Retrieval-Augmented Generation (RAG) has emerged as a widely adopted approach for knowledge injection during large language model (LLM) inference in recent years. However, due to their limited ability to exploit fine-grained inter-document relationships, current RAG implementations face challenges in effectively addressing the retrieved noise and redundancy content, which may cause error in the generation results. To address these limitations, we propose an Efficient Dynamic Clustering-based document Compression framework (EDC2-RAG) that utilizes latent inter-document relationships while simultaneously removing irrelevant information and redundant content. We validate our approach, built upon GPT-3.5-Turbo and GPT-4o-mini, on widely used knowledge-QA and Hallucination-Detection datasets. Experimental results show that our method achieves consistent performance improvements across various scenarios and experimental settings, demonstrating strong robustness and applicability. Our code and datasets are available at https://github.com/Tsinghua-dhy/EDC-2-RAG.

Visual hallucination refers to the perception of recognizable things that are not present. These phenomena are commonly linked to a range of neurological/psychiatric disorders. Despite ongoing research, the mechanisms through which the visual system generates hallucinations from real-world environments are still not well understood. Abnormal interactions between different regions of the brain responsible for perception are known to contribute to the occurrence of visual hallucinations. In this study, we propose and extend a generative neural network-based framework to address challenges within the visual system, aiming to create goal-driven models inspired by neurobiological mechanisms of visual hallucinations. We focus on the adversarial interactions between the visual system and the frontal lobe regions, proposing the Hallu-GAN model to suggest how these interactions can give rise to visual hallucinations. The architecture of the Hallu-GAN model is based on generative adversarial networks. Our simulation results indicate that disturbances in the ventral stream can lead to visual hallucinations. To further analyze the impact of other brain regions on the visual system, we extend the Hallu-GAN model by adding EEG data from individuals. This extended model, referred to as Hallu-GAN+, enables the examination of both hallucinating and non-hallucinating states. By training the Hallu-GAN+ model with EEG data from an individual with Charles Bonnet syndrome, we demonstrated its utility in analyzing the behavior of those experiencing hallucinations. Our simulation results confirmed the capability of the proposed model in resembling the visual system in both healthy and hallucinating states.

In this work, we demonstrate that a simple two-layer neural network with standard activation functions can learn an arbitrary word operation in any finite group, provided sufficient width is available and exhibits grokking while doing so. To explain the mechanism by which this is achieved, we reframe the problem as that of learning a particular $3$-tensor, which we show is typically of low rank. A key insight is that low-rank implementations of this tensor can be obtained by decomposing it along triplets of basic self-conjugate representations of the group and leveraging the fusion structure to rule out many components. Focusing on a phenomenologically similar but more tractable surrogate model, we show that the network is able to find such low-rank implementations (or approximations thereof), thereby using limited width to approximate the word-tensor in a generalizable way. In the case of the simple multiplication word, we further elucidate the form of these low-rank implementations, showing that the network effectively implements efficient matrix multiplication in the sense of Strassen. Our work also sheds light on the mechanism by which a network reaches such a solution under gradient descent.

Reinforcement learning with verifiable rewards (RL VR) has achieved remarkable success in enhancing the reasoning capabilities of large language models (LLMs). However, existing RL VR methods often suffer from exploration inefficiency due to mismatches between the training data's difficulty and the model's capability. LLMs fail to discover viable reasoning paths when problems are overly difficult, while learning little new capability when problems are too simple. Building on this analysis, we propose SEELE, a novel supervision-aided RL VR framework that dynamically adjusts problem difficulty to stay within the high-efficiency region. Unlike previous hint-based approaches, SEELE deliberately and adaptively adjusts the hint length for each problem to achieve an optimal difficulty. To determine the optimal hint length, SEELE employs a multi-round rollout sampling strategy. In each round, it fits an item response theory model to the accuracy-hint pairs collected in preceding rounds to predict the required hint length for the next round. Experimental results show that SEELE outperforms Group Relative Policy Optimization (GRPO) and Supervised Fine-tuning (SFT) by +11.8 and +10.5 points, respectively, and surpasses the best previous supervision-aided approach by +3.6 points on average across six math reasoning benchmarks.

We propose HyGLAD, a novel algorithm that automatically builds a set of interpretable patterns that model event data. These patterns can then be used to detect event-based anomalies in a stationary system, where any deviation from past behavior may indicate malicious activity. The algorithm infers equivalence classes of entities with similar behavior observed from the events, and then builds regular expressions that capture the values of those entities. As opposed to deep-learning approaches, the regular expressions are directly interpretable, which also translates to interpretable anomalies. We evaluate HyGLAD against all 7 unsupervised anomaly detection methods from DeepOD on five datasets from real-world systems. The experimental results show that on average HyGLAD outperforms existing deep-learning methods while being an order of magnitude more efficient in training and inference (single CPU vs GPU). Precision improved by 1.2x and recall by 1.3x compared to the second-best baseline.

