Shikui Tu

Wenjing Huang, Shikui Tu, Lei Xu

Diffusion models have demonstrated their ability to generate diverse and high-quality images, sparking considerable interest in their potential for real image editing applications. However, existing diffusion-based approaches for local image editing often suffer from undesired artifacts due to the latent-level blending of the noised target images and diffusion latent variables, which lack the necessary semantics for maintaining image consistency. To address these issues, we propose PFB-Diff, a Progressive Feature Blending method for Diffusion-based image editing. Unlike previous methods, PFB-Diff seamlessly integrates text-guided generated content into the target image through multi-level feature blending. The rich semantics encoded in deep features and the progressive blending scheme from high to low levels ensure semantic coherence and high quality in edited images. Additionally, we introduce an attention masking mechanism in the cross-attention layers to confine the impact of specific words to desired regions, further improving the performance of background editing and multi-object replacement. PFB-Diff can effectively address various editing tasks, including object/background replacement and object attribute editing. Our method demonstrates its superior performance in terms of editing accuracy and image quality without the need for fine-tuning or training. Our implementation is available at https://github.com/CMACH508/PFB-Diff.

Guang Lin, Shikui Tu, Lei Xu

Generating molecules that simultaneously satisfy drug-like properties and conform to the 3D structure of a target protein is a core challenge in structure-based drug design (SBDD). Existing generative approaches, however, often rely on costly post-hoc processing during Sampling or require carefully curated datasets during training, yet still achieve modest gains. These limitations are especially pronounced in multi-objective settings, where balancing conflicting criteria remains a core challenge. To address these challenges, We propose FTDiff, a reinforcement learning fine-tuning framework tailored for diffusion-based molecular generation under structural constraints. To ensure stable and sample-efficient optimization, FTDiff adopts a group relative policy optimization (GRPO) style strategy. Furthermore, FTDiff builds upon a time-free pretrained diffusion model and incorporates a fast sampling mechanism that reduces the number of denoising steps, significantly accelerating both training and inference while maintaining generation quality. By optimizing a fixed threshold-aware reward, FTDiff effectively guides the model to produce valid, diverse, and high- quality molecules that balance multiple drug design objectives. Extensive experiments on benchmark datasets demonstrate that FTDiff consistently outperforms prior methods, without requiring expensive post-hoc optimization or intricate data engineering.

Tengjie Li, Shikui Tu, Lei Xu

Recent advances in vision-language models have facilitated progress in sketch generation. However, existing specialized methods primarily focus on generic synthesis and lack mechanisms for precise control over sketch styles. In this work, we propose a training-free framework based on diffusion models that enables explicit style guidance via textual prompts and referenced style sketches. Unlike previous style transfer methods that overwrite key and value matrices in self-attention, we incorporate the reference features as auxiliary information with linear smoothing and leverage a style-content guidance mechanism. This design effectively reduces content leakage from reference sketches and enhances synthesis quality, especially in cases with low structural similarity between reference and target sketches. Furthermore, we extend our framework to support controllable multi-style generation by integrating features from multiple reference sketches, coordinated via a joint AdaIN module. Extensive experiments demonstrate that our approach achieves high-quality sketch generation with accurate style alignment and improved flexibility in style control. The official implementation of M3S is available at https://github.com/CMACH508/M3S.

Wenjing Huang, Shikui Tu, Lei Xu

Semantic face editing has achieved substantial progress in recent years. Known as a growingly popular method, latent space manipulation performs face editing by changing the latent code of an input face to liberate users from painting skills. However, previous latent space manipulation methods usually encode an entire face into a single low-dimensional embedding, which constrains the reconstruction capacity and the control flexibility of facial components, such as eyes and nose. This paper proposes IA-FaceS as a bidirectional method for disentangled face attribute manipulation as well as flexible, controllable component editing without the need for segmentation masks or sketches in the original image. To strike a balance between the reconstruction capacity and the control flexibility, the encoder is designed as a multi-head structure to yield embeddings for reconstruction and control, respectively: a high-dimensional tensor with spatial properties for consistent reconstruction and four low-dimensional facial component embeddings for semantic face editing. Manipulating the separate component embeddings can help achieve disentangled attribute manipulation and flexible control of facial components. To further disentangle the highly-correlated components, a component adaptive modulation (CAM) module is proposed for the decoder. The semantic single-eye editing is developed for the first time without any input visual guidance, such as segmentation masks or sketches. According to the experimental results, IA-FaceS establishes a good balance between maintaining image details and performing flexible face manipulation. Both quantitative and qualitative results indicate that the proposed method outperforms the other techniques in reconstruction, face attribute manipulation, and component transfer.

