framework employs a multi-granularity contrastive objective that jointly aligns (1) global class tokens across co-temporal AIA-HMI pairs to enhance temporal discrimination, (2) local patch tokens at fixed spatial indices to enforce position-consistent, modality-invariant features, and (3) intra-sample patches across different spatial locations to preserve fine-grained spatial structure. We train both CNN- and Vision Transformer-based autoencoders and demonstrate their effectiveness on two downstream tasks: cross-modal translation between HMI and AIA passbands via ControlNet, and full-disk flare classification. Experimental results show that SolarCHIP achieves state-of-the-art performance across both tasks, with particularly strong gains in low-resource settings where labeled data is limited. Ablation studies confirm that each contrastive component contributes essential discriminative capacity at different granularities. By publicly releasing pretrained weights and training code, we provide the heliophysics community with a practical, plug-and-play feature extractor that reduces computational requirements, improves label efficiency, and establishes a reusable foundation for diverse solar imaging applications.

Bin Pan, Shiyu Shen, Zongbin Wang et al.

Domain generalization seeks to develop models trained on a limited set of source domains that are capable of generalizing effectively to unseen target domains. While the predominant approach leverages large-scale pre-trained vision models as initialization, recent studies have highlighted that full fine-tuning can compromise the intrinsic generalization capabilities of these models. To address this limitation, parameter-efficient adaptation strategies have emerged, wherein only a subset of model parameters is selectively fine-tuned, thereby balancing task adaptation with the preservation of generalization. Motivated by this paradigm, we introduce Joint Parameter Selection (JPS), a novel method that restricts updates to a small, sparse subset of parameters, thereby retaining and harnessing the generalization strength of pre-trained models. Theoretically, we establish a generalization error bound that explicitly accounts for the sparsity of parameter updates, thereby providing a principled justification for selective fine-tuning. Practically, we design a selection mechanism employing dual operators to identify and update parameters exhibiting consistent and significant gradients across all source domains. Extensive benchmark experiments demonstrate that JPS achieves superior performance compared to state-of-the-art domain generalization methods, substantiating both the efficiency and efficacy of the proposed approach.

Shiyu Shen, Bin Pan, Ziye Zhang et al.

We address hyperspectral image (HSI) synthesis, a problem that has garnered growing interest yet remains constrained by the conditional generative paradigms that limit sample diversity. While diffusion models have emerged as a state-of-the-art solution for high-fidelity image generation, their direct extension from RGB to hyperspectral domains is challenged by the high spectral dimensionality and strict physical constraints inherent to HSIs. To overcome the challenges, we introduce a diffusion framework explicitly guided by hyperspectral unmixing. The approach integrates two collaborative components: (i) an unmixing autoencoder that projects generation from the image domain into a low-dimensional abundance manifold, thereby reducing computational burden while maintaining spectral fidelity; and (ii) an abundance diffusion process that enforces non-negativity and sum-to-one constraints, ensuring physical consistency of the synthesized data. We further propose two evaluation metrics tailored to hyperspectral characteristics. Comprehensive experiments, assessed with both conventional measures and the proposed metrics, demonstrate that our method produces HSIs with both high quality and diversity, advancing the state of the art in hyperspectral data generation.

Shiyu Shen, Bin Pan, Guirong Xue

City-scale traffic volume prediction plays a pivotal role in intelligent transportation systems, yet remains a challenge due to the inherent incompleteness and bias in observational data. Although deep learning-based methods have shown considerable promise, most existing approaches produce deterministic point estimates, thereby neglecting the uncertainty arising from unobserved traffic flows. Furthermore, current models are typically trained in a city-specific manner, which hinders their generalizability and limits scalability across diverse urban contexts. To overcome these limitations, we introduce TrafficPPT, a Pretrained Probabilistic Transformer designed to model traffic volume as a distributional aggregation of trajectories. Our framework fuses heterogeneous data sources-including real-time observations, historical trajectory data, and road network topology-enabling robust and uncertainty-aware traffic inference. TrafficPPT is initially pretrained on large-scale simulated data spanning multiple urban scenarios, and later fine-tuned on target cities to ensure effective domain adaptation. Experiments on real-world datasets show that TrafficPPT consistently surpasses state-of-the-art baselines, particularly under conditions of extreme data sparsity. Code will be open.

