Large Vision-Language Models (LVLMs) have recently advanced robotic manipulation by leveraging vision for scene perception and language for instruction following. However, existing methods rely heavily on costly human-annotated training datasets, which limits their generalization and causes them to struggle in out-of-domain (OOD) scenarios, reducing real-world adaptability. To address these challenges, we propose ManipLVM-R1, a novel reinforcement learning framework that replaces traditional supervision with Reinforcement Learning using Verifiable Rewards (RLVR). By directly optimizing for task-aligned outcomes, our method enhances generalization and physical reasoning while removing the dependence on costly annotations. Specifically, we design two rule-based reward functions targeting key robotic manipulation subtasks: an Affordance Perception Reward to enhance localization of interaction regions, and a Trajectory Match Reward to ensure the physical plausibility of action paths. These rewards provide immediate feedback and impose spatial-logical constraints, encouraging the model to go beyond shallow pattern matching and instead learn deeper, more systematic reasoning about physical interactions. Experimental results show that ManipLVM-R1 achieves substantial performance gains across multiple manipulation tasks, using only 50% of the training data while achieving strong generalization to OOD scenarios. We further analyze the benefits of our reward design and its impact on task success and efficiency.

Zirui Song, Qian Jiang, Mingxuan Cui, Mingzhe Li, Lang Gao, Zeyu Zhang, Zixiang Xu, Yanbo Wang, Guangxian Ouyang, Zhenhao Chen, Xiuying Chen. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Abstract Given a finite abelian group G and a subset $$J\subset G$$ J ⊂ G with $$0\in J$$ 0 ∈ J , let $$D_{G}(J,N)$$ D G ( J , N ) be the maximum size of $$A\subset G^{N}$$ A ⊂ G N such that the difference set $$A-A$$ A - A and $$J^{N}$$ J N have no non-trivial intersection. Recently, this extremal problem has been widely studied for different groups G and subsets J . In this paper, we generalize and improve the relevant results by Alon and by Hegedűs by building a bridge between this problem and cyclotomic polynomials with the help of algebraic graph theory. In particular, we construct infinitely many non-trivial families of G and J for which the current known upper bounds on $$D_{G}(J, N)$$ D G ( J , N ) can be improved exponentially.

Zixiang Xu, Yanbo Wang, Yue Huang, Xiuying Chen, Jieyu Zhao, Meng Jiang, Xiangliang Zhang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Alzheimer's Disease (AD) is the most prevalent neurodegenerative disorder in aging populations, posing a significant and escalating burden on global healthcare systems. While Multi-Tusk Learning (MTL) has emerged as a powerful computational paradigm for modeling longitudinal AD data, existing frameworks do not account for the time-varying nature of feature correlations. To address this limitation, we propose a novel MTL framework, named Feature Similarity Laplacian graph Multi-Task Learning (MTL-FSL). Our framework introduces a novel Feature Similarity Laplacian (FSL) penalty that explicitly models the time-varying relationships between features. By simultaneously considering temporal smoothness among tasks and the dynamic correlations among features, our model enhances both predictive accuracy and biological interpretability. To solve the non-smooth optimization problem arising from our proposed penalty terms, we adopt the Alternating Direction Method of Multipliers (ADMM) algorithm. Experiments conducted on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset demonstrate that our proposed MTLFSL framework achieves state-of-the-art performance, outperforming various baseline methods. The implementation source can be found at https://github.com/huatxxx/MTL-FSL.

Large Language Models (LLMs) often struggle to maintain their original performance when faced with semantically coherent but task-irrelevant contextual information. Although prior studies have explored this issue using fixed-template or retrieval-based distractions, such static methods show limited effectiveness against contemporary models. To address this problem, we propose a dynamic distraction generation framework based on tree search, where the generation process is guided by model behavior. Without modifying the original question or answer, the method efficiently produces challenging adaptive distractions across multiple datasets, enabling systematic stress testing of LLMs’ contextual robustness. Experiments on four benchmarks demonstrate that the generated distractions lead to an average performance drop of over 45\% for mainstream models. Further comparisons of mitigation strategies show that prompt-based optimization methods yield limited gains, whereas post-training approaches (e.g., DPO) significantly enhance the model's contextual robustness. The results indicate that these issues do not stem from knowledge deficits in LLMs, but from a fundamental inability to maintain consistent reasoning under contextual distraction, posing a major challenge to the reliability of LLMs in real-world applications.