Sicong Zang, Shikui Tu, Lei Xu

Graphic sketch representations are effective for representing sketches. Existing methods take the patches cropped from sketches as the graph nodes, and construct the edges based on sketch's drawing order or Euclidean distances on the canvas. However, the drawing order of a sketch may not be unique, while the patches from semantically related parts of a sketch may be far away from each other on the canvas. In this paper, we propose an order-invariant, semantics-aware method for graphic sketch representations. The cropped sketch patches are linked according to their global semantics or local geometric shapes, namely the synonymous proximity, by computing the cosine similarity between the captured patch embeddings. Such constructed edges are learnable to adapt to the variation of sketch drawings, which enable the message passing among synonymous patches. Aggregating the messages from synonymous patches by graph convolutional networks plays a role of denoising, which is beneficial to produce robust patch embeddings and accurate sketch representations. Furthermore, we enforce a clustering constraint over the embeddings jointly with the network learning. The synonymous patches are self-organized as compact clusters, and their embeddings are guided to move towards their assigned cluster centroids. It raises the accuracy of the computed synonymous proximity. Experimental results show that our method significantly improves the performance on both controllable sketch synthesis and sketch healing.

Yupei Yang, Biwei Huang, Fan Feng et al.

General intelligence requires quick adaption across tasks. While existing reinforcement learning (RL) methods have made progress in generalization, they typically assume only distribution changes between source and target domains. In this paper, we explore a wider range of scenarios where not only the distribution but also the environment spaces may change. For example, in the CoinRun environment, we train agents from easy levels and generalize them to difficulty levels where there could be new enemies that have never occurred before. To address this challenging setting, we introduce a causality-guided self-adaptive representation-based approach, called CSR, that equips the agent to generalize effectively across tasks with evolving dynamics. Specifically, we employ causal representation learning to characterize the latent causal variables within the RL system. Such compact causal representations uncover the structural relationships among variables, enabling the agent to autonomously determine whether changes in the environment stem from distribution shifts or variations in space, and to precisely locate these changes. We then devise a three-step strategy to fine-tune the causal model under different scenarios accordingly. Empirical experiments show that CSR efficiently adapts to the target domains with only a few samples and outperforms state-of-the-art baselines on a wide range of scenarios, including our simulated environments, CartPole, CoinRun and Atari games.

Yupei Yang, Biwei Huang, Shikui Tu et al.

The effectiveness of model training heavily relies on the quality of available training resources. However, budget constraints often impose limitations on data collection efforts. To tackle this challenge, we introduce causal exploration in this paper, a strategy that leverages the underlying causal knowledge for both data collection and model training. We, in particular, focus on enhancing the sample efficiency and reliability of the world model learning within the domain of task-agnostic reinforcement learning. During the exploration phase, the agent actively selects actions expected to yield causal insights most beneficial for world model training. Concurrently, the causal knowledge is acquired and incrementally refined with the ongoing collection of data. We demonstrate that causal exploration aids in learning accurate world models using fewer data and provide theoretical guarantees for its convergence. Empirical experiments, on both synthetic data and real-world applications, further validate the benefits of causal exploration.

Yiwei Wu, Yifan Wu, Yunhao Xiong et al.

Constructing behavioral-level chiplet models (e.g., SystemC) is crucial for early-stage heterogeneous architecture exploration. Traditional manual modeling is notoriously time-consuming and error-prone. Recently, Large Language Models (LLMs) have demonstrated immense potential in automating hardware code generation. However, existing LLM-assisted design frameworks predominantly target highly structured or well-defined design specifications. In practical engineering scenarios, raw datasheets typically encompass lengthy, complex, and highly unstructured information. Consequently, reliable code generation directly from these raw datasheets suffers from severe challenges, including context vanishing and logical hallucinations.To overcome this critical bottleneck, this paper proposes DS2SC-Agent(Datasheet-to-SystemC-Agent): the first end-to-end, fully automated generation pipeline that translates raw datasheets directly into SystemC chiplet models. This system establishes a highly efficient multi-agent collaborative framework. By decoupling the intricate modeling tasks, the proposed pipeline orchestrates a fully automated workflow encompassing unstructured long-document parsing, SystemC core code construction, testbench stimulus generation, and adaptive closed-loop debugging. We comprehensively evaluate the proposed framework on representative single-function chiplets across the analog, digital, and radio frequency (RF) domains--specifically, a Limiting Amplifier (LA), a Fast Fourier Transform (FFT) module, and a Power Amplifier (PA). The evaluation demonstrates that our pipeline seamlessly processes complex real-world datasheets to consistently generate functionally correct SystemC models. This provides a highly efficient and reliable paradigm for agile model library construction while drastically minimizing manual intervention.