Qiaoying Qu, Shiyu Shen

Autoregressive conditional image generation algorithms are capable of generating photorealistic images that are consistent with given textual or image conditions, and have great potential for a wide range of applications. Nevertheless, the majority of popular autoregressive image generation methods rely heavily on vector quantization, and the inherent discrete characteristic of codebook presents a considerable challenge to achieving high-quality image generation. To address this limitation, this paper introduces a novel conditional introduction network for continuous masked autoregressive models. The proposed self-control network serves to mitigate the negative impact of vector quantization on the quality of the generated images, while simultaneously enhancing the conditional control during the generation process. In particular, the self-control network is constructed upon a continuous mask autoregressive generative model, which incorporates multimodal conditional information, including text and images, into a unified autoregressive sequence in a serial manner. Through a self-attention mechanism, the network is capable of generating images that are controllable based on specific conditions. The self-control network discards the conventional cross-attention-based conditional fusion mechanism and effectively unifies the conditional and generative information within the same space, thereby facilitating more seamless learning and fusion of multimodal features.

Zongbin Wang, Bin Pan, Shiyu Shen et al.

Domain generalization (DG) aims to train a model from limited source domains, allowing it to generalize to unknown target domains. Typically, DG models only employ large-scale pre-trained models during the initialization of fine-tuning. However, large-scale pre-trained models already possess the ability to resist domain shift. If we reference pre-trained models continuously during fine-tuning to maintain this ability, it could further enhance the generalization ability of the DG model. For this purpose, we introduce a new method called Fine-Tune with Large-scale pre-trained Priors (FT-LP), which incorporates the pre-trained model as a prior into the DG fine-tuning process, ensuring that the model refers to its pre-trained model at each optimization step. FT-LP comprises a theoretical framework and a simple implementation strategy. In theory, we verify the rationality of FT-LP by introducing a generalization error bound with the pre-trained priors for DG. In implementation, we utilize an encoder to simulate the model distribution, enabling the use of FT-LP when only pre-trained weights are available. In summary, we offer a new fine-tuning method for DG algorithms to utilize pre-trained models throughout the fine-tuning process. Through experiments on various datasets and DG models, our proposed method exhibits significant improvements, indicating its effectiveness.

Shiyu Shen, Feng He, Zhichuang Liang et al.

In this work, we make \emph{systematic} optimizations of key encapsulation mechanisms (KEM) based on module learning-with-errors (MLWE), covering algorithmic design, fundamental operation of number-theoretic transform (NTT), approaches to expanding encapsulated key size, and optimized implementation coding. We focus on Kyber (now in the Round-3 finalist of NIST PQC standardization) and Aigis (a variant of Kyber proposed at PKC 2020). By careful analysis, we first observe that the algorithmic design of Kyber and Aigis can be optimized by the mechanism of asymmetric key consensus with noise (AKCN) proposed in \cite{JZ16,JZ19}. Specifically, the decryption process can be simplified with AKCN, leading to a both faster and less error-prone decryption process. Moreover, the AKCN-based optimized version has perfect compatibility with the deployment of Kyber/Aigis in reality, as they can run on the same parameters, the same public key, and the same encryption process. We make a systematic study of the variants of NTT proposed in recent years for extending its applicability scope, make concrete analysis of their exact computational complexity, and in particular show their equivalence. We then present a new variant named hybrid-NTT (H-NTT), combining the advantages of existing NTT methods, and derive its optimality in computational complexity. The H-NTT technique not only has larger applicability scope but also allows for modular and unified implementation codes of NTT operations even with varying module dimensions. We analyze and compare the different approaches to expand the size of key to be encapsulated (specifically, 512-bit key for dimension of 1024), and conclude with the most economic approach. To mitigate the compatibility issue in implementations we adopt the proposed H-NTT method.