Agent systems based on large language models (LLMs) have shown great potential in complex reasoning tasks, but building efficient and generalizable workflows remains a major challenge. Most existing approaches rely on manually designed processes, which limits their adaptability across different tasks. While a few methods attempt automated workflow generation, they are often tied to specific datasets or query types and make limited use of intermediate feedback, reducing system robustness and reasoning depth. Moreover, their operations are typically predefined and inflexible. To address these limitations, we propose **DyFlow**, a dynamic workflow generation framework that adaptively constructs and adjusts reasoning procedures based on task requirements and real-time intermediate feedback, thereby enhancing cross-task generalization. DyFlow consists of two core components: a designer and an executor. The designer decomposes complex problems into a sequence of sub-goals defined by high-level objectives and dynamically plans the next steps based on intermediate outputs and feedback. These plans are then carried out by the executor, which executes each operation using dynamic operators with context-aware parameterization, enabling flexible and semantically grounded reasoning. We systematically evaluate DyFlow across diverse domains, including social reasoning, biomedical tasks, mathematical problem solving, and code generation. Results demonstrate that DyFlow significantly outperforms existing baselines, achieving substantial Pass@k improvements and exhibiting robust generalization across diverse domains.

Function-correcting codes (FCCs), introduced by Lenz, Bitar, Wachter-Zeh, and Yaakobi, protect specific function values of a message rather than the entire message. A central challenge is determining the optimal redundancy—the minimum additional information required to recover function values amid errors. This redundancy depends on both the number of correctable errorstand the structure of message vectors yielding identical function values. While prior works established bounds, key questions remain, such as the optimal redundancy for functions like Hamming weight and Hamming weight distribution, along with efficient code constructions. In this paper, we make the following contributions: 1) For the Hamming weight function, we improve the lower bound on optimal redundancy from 10(t-1)/3 to 4t− 4/3√6t+ 2 + 2. On the other hand, we provide a systematical approach to constructing explicit FCCs via a novel connection with Gray codes, which also improve the previous upper bound from 4t-2/1−2√ln(2t)/(2t) to 4t− [logt]. Consequently, we almost determine the optimal redundancy for Hamming weight function. 2) The Hamming weight distribution function is defined by the value of Hamming weight divided by a given integerT∈ N. Previous work established that the optimal redundancy is 2twhenT> 2t, while the caseT≤ 2tremained unclear. We show that the optimal redundancy remains 2twhenT≥t+ 1. However, in the surprising regime whereT=o(t), we achieve near-optimal redundancy of 4t−o(t). Our results reveal a significant distinction in behavior of redundancy for distinct choices of T.

We establish a novel connection between the well-known chromatic threshold problem in extremal combinatorics and the celebrated (p, q)-theorem in discrete geometry. In particular, for a graph G with bounded clique number and a natural density condition, we prove a (p, q)-theorem for an abstract convexity space associated with G. Our result strengthens those of Thomassen and Nikiforov on the chromatic threshold of cliques. Our (p, q)-theorem can also be viewed as a χ-boundedness result for (what we call) ultra maximal Kr-free graphs. We further show that the graphs under study are blow-ups of constant size graphs, improving a result of Oberkampf and Schacht on homomorphism threshold of cliques. Our result unravels the cause underpinning such a blow-up phenomenon, differentiating the chromatic and homomorphism threshold problems for cliques. Our result implies that for the homomorphism threshold problem, rather than the minimum degree condition usually considered in the literature, the decisive factor is a clique density condition on co-neighborhoods of vertices. More precisely, we show that if an n-vertex Kr-free graph G satisfies that the common neighborhood of every pair of non-adjacent vertices induces a subgraph with Kr−2-density at least ε > 0, then G must be a blow-up of some Kr-free graph F on at most 2 O(rε log 1ε ) vertices. Furthermore, this single exponential bound is optimal.