Yupei Yang, Lin Yang, Wanxi Deng et al.

A reliable reward model is essential for aligning large language models with human preferences through reinforcement learning from human feedback. However, standard reward models are susceptible to spurious features that are not causally related to human labels. This can lead to reward hacking, where high predicted reward does not translate into better behavior. In this work, we address this problem from a causal perspective by proposing a factored representation learning framework that decomposes the model's contextual embedding into (1) causal factors that are sufficient for reward prediction and (2) non-causal factors that capture reward-irrelevant attributes such as length or sycophantic bias. The reward head is then constrained to depend only on the causal component. In addition, we introduce an adversarial head trained to predict reward from the non-causal factors, while applying gradient reversal to discourage them from encoding reward-relevant information. Experiments on both mathematical and dialogue tasks demonstrate that our method learns more robust reward models and consistently improves downstream RLHF performance over state-of-the-art baselines. Analyses on length and sycophantic bias further validate the effectiveness of our method in mitigating reward hacking behaviors.

Yupei Yang, Lin Yang, Wanxi Deng et al.

Reranking is a critical component in many information retrieval pipelines. Despite remarkable progress in text-only settings, multimodal reranking remains challenging, particularly when the candidate set contains hybrid text and image items. A key difficulty is the modality gap: a text reranker is intrinsically closer to text candidates than to image candidates, leading to biased and suboptimal cross-modal ranking. Vision-language models (VLMs) mitigate this gap through strong cross-modal alignment and have recently been adopted to build multimodal rerankers. However, most VLM-based rerankers encode all candidates as images, and treating text as images introduces substantial computational overhead. Meanwhile, existing open-source multimodal rerankers are typically trained on general-domain data and often underperform in domain-specific scenarios. To address these limitations, we propose UniRank, a VLM-based reranking framework that natively scores and orders hybrid text-image candidates without any modality conversion. Building on this hybrid scoring interface, UniRank provides an end-to-end domain adaptation pipeline that includes: (1) an instruction-tuning stage that learns calibrated cross-modal relevance scoring by mapping label-token likelihoods to a unified scalar score; and (2) a hard-negative-driven preference alignment stage that constructs in-domain pairwise preferences and performs query-level policy optimization through reinforcement learning from human feedback (RLHF). Extensive experiments on scientific literature retrieval and design patent search demonstrate that UniRank consistently outperforms state-of-the-art baselines, improving Recall@1 by 8.9% and 7.3%, respectively.

Hao Qian, Shikui Tu, Lei Xu

Diffusion and flow matching models have recently emerged as promising approaches for peptide binder design. Despite their progress, these models still face two major challenges. First, categorical sampling of discrete residue types collapses their continuous parameters into onehot assignments, while continuous variables (e.g., atom positions) evolve smoothly throughout the generation process. This mismatch disrupts the update dynamics and results in suboptimal performance. Second, current models assume unimodal distributions for side-chain torsion angles, which conflicts with the inherently multimodal nature of side chain rotameric states and limits prediction accuracy. To address these limitations, we introduce PepBFN, the first Bayesian flow network for full atom peptide design that directly models parameter distributions in fully continuous space. Specifically, PepBFN models discrete residue types by learning their continuous parameter distributions, enabling joint and smooth Bayesian updates with other continuous structural parameters. It further employs a novel Gaussian mixture based Bayesian flow to capture the multimodal side chain rotameric states and a Matrix Fisher based Riemannian flow to directly model residue orientations on the $\mathrm{SO}(3)$ manifold. Together, these parameter distributions are progressively refined via Bayesian updates, yielding smooth and coherent peptide generation. Experiments on side chain packing, reverse folding, and binder design tasks demonstrate the strong potential of PepBFN in computational peptide design.

Jingyuan Zhou, Hao Qian, Shikui Tu et al.

