Language model inversion (LMI), i.e., recovering hidden prompts from outputs, emerges as a concrete threat to user privacy and system security. We recast LMI as reusing the LLM's own latent space and propose the Invariant Latent Space Hypothesis (ILSH): (1) diverse outputs from the same source prompt should preserve consistent semantics (source invariance), and (2) inputoutput cyclic mappings should be self-consistent within a shared latent space (cyclic invariance). Accordingly, we present Inv2A, which treats the LLM as an invariant decoder and learns only a lightweight inverse encoder that maps outputs to a denoised pseudo-representation. When multiple outputs are available, they are sparsely concatenated at the representation layer to increase information density. Training proceeds in two stages: contrastive alignment (source invariance) and supervised reinforcement (cyclic invariance). An optional training-free neighborhood search can refine local performance. Across 9 datasets covering user and system prompt scenarios, Inv2A outperforms baselines by an average of 4.77% BLEU score while reducing dependence on large inverse corpora. Our analysis further shows that prevalent defenses provide limited protection, underscoring the need for stronger strategies.

Predictive modeling on web-scale tabular data presents significant scalability challenges for industrial applications, often involving billions of instances and hundreds of heterogeneous numerical features. The inherent complexities of these features—characterized by anisotropy, heavy-tailed distributions, and non-stationarity—not only impose bottlenecks on the training efficiency and scalability of mainstream models like Gradient Boosting Decision Trees (GBDTs), but also compel practitioners into laborious, inefficient, and expert-dependent manual feature engineering. To systematically address this challenge, we introduce KMLP, a novel hybrid deep architecture. KMLP synergistically integrates a shallow Kolmogorov-Arnold Network (KAN) as a front-end with a Gated Multilayer Perceptron (gMLP) as the backbone. The KAN front-end leverages its learnable activation functions to automatically model complex non-linear transformations for each input feature in an end-to-end manner, thereby automating feature representation learning. Subsequently, the gMLP backbone efficiently captures high-order interactions among these refined representations. Extensive experiments on multiple public benchmarks and an ultra-large-scale industrial web dataset with billions of samples demonstrate that KMLP achieves state-of-the-art (SOTA) performance. Crucially, our findings reveal that KMLP's performance advantage over strong baselines like GBDTs becomes more pronounced as the data scale increases. This validates KMLP as a scalable and adaptive deep learning paradigm, offering a promising path forward for modeling large-scale, dynamic web tabular data.

In-context learning (ICL) has demonstrated significant potential in enhancing the capabilities of large language models (LLMs) during inference.It's well-established that ICL heavily relies on selecting effective demonstrations to generate outputs that better align with the expected results.As for demonstration selection, previous approaches have typically relied on intuitive metrics to evaluate the effectiveness of demonstrations, which often results in limited robustness and poor crossmodel generalization capabilities.To tackle these challenges, we propose a novel method, Demonstration Validation (D.Va), which integrates a demonstration validation perspective into this field.By introducing the demonstration validation mechanism, our method effectively identifies demonstrations that are both effective and highly generalizable.D.Va surpasses all existing demonstration selection techniques across both natural language understanding (NLU) and natural language generation (NLG) tasks.Additionally, we demonstrate the robustness and generalizability of our approach across various language models with different retrieval models.The code is available at https://github.com/Cheungki/dva.