The centralized coded caching focuses on reducing the network burden in peak times in a wireless network system. In this paper, motivated by the study of the only rainbow 3-term arithmetic progressions set, we propose a combinatorial framework for constructing coded caching schemes. This framework builds bridges between coded caching schemes and lots of combinatorial objects due to the freedom of the choices of families and binary operations. We prove that any scheme based on a placement delivery array (PDA) can be represented by a rainbow scheme under this framework and lots of other known schemes can also be included in this framework. Moreover, we also present a new coded caching scheme with linear subpacketization and near constant rate using the only rainbow 3-term arithmetic progressions set. Finally, we modify the framework to be applicable to the coded caching problem in Device-to-Device (D2D) networks and the distributed computing problem.

Recently, deep convolution neural networks (CNNs) steered face super-resolution methods have achieved great progress in restoring degraded facial details by joint training with facial priors. However, these methods have some obvious limitations. On the one hand, multi-task joint learning requires additional marking on the dataset, and the introduced prior network will significantly increase the computational cost of the model. On the other hand, the limited receptive field of CNN will reduce the fidelity and naturalness of the reconstructed facial images, resulting in suboptimal reconstructed images. In this work, we propose an efficient CNN-Transformer Cooperation Network (CTCNet) for face super-resolution tasks, which uses the multi-scale connected encoder-decoder architecture as the backbone. Specifically, we first devise a novel Local-Global Feature Cooperation Module (LGCM), which is composed of a Facial Structure Attention Unit (FSAU) and a Transformer block, to promote the consistency of local facial detail and global facial structure restoration simultaneously. Then, we design an efficient Feature Refinement Module (FRM) to enhance the encoded features. Finally, to further improve the restoration of fine facial details, we present a Multi-scale Feature Fusion Unit (MFFU) to adaptively fuse the features from different stages in the encoder procedure. Extensive evaluations on various datasets have assessed that the proposed CTCNet can outperform other state-of-the-art methods significantly. Source code will be available at https://github.com/IVIPLab/CTCNet.

Virtual Reality (VR) technology has been shown to achieve remarkable results in multiple fields. Due to the nature of the immersive medium of Virtual Reality it logically follows that it can be used as a high-quality educational tool as it offers potentially a higher bandwidth than other mediums such as text, pictures and videos. This short paper illustrates the development of a climate change educational awareness application for virtual reality to simulate virtual scenes of local scenery and sea level rising until 2100 using prediction data. The paper also reports on the current in progress work of porting the system to Augmented Reality (AR) and future work to evaluate the system.

Light Field Displays (LFD) offer the potential for a true window into a virtual world without any form of headset. However, the vast majority of LFD only provide motion parallax along the x-axis (horizontal) due to its domination of human 3D perception. The additional motion parallax along y and the z-axis are rarely or never achieved and let alone evaluated. This paper proposed the first approach that provided the real full-motion parallax covering the z-axis. Moreover, this generic approach enables on-demand y and z-axis motion parallax on any off-the-shelf LFD. A prototype was created according to the proposed approach, and with the use of it, an experiment with two tasks was carried out among 24 participants. This pioneering study explored the effect of the motion parallax along the additional axes. Subjective and objective metrics were collected to measure participants’ 3D perception on four viewing conditions (LFD with motion parallax along the x-axis, x-and-y-axis, x-and-z-axis, and x-y-and-z-axis). The study manifested three main findings: 1) The additional y-axis and z-axis motion parallax increased viewers’ engagement with the LFD. 2) LFD with full-motion parallax provided the optimal user experience. 3) The additional motion parallax along with the y-axis increased viewers’ user experience more than the z-axis.

Augmented Reality (AR) has shown significant potential in supporting in-situ decision-making in various application areas. Many prior works have shown how AR can visualize the decision support data in various contexts. However, prior research about AR-based decision support systems rarely explored how the explanations were visualized. Providing context-aware explanations within AR-based recommendation systems may help users instantly understand the recommendations they have been given. Therefore, this paper presents the world-first user study exploring AR explanation visualization designs. Three feature importance analysis visualizations that apply different color-coding and size-scaling strategies were designed to explain the recommendations provided by a context-aware AR shopping assistant system. Twenty-four participants were recruited to evaluate these three explanations in a shopping scenario. The results revealed novel findings that could help guide the appropriate utilization of descriptive parameters when designing AR explanation artifacts. The results also show the potential of providing intuitive visualization to explain recommendations in AR.