Large Language Models (LLMs) as stochastic systems may generate numbers that deviate from available data, a failure known as \emph{numeric hallucination}. Existing safeguards -- retrieval-augmented generation, citations, and uncertainty estimation -- improve transparency but cannot guarantee fidelity: fabricated or misquoted values may still be displayed as if correct. We propose \textbf{Proof-Carrying Numbers (PCN)}, a presentation-layer protocol that enforces numeric fidelity through mechanical verification. Under PCN, numeric spans are emitted as \emph{claim-bound tokens} tied to structured claims, and a verifier checks each token under a declared policy (e.g., exact equality, rounding, aliases, or tolerance with qualifiers). Crucially, PCN places verification in the \emph{renderer}, not the model: only claim-checked numbers are marked as verified, and all others default to unverified. This separation prevents spoofing and guarantees fail-closed behavior. We formalize PCN and prove soundness, completeness under honest tokens, fail-closed behavior, and monotonicity under policy refinement. PCN is lightweight and model-agnostic, integrates seamlessly into existing applications, and can be extended with cryptographic commitments. By enforcing verification as a mandatory step before display, PCN establishes a simple contract for numerically sensitive settings: \emph{trust is earned only by proof}, while the absence of a mark communicates uncertainty.

Encoder-only languages models are frequently used for a variety of standard machine learning tasks, including classification and retrieval. However, there has been a lack of recent research for encoder models, especially with respect to multilingual models. We introduce mmBERT, an encoder-only language model pretrained on 3T tokens of multilingual text in over 1800 languages. To build mmBERT we introduce several novel elements, including an inverse mask ratio schedule and an inverse temperature sampling ratio. We add over 1700 low-resource languages to the data mix only during the decay phase, showing that it boosts performance dramatically and maximizes the gains from the relatively small amount of training data. Despite only including these low-resource languages in the short decay phase we achieve similar classification performance to models like OpenAI's o3 and Google's Gemini 2.5 Pro. Overall, we show that mmBERT significantly outperforms the previous generation of models on classification and retrieval tasks -- on both high and low-resource languages.

Medical image segmentation is a critical task in clinical workflows, particularly for the detection and delineation of pathological regions. While convolutional architectures like U-Net have become standard for such tasks, their limited receptive field restricts global context modeling. Recent efforts integrating transformers have addressed this, but often result in deep, computationally expensive models unsuitable for real-time use. In this work, we present a novel end-to-end lightweight architecture designed specifically for real-time binary medical image segmentation. Our model combines a Swin Transformer-like encoder with a U-Net-like decoder, connected via skip pathways to preserve spatial detail while capturing contextual information. Unlike existing designs such as Swin Transformer or U-Net, our architecture is significantly shallower and competitively efficient. To improve the encoder's ability to learn meaningful features without relying on large amounts of labeled data, we first train it using Barlow Twins, a self-supervised learning method that helps the model focus on important patterns by reducing unnecessary repetition in the learned features. After this pretraining, we fine-tune the entire model for our specific task. Experiments on benchmark binary segmentation tasks demonstrate that our model achieves competitive accuracy with substantially reduced parameter count and faster inference, positioning it as a practical alternative for deployment in real-time and resource-limited clinical environments. The code for our method is available at Github repository: https://github.com/mkianih/Barlow-Swin.

Simulating the dynamics of open quantum systems with spatial structure and external control is an important challenge in quantum information science. Classical numerical solvers for such systems require integrating coupled master and field equations, which is computationally demanding for simulation and optimization tasks and often precluding real-time use in network-scale simulations or feedback control. We introduce a regional attention-based neural architecture that learns the spatiotemporal dynamics of structured open quantum systems. The model incorporates translational invariance of physical laws as an inductive bias to achieve scalable complexity, and supports conditioning on time-dependent global control parameters. We demonstrate learning on two representative systems: a driven dissipative single qubit and an electromagnetically induced transparency (EIT) quantum memory. The model achieves high predictive fidelity under both in-distribution and out-of-distribution control protocols, and provides substantial acceleration up to three orders of magnitude over numerical solvers. These results demonstrate that the architecture establishes a general surrogate modeling framework for spatially structured open quantum dynamics, with immediate relevance to large-scale quantum network simulation, quantum repeater and protocol design, real-time experimental optimization, and scalable device modeling across diverse light-matter platforms.

Test-time scaling increases inference-time computation by allowing models to generate long reasoning chains, and has shown strong performance across many domains. However, in this work, we show that this approach is not yet effective for knowledge-intensive tasks, where high factual accuracy and low hallucination rates are essential. We conduct a comprehensive evaluation of test-time scaling using 12 reasoning models on two knowledge-intensive benchmarks. Our results reveal that increasing test-time computation does not consistently improve accuracy and, in many cases, it even leads to more hallucinations. We then analyze how extended reasoning affects hallucination behavior. We find that reduced hallucinations often result from the model choosing to abstain after thinking more, rather than from improved factual recall. Conversely, for some models, longer reasoning encourages attempts on previously unanswered questions, many of which result in hallucinations. Case studies show that extended reasoning can induce confirmation bias, leading to overconfident hallucinations. Despite these limitations, we observe that compared to non-thinking, enabling thinking remains beneficial. Code and data are available at https://github.com/XuZhao0/tts-knowledge