Recent advances in spatial transcriptomics (ST) and cost reductions have enabled large-scale multi-slice ST data generation, enhancing the statistical power to detect subtle biological signals. However, cross-slice inconsistencies and data quality variability present significant analytical challenges. To overcome these limitations, we developed MUSE, a computational framework designed for multislice joint embedding, spatial domain identification, and gene expression imputation. Specifically, MUSE integrates a two-module architecture to ensure robust cross-slice alignment and data harmonization. The alignment module models each slice as a graph and employs optimal transport to align cells across slices while preserving spatial continuity. The optimization module further refines integration by incorporating an alignment loss, allowing lower-quality data to leverage structural information from higher-quality slices. Additionally, MUSE generates virtual neighbors from aligned cells, enriching contextual information and mitigating data sparsity. These design principles enable seamless integration with existing single-slice methods, extending their applicability to multi-slice ST analysis. To comprehensively evaluate its performance, we applied MUSE to 12 real and 48 simulated datasets spanning a range of data qualities. Across all metrics, MUSE consistently outperformed existing methods in cross-slice consistency, spatial domain identification, and gene expression imputation. To promote accessibility and adoption, we provide MUSE as an open-source software package. As multi-slice ST datasets become increasingly prevalent, MUSE provides a robust and extensible framework designed to effectively integrate growing numbers of slices, thereby advancing the analysis of tissue architectures and spatial gene expression in complex biological systems.

Tabular data, which accounts for over 80% of enterprise data assets, is vital in various fields. With growing concerns about privacy protection and data-sharing restrictions, generating high-quality synthetic tabular data has become essential. Recent advancements show that large language models (LLMs) can effectively generate realistic tabular data by leveraging semantic information and overcoming the challenges of high-dimensional data that arise from one-hot encoding. However, current methods do not fully utilize the rich information available in tables. To address this, we introduce AI Generative Table based on prompt enhancement, a novel approach that utilizes metadata information, such as table descriptions and schemas, as prompts to generate ultra-high-quality synthetic data. To overcome the token limit constraints of LLMs, we propose long-token partitioning algorithms that enable AIGT to model tables of any scale. AIGT achieves state-of-the-art performance on 14 out of 20 public datasets and two real industry datasets within the Alipay risk control system.

The recent large-scale text-to-image generative models have attained unprecedented performance, while people established *adaptor* modules like LoRA and DreamBooth to extend this performance to even more unseen concept tokens. However, we empirically find that this workflow often fails to accurately depict the *out-of-distribution* concepts. This failure is highly related to the low quality of training data. To resolve this, we present a framework called Controllable Adaptor Towards Out-of-Distribution Concepts (CATOD). Our framework follows the active learning paradigm which includes high-quality data accumulation and adaptor training, enabling a finer-grained enhancement of generative results. The *aesthetics* score and *concept-matching* score are two major factors that impact the quality of synthetic results. One key component of CATOD is the weighted scoring system that automatically balances between these two scores and we also offer comprehensive theoretical analysis for this point. Then, it determines how to select data and schedule the adaptor training based on this scoring system. The extensive results show that CATOD significantly outperforms the prior approaches with an 11.10 boost on the CLIP score and a 33.08% decrease on the CMMD metric.

Graph Neural Networks (GNNs) are powerful tools for graph representation learning. Despite their rapid development, GNNs also face some challenges, such as over-fitting, over-smoothing, and non-robustness. Previous works indicate that these problems can be alleviated by random dropping methods, which integrate augmented data into models by randomly masking parts of the input. However, some open problems of random dropping on GNNs remain to be solved. First, it is challenging to find a universal method that are suitable for all cases considering the divergence of different datasets and models. Second, augmented data introduced to GNNs causes the incomplete coverage of parameters and unstable training process. Third, there is no theoretical analysis on the effectiveness of random dropping methods on GNNs. In this paper, we propose a novel random dropping method called DropMessage, which performs dropping operations directly on the propagated messages during the message-passing process. More importantly, we find that DropMessage provides a unified framework for most existing random dropping methods, based on which we give theoretical analysis of their effectiveness. Furthermore, we elaborate the superiority of DropMessage: it stabilizes the training process by reducing sample variance; it keeps information diversity from the perspective of information theory, enabling it become a theoretical upper bound of other methods. To evaluate our proposed method, we conduct experiments that aims for multiple tasks on five public datasets and two industrial datasets with various backbone models. The experimental results show that DropMessage has the advantages of both effectiveness and generalization, and can significantly alleviate the problems mentioned above. A detailed version with full appendix can be found on arXiv: https://arxiv.org/abs/2204.10037.