In this paper, we first study a new extremal problem recently posed by Conlon and Tyomkyn [ Repeated Patterns in Proper Colourings, preprint, https://arxiv.org/abs/2002.00921 (2020)]. Given a graph $H$ and an integer $k\geqslant 2$, let $f_{k}(n,H)$ be the smallest number of colors $c$ such that there exists a proper edge coloring of the complete graph $K_{n}$ with $c$ colors containing no $k$ vertex-disjoint color-isomorphic copies of $H$. Using algebraic properties of polynomials over finite fields, we give an explicit proper edge coloring of $K_{n}$ and show that $f_{k}(n, C_{4})=\Theta(n)$ when $k\geqslant 3$ and $n\rightarrow\infty$. The methods we used in the edge coloring may be of some independent interest. We also consider a related generalized Ramsey problem. For given graphs $G$ and $H,$ let $r(G,H,q)$ be the minimum number of edge colors (not necessarily proper) of $G$, such that the edges of every copy of $H\subseteq G$ together receive at least $q$ distinct colors. Establishing the relation to the Turán number of specified bipartite graphs, we obtain some general lower bounds for $r(K_{n,n},K_{s,t},q)$ with a broad range of $q$.

For a subset $I\subseteq \mathbb{F}_{q}^{n}$, let $\Delta(I)$ be the set of distances determined by the elements of $I.$ The Erdös--Falconer distance problem in $\mathbb{F}_{q}^{n}$ asks for a threshold on the cardinality $|I|$ so that $\Delta(I)$ contains a positive proportion of the whole distance set. In this paper, we consider the analogous question under Hamming distance, which is the most important metric in coding theory. When $q\geqslant 4$ is a fixed prime power and $n$ goes to infinity, our main result shows that, for arbitrary positive proportion $\alpha,$ we can find $\alpha n$ distinct Hamming distances in $\Delta(I)$ if $|I|>q^{(1-\beta)\cdot n},$ where $\beta$ is a positive number depending on $\alpha.$ Unlike using Fourier analytical method as usual, our main tools include the celebrated dependent random choice and some results from additive number theory and coding theory. Hence our bound is much smaller than the previously known bound which was obtained by Fourier analytic machinery.

BACKGROUND: Genome-scale models of metabolism have only been analyzed with the constraint-based modelling philosophy. Some gene deletion studies on in silico organism models at genome-scale have been made, but most of them were from the aspects of distinguishing lethal and non-lethal genes or growth rate. The impact of gene deletion on flux redistribution, the functions and characters of key genes, and the performance of different reactions in entire gene deletion still lack research. RESULTS: Three main researches have been done into the metabolism of E. coli in gene deletion. The first work was about finding key genes and subsystems: First, by calculating the deletion impact p of whole 1261 genes, one by one, on the metabolic flux redistribution of E. coli_iAF1260, we can find that p is more detailed in describing the change of organism's metabolism. Next, we sought out 195 important (high-p) genes, and they are more than essential genes (growth rate f becomes zero if deleting). So we speculated that under some circumstances and when an important gene is deleted, a big change in the metabolic system of E. coli has taken place and E. coli may use other reaction ways to strive to live. Further, by determining the functional subsystems to which 195 key genes belong, we found that their distribution to subsystems was not even and most of them were related to just three subsystems and that all of the 8 important but not essential genes appear just in "Oxidative Phosphorylation". Our second work was about p's three characters: We analyzed the correlation between p and d (connection degree of one gene) and the correlation between p and vgene (flux sum controlled by one gene), and found that both of them are not of linear correlation, but the correlation between p and f is of highly linear correlation. The third work was about highly-affected reactions: We found 16 reactions with more than 2000 Rg value (measuring the impact that a reaction is gotten in the whole 1261 gene deletion). We speculated that highly-affected reactions involve in the metabolism of basic biomasses. CONCLUSION: To sum up, these results we obtained have biological significances and our researches will shed new light on the future researches.