Structure-based drug design (SBDD), aiming to generate 3D molecules with high binding affinity toward target proteins, is a vital approach in novel drug discovery. Although recent generative models have shown great potential, they suffer from unstable probability dynamics and mismatch between generated molecule size and the protein pockets geometry, resulting in inconsistent quality and off-target effects. We propose PAFlow, a novel target-aware molecular generation model featuring prior interaction guidance and a learnable atom number predictor. PAFlow adopts the efficient flow matching framework to model the generation process and constructs a new form of conditional flow matching for discrete atom types. A protein-ligand interaction predictor is incorporated to guide the vector field toward higher-affinity regions during generation, while an atom number predictor based on protein pocket information is designed to better align generated molecule size with target geometry. Extensive experiments on the CrossDocked2020 benchmark show that PAFlow achieves a new state-of-the-art in binding affinity (up to -8.31 Avg. Vina Score), simultaneously maintains favorable molecular properties.

Yupei Yang, Fan Feng, Lin Yang et al.

Relation extraction enables the construction of structured knowledge for many downstream applications. While large language models (LLMs) have shown great promise in this domain, most existing methods concentrate on relation classification, which predicts the semantic relation type between a related entity pair. However, we observe that LLMs often struggle to reliably determine whether a relation exists, especially in cases involving complex sentence structures or intricate semantics, which leads to spurious predictions. Such hallucinations can introduce noisy edges in knowledge graphs, compromising the integrity of structured knowledge and downstream reliability. To address these challenges, we propose DEPTH, a framework that integrates Dependency-aware sEntence simPlification and Two-tiered Hierarchical refinement into the relation extraction pipeline. Given a sentence and its candidate entity pairs, DEPTH operates in two stages: (1) the Grounding module extracts relations for each pair by leveraging their shortest dependency path, distilling the sentence into a minimal yet coherent relational context that reduces syntactic noise while preserving key semantics; (2) the Refinement module aggregates all local predictions and revises them based on a holistic understanding of the sentence, correcting omissions and inconsistencies. We further introduce a causality-driven reward model that mitigates reward hacking by disentangling spurious correlations, enabling robust fine-tuning via reinforcement learning with human feedback. Experiments on six benchmarks demonstrate that DEPTH reduces the average hallucination rate to 7.0\% while achieving a 17.2\% improvement in average F1 score over state-of-the-art baselines.

Dengdeng Huang, Shikui Tu

Deep generative models provide a promising approach to de novo 3D peptide design. Most of them jointly model the distributions of peptide's position, orientation, and conformation, attempting to simultaneously converge to the target pocket. However, in the early stage of docking, optimizing conformation-only modalities such as rotation and torsion can be physically meaningless, as the peptide is initialized far from the protein pocket and no interaction field is present. We define this problem as the multimodal temporal inconsistency problem and claim it is a key factor contributing to low binding affinity in generated peptides. To address this challenge, we propose THFlow, a novel flow matching-based multimodal generative model that explicitly models the temporal hierarchy between peptide position and conformation. It employs a polynomial based conditional flow to accelerate positional convergence early on, and later aligns it with rotation and torsion for coordinated conformation refinement under the emerging interaction field. Additionally, we incorporate interaction-related features, such as polarity, to further enhance the model's understanding of peptide-protein binding. Extensive experiments demonstrate that THFlow outperforms existing methods in generating peptides with superior stability, affinity, and diversity, offering an effective and accurate solution for advancing peptide-based therapeutic development.

Peiying Li, Shikui Tu, Lei Xu

Current face recognition tasks are usually carried out on high-quality face images, but in reality, most face images are captured under unconstrained or poor conditions, e.g., by video surveillance. Existing methods are featured by learning data uncertainty to avoid overfitting the noise, or by adding margins to the angle or cosine space of the normalized softmax loss to penalize the target logit, which enforces intra-class compactness and inter-class discrepancy. In this paper, we propose a deep Rival Penalized Competitive Learning (RPCL) for deep face recognition in low-resolution (LR) images. Inspired by the idea of the RPCL, our method further enforces regulation on the rival logit, which is defined as the largest non-target logit for an input image. Different from existing methods that only consider penalization on the target logit, our method not only strengthens the learning towards the target label, but also enforces a reverse direction, i.e., becoming de-learning, away from the rival label. Comprehensive experiments demonstrate that our method improves the existing state-of-the-art methods to be very robust for LR face recognition.