Unsupervised domain adaptation (UDA) is a pivotal form in machine learning to extend the in-domain model to the distinctive target domains where the data distributions differ. Most prior works focus on capturing the inter-domain transferability but largely overlook rich intra-domain structures, which empirically results in even worse discriminability. In this work, we introduce a novel graph SPectral Alignment (SPA) framework to tackle the tradeoff. The core of our method is briefly condensed as follows: (i)-by casting the DA problem to graph primitives, SPA composes a coarse graph alignment mechanism with a novel spectral regularizer towards aligning the domain graphs in eigenspaces; (ii)-we further develop a fine-grained message propagation module --- upon a novel neighbor-aware self-training mechanism --- in order for enhanced discriminability in the target domain. On standardized benchmarks, the extensive experiments of SPA demonstrate that its performance has surpassed the existing cutting-edge DA methods. Coupled with dense model analysis, we conclude that our approach indeed possesses superior efficacy, robustness, discriminability, and transferability. Code and data are available at: https://github.com/CrownX/SPA.

Because contractions signal the approach of labor, pregnant women-especially primigravidas (i.e., women pregnant for the first time)-usually go to the hospital to seek medical intervention when they begin experiencing contractions, which is not conductive to good perinatal outcomes. Conventionally, uterine contraction monitoring requires specialized medical devices and relies on the doctor's clinical experience. Therefore, exploring an objective method to detect labor onset at home and avoid early hospital admission has essential importance. In this article, a labor progress monitoring system based on a sensing device, edge service, and Internet of things (IoT) platform is proposed, aiming to suggest suitable hospital admission times for low-risk primigravidas. The pregnant woman places the sensing device on her abdomen with the help of a belt to detect contraction activities. An intelligent edge service for contraction classification is deployed on a mobile phone. The system's artificial intelligence (AI)-assisted algorithm is lightweight, with 670 kB and 194 kB of memory dedicated to a convolutional neural network and long short-term memory, respectively. It classifies the pregnant woman as deferred admission, optional admission, or recommended admission according to different contraction states. An IoT platform connected to the hospital is implemented, providing professional suggestions from doctors. The test set collected in an emergency clinic shows that the proposed system can reach a classification accuracy of more than 96%. In conclusion, the proposed system enables remote labor progress monitoring at home and avoids early hospital admission.

Decoding and repair schemes are proposed for shift-exclusive-or (shift-XOR) product-matrix (PM) regenerating codes, which outperform the existing schemes in terms of both communication and computation costs. In particular, for the shift-XOR minimum bandwidth regenerating (MBR) codes, our decoding and repair schemes have the optimal transmission bandwidth and can be implemented in-place without extra storage space for intermediate XOR results. Technically, our schemes involve an in-place algorithm for solving a system of shift-XOR equations, called shift-XOR elimination, which does not have the bandwidth overhead generated by shift operations as in the previous zigzag algorithm and has lower computation complexities compared with the zigzag algorithm. The decoding and repair of shift-XOR MBR/MSR codes are decomposed into a sequence of systems of shift-XOR equations, and hence can be solved by a sequence of calls to the shift-XOR elimination. As the decompositions of the decoding and repair depend only on the PM construction, but not the specific shift and XOR operations, our decoding and repair schemes can be extended to other MBR/MSR codes using the PM construction. Due to its fundamental role, the shift-XOR elimination is of independent interest.

A cyclically symmetric entropy inequality is of the form hO≥chO′, where hOand hO′are two cyclic orbit entropy terms. A computational approach is formulated for bounding the extremal value of c̄, which is denoted by c̄O,O′. For two non-empty orbits O and O′ of a cyclic group, it is said that O dominates O′ if c̄O,O′= 1. Special attention is paid to characterizing such dominance relationship, and a graphical method is developed for that purpose.

