Despite neural ordinary differential equations (Neural ODEs) exhibiting intrinsic robustness under input perturbations due to their dynamical systems nature, recent approaches often involve imposing Lyapunov-based stability conditions to provide formal robustness guarantees. However, a fundamental challenge remains: the tension between robustness and accuracy, primarily stemming from the difficulty in imposing appropriate stability conditions. To address this, we propose an adaptive stable learning framework named Zubov-Net, which innovatively reformulates Zubov's equation into a consistency characterization between regions of attraction (RoAs) and prescribed RoAs (PRoAs). Building on this consistency, we introduce a new paradigm for actively controlling the geometry of RoAs by directly optimizing PRoAs to reconcile accuracy and robustness. Our approach is realized through tripartite losses (consistency, classification, and separation losses) and a parallel boundary sampling algorithm that co-optimizes the Neural ODE and the Lyapunov function. To enhance the discriminativity of Lyapunov functions, we design an input-attention-based convex neural network via a softmax attention mechanism that focuses on equilibrium-relevant features and also serves as weight normalization to maintain training stability in deep architectures. Theoretically, we prove that minimizing the tripartite loss guarantees consistent alignment of PRoAs-RoAs, trajectory stability, and non-overlapping PRoAs. Moreover, we establish stochastic convex separability with tighter probability bounds and fewer dimensionality requirements to justify the convex design in Lyapunov functions. Experimentally, Zubov-Net maintains high classification accuracy while significantly improving robustness against various stochastic noises and adversarial attacks.

We propose ILP-CoT, a method that bridges Inductive Logic Programming (ILP) and Multimodal Large Language Models (MLLMs) for abductive logical rule induction. The task involves both discovering logical facts and inducing logical rules from a small number of unstructured textual or visual inputs, which still remain challenging when solely relying on ILP, due to the requirement of specified background knowledge and high computational cost, or MLLMs, due to the appearance of perceptual hallucinations. Based on the key observation that MLLMs could propose structure-correct rules even under hallucinations, our approach automatically builds ILP tasks with pruned search spaces based on the rule structure proposals from MLLMs, and utilizes ILP system to output rules built upon rectified logical facts and formal inductive reasoning. Its effectiveness is verified through challenging logical induction benchmarks, as well as a potential application of our approach, namely text-to-image customized generation with rule induction. Our code and data are released at https://github.com/future-item/ILP-CoT.

Multimodal large language models (MLLMs) are well suited to image aesthetic assessment, as they can capture high-level aesthetic features leveraging their cross-modal understanding capacity. However, the scarcity of multimodal aesthetic reasoning data and the inherently subjective nature of aesthetic judgment make it difficult for MLLMs to generate accurate aesthetic judgments with interpretable rationales. To this end, we propose Aes-R1, a comprehensive aesthetic reasoning framework with reinforcement learning (RL). Concretely, Aes-R1 integrates a pipeline, AesCoT, to construct and filter high-quality chain-of-thought aesthetic reasoning data used for cold-start. After teaching the model to generate structured explanations prior to scoring, we then employ the Relative-Absolute Policy Optimization (RAPO), a novel RL algorithm that jointly optimizes absolute score regression and relative ranking order, improving both per-image accuracy and cross-image preference judgments. Aes-R1 enables MLLMs to generate grounded explanations alongside faithful scores, thereby enhancing aesthetic scoring and reasoning in a unified framework. Extensive experiments demonstrate that Aes-R1 improves the backbone's average PLCC/SRCC by 47.9%/34.8%, surpassing state-of-the-art baselines of similar size. More ablation studies validate Aes-R1's robust generalization under limited supervision and in out-of-distribution scenarios.

Low-Rank Adaptation (LoRA) is a widely adopted parameter-efficient fine-tuning method for large language models. However, its linear nature limits expressiveness. We propose LoRAN, a non-linear extension of LoRA that applies lightweight transformations to the low-rank updates. We further introduce Sinter, a sine-based activation that adds structured perturbations without increasing parameter count. Experiments across summarization and classification tasks show that LoRAN consistently improves over QLoRA. Ablation studies reveal that Sinter outperforms standard activations such as Sigmoid, ReLU, and Tanh, highlighting the importance of activation design in lowrank tuning.

Retrieval-Augmented Generation (RAG) is a framework for grounding Large Language Models (LLMs) in external, up-to-date information. However, recent advancements in context window size allow LLMs to process inputs of up to 128K tokens or more, offering an alternative strategy: supplying the full document context directly to the model, rather than relying on RAG to retrieve a subset of contexts. Nevertheless, this emerging alternative strategy has notable limitations: (i) it is token-inefficient to handle large and potentially redundant contexts; (ii) it exacerbates the `lost in the middle' phenomenon; and (iii) under limited model capacity, it amplifies distraction, ultimately degrading LLM output quality. In this paper, we propose LDAR (Learning Distraction-Aware Retrieval), an adaptive retriever that learns to retrieve contexts in a way that mitigates interference from distracting passages, thereby achieving significantly higher performance with reduced token usage compared to long-context approaches. Extensive experiments across diverse LLM architectures and six knowledge-intensive benchmarks demonstrate the effectiveness and robustness of our approach, highlighting the importance of balancing the trade-off between information coverage and distraction.

Multi-Turn Long-Form Question Answering (MT-LFQA) is a key application paradigm of Large Language Models (LLMs) in knowledge-intensive domains. However, existing benchmarks are limited to single-turn dialogue, while multi-turn dialogue benchmarks typically assess other orthogonal capabilities rather than knowledge-intensive factuality. To bridge this critical gap, we introduce \textbf{KnowMT-Bench}, the \textit{first-ever} benchmark designed to systematically evaluate MT-LFQA for LLMs across knowledge-intensive fields, including medicine, finance, and law. To faithfully assess the model's real-world performance, KnowMT-Bench employs a dynamic evaluation setting where models generate their own multi-turn dialogue histories given logically progressive question sequences. The factual capability and information delivery efficiency of the \textit{final-turn} answer are then evaluated using a human-validated automated pipeline. Our experiments reveal that multi-turn contexts degrade performance: factual capability declines due to the contextual noise from self-generated histories, while information efficiency drops as models become more verbose with increasing dialogue length. We then investigate mitigation strategies, demonstrating that retrieval-augmented generation (RAG) can effectively alleviate and even reverse this factual degradation. These findings underscore the importance of our benchmark in evaluating and enhancing the conversational factual capabilities of LLMs in real-world knowledge-intensive applications. Code is available at \href{https://github.com/hardenyu21/KnowMT-Bench}{\textcolor{cyan}{\texttt{KnowMT-Bench}}}.

As a model-agnostic approach to long context modeling, multi-agent systems can process inputs longer than a large language model's context window without retraining or architectural modifications. However, their performance often heavily relies on hand-crafted multi-agent collaboration strategies and prompt engineering, which limit generalizability. In this work, we introduce a principled framework that formalizes the model-agnostic long context modeling problem as a compression problem, yielding an information-theoretic compression objective. Building on this framework, we propose Graph of Agents (GoA), which dynamically constructs an input-dependent collaboration structure that maximizes this objective. For Llama 3.1 8B and Qwen3 8B across six document question answering benchmarks, GoA improves the average $F_1$ score of retrieval-augmented generation by 5.7\% and a strong multi-agent baseline using a fixed collaboration structure by 16.35\%, respectively. Even with only a 2K context window, GoA surpasses the 128K context window Llama 3.1 8B on LongBench, showing a dramatic increase in effective context length. Our source code is available at https://github.com/tjoo512/graph-of-agents.

Model integrity of Large language models (LLMs) has become a pressing security concern with their massive online deployment. Prior Bit-Flip Attacks (BFAs)--a class of popular AI weight memory fault-injection techniques--can severely compromise Deep Neural Networks (DNNs): as few as tens of bit flips can degrade accuracy toward random guessing. Recent studies extend BFAs to LLMs and reveal that, despite the intuition of better robustness from modularity and redundancy, only a handful of adversarial bit flips can also cause LLMs' catastrophic accuracy degradation. However, existing BFA methods typically focus on either integer or floating-point models separately, limiting attack flexibility. Moreover, in floating-point models, random bit flips often cause perturbed parameters to extreme values (e.g., flipping in exponent bit), making it not stealthy and leading to numerical runtime error (e.g., invalid tensor values (NaN/Inf)). In this work, for the first time, we propose SBFA (Sneaky Bit-Flip Attack), which collapses LLM performance with only one single bit flip while keeping perturbed values within benign layer-wise weight distribution. It is achieved through iterative searching and ranking through our defined parameter sensitivity metric, ImpactScore, which combines gradient sensitivity and perturbation range constrained by the benign layer-wise weight distribution. A novel lightweight SKIP searching algorithm is also proposed to greatly reduce searching complexity, which leads to successful SBFA searching taking only tens of minutes for SOT A LLMs. Across Qwen, LLaMA, and Gemma models, with only one single bit flip, SBFA successfully degrades accuracy to below random levels on MMLU and SST -2 in both BF16 and INT8 data formats. Remarkably, flipping a single bit out of billions of parameters reveals a severe security concern of SOT A LLM models.

We study the problem of online multi-agent reinforcement learning (MARL) in environments with sparse rewards, where reward feedback is not provided at each interaction but only revealed at the end of a trajectory. This setting, though realistic, presents a fundamental challenge: the lack of intermediate rewards hinders standard MARL algorithms from effectively guiding policy learning. To address this issue, we propose a novel framework that integrates online inverse preference learning with multi-agent on-policy optimization into a unified architecture. At its core, our approach introduces an implicit multi-agent reward learning model, built upon a preference-based value-decomposition network, which produces both global and local reward signals. These