Zhihao Xing, Shikui Tu, Lei Xu

We present a self-learning approach that combines deep reinforcement learning and Monte Carlo tree search to solve the traveling salesman problem. The proposed approach has two advantages. First, it adopts deep reinforcement learning to compute the value functions for decision, which removes the need of hand-crafted features and labelled data. Second, it uses Monte Carlo tree search to select the best policy by comparing different value functions, which increases its generalization ability. Experimental results show that the proposed method performs favorably against other methods in small-to-medium problem settings. And it shows comparable performance as state-of-the-art in large problem setting.

Wenjing Huang, Shikui Tu, Lei Xu

Proposed in 1991, Least Mean Square Error Reconstruction for self-organizing network, shortly Lmser, was a further development of the traditional auto-encoder (AE) by folding the architecture with respect to the central coding layer and thus leading to the features of symmetric weights and neurons, as well as jointly supervised and unsupervised learning. However, its advantages were only demonstrated in a one-hidden-layer implementation due to the lack of computing resources and big data at that time. In this paper, we revisit Lmser from the perspective of deep learning, develop Lmser network based on multiple convolutional layers, which is more suitable for image-related tasks, and confirm several Lmser functions with preliminary demonstrations on image recognition, reconstruction, association recall, and so on. Experiments demonstrate that Lmser indeed works as indicated in the original paper, and it has promising performance in various applications.

Xiaojie Gao, Shikui Tu, Lei Xu

We propose a planning-based method to teach an agent to manage portfolio from scratch. Our approach combines deep reinforcement learning techniques with search techniques like AlphaGo. By uniting the advantages in A* search algorithm with Monte Carlo tree search, we come up with a new algorithm named A* tree search in which best information is returned to guide next search. Also, the expansion mode of Monte Carlo tree is improved for a higher utilization of the neural network. The suggested algorithm can also optimize non-differentiable utility function by combinatorial search. This technique is then used in our trading system. The major component is a neural network that is trained by trading experiences from tree search and outputs prior probability to guide search by pruning away branches in turn. Experimental results on simulated and real financial data verify the robustness of the proposed trading system and the trading system produces better strategies than several approaches based on reinforcement learning.

Minchao Li, Shikui Tu, Lei Xu

Neural style transfer has been demonstrated to be powerful in creating artistic image with help of Convolutional Neural Networks (CNN). However, there is still lack of computational analysis of perceptual components of the artistic style. Different from some early attempts which studied the style by some pre-processing or post-processing techniques, we investigate the characteristics of the style systematically based on feature map produced by CNN. First, we computationally decompose the style into basic elements using not only spectrum based methods including Fast Fourier Transform (FFT), Discrete Cosine Transform (DCT) but also latent variable models such Principal Component Analysis (PCA), Independent Component Analysis (ICA). Then, the decomposition of style induces various ways of controlling the style elements which could be embedded as modules in state-of-the-art style transfer algorithms. Such decomposition of style brings several advantages. It enables the computational coding of different artistic styles by our style basis with similar styles clustering together, and thus it facilitates the mixing or intervention of styles based on the style basis from more than one styles so that compound style or new style could be generated to produce styled images. Experiments demonstrate the effectiveness of our method on not only painting style transfer but also sketch style transfer which indicates possible applications on picture-to-sketch problems.

Yajing Chen, Shikui Tu, Yuqi Yi et al.

Sketch is an important media for human to communicate ideas, which reflects the superiority of human intelligence. Studies on sketch can be roughly summarized into recognition and generation. Existing models on image recognition failed to obtain satisfying performance on sketch classification. But for sketch generation, a recent study proposed a sequence-to-sequence variational-auto-encoder (VAE) model called sketch-rnn which was able to generate sketches based on human inputs. The model achieved amazing results when asked to learn one category of object, such as an animal or a vehicle. However, the performance dropped when multiple categories were fed into the model. Here, we proposed a model called sketch-pix2seq which could learn and draw multiple categories of sketches. Two modifications were made to improve the sketch-rnn model: one is to replace the bidirectional recurrent neural network (BRNN) encoder with a convolutional neural network(CNN); the other is to remove the Kullback-Leibler divergence from the objective function of VAE. Experimental results showed that models with CNN encoders outperformed those with RNN encoders in generating human-style sketches. Visualization of the latent space illustrated that the removal of KL-divergence made the encoder learn a posterior of latent space that reflected the features of different categories. Moreover, the combination of CNN encoder and removal of KL-divergence, i.e., the sketch-pix2seq model, had better performance in learning and generating sketches of multiple categories and showed promising results in creativity tasks.