In this paper, refined recovery and repair schemes are proposed for a storage system using the XOR-based MBR regenerating storage code proposed by Hou et al. Our schemes have zero transmission overhead for both recovery and repair, i.e., the total number of transmitted bits for repair/recovery is exactly equal to the total number of bits repaired/recovered. Further, our schemes use mainly XOR operations and have lower complexity than that of the previous schemes. Moreover, our schemes require only a small amount of auxiliary space, which qualifies our schemes as in-place.

This paper considers the error-detecting codes between two nodes. There are three transmissions between these two nodes: node A and node B. The first transmission can transmit one symbol within the alphabet {0, 1, ... , q1-1} from node A to node B; the second transmission can transmit one symbol within the alphabet {0, 1, ... , q2-1} from node B to node A; and the third transmission can transmit one symbol within the alphabet {0, 1, ... , q3-1} from node A to node B. One of the three transmissions may be erroneous. We prove that the maximum size of all 1-error-detecting codes over these three transmissions is min{q1, q2(q3-1)}.

This paper studies the allocation of information flows in noiseless, memoryless communication networks in the presence of omniscient Byzantine adversary. In such networks, adversary may maliciously modify some edge-flows, and legitimate users should resort to network error correction strategies to transmit data reliably. Unlike prior papers, which focused on the capacities of the networks, we consider the expense of resources used by the flow. Hereby, this paper uses an optimization problem to define the concept of minimum cost network error correction flows. We provide a necessary and sufficient condition of feasibility of the allocation problem, and derive a cut-set outer bound on the feasible region. Using this cut-set bound, we can find the minimum cost network error correction flow in some instances. Moreover, we also consider the relationship between incoming edge-flows and outgoing edge-flows of a vertex. As for the directed acyclic graphs, we propose an algorithm to allocate the network error correction flow. This algorithm is with polynomial time complexity, and proves to be optimal when recoding at intermediate nodes is forbidden. Additionally, in order to justify the necessity of recoding at intermediate nodes, we analyze the benefit of intermediate recoding. On the one hand, we construct a series of instances to prove that intermediate recoding can bring enormous benefits in some networks. On the other hand, numerical analysis shows that the benefit is modest in small random graphs.

In distributed multilevel diversity coding, K correlated sources (each with K components) are encoded in a distributed manner such that, given the outputs from any α encoders, the decoder can reconstruct the first α components of each of the corresponding α sources. For this problem, the optimality of a multilayer Slepian-Wolf coding scheme based on binning and superposition is established when K ≤ 3. The same conclusion is shown to hold for general K under a certain symmetry condition, which generalizes a celebrated result by Yeung and Zhang.

This paper considers interactive transmissions in the presence of omniscient Byzantine attacks. Unlike prior papers, it is assumed that the number of transmissions, the number of erroneous transmissions therein, and the direction of each transmission are predetermined. Besides, the size of the alphabet in each transmission is unequal and predefined. Using these transmissions, two nodes communicate interactively to send a message. In this model, both attack strategies and coding bounds are considered. Although the codebook can not fully describe the interactive code, we still assert the existence of successful attack strategies according to the relations between codewords in the codebook. Furthermore, to ensure that the code is able to detect or correct a given number of transmission errors, upper bounds on the size of code are derived. Finally, the tightness of the bounds is discussed.

This paper proposes a novel recovery scheme for the XOR-based storage codes with the increasing-difference property. For a message of kL bits stored in n storage nodes, a data collector connects any k out of the n storage nodes to recover the message. In our scheme, the data collector acquires exactly L bits for each node, so that no transmission overhead exists. Furthermore, we propose an in-place decoding algorithm that acquires less auxiliary space than the existing decoding algorithm, and the decoding computational complexity of our decoding algorithm is the same as the existing decoding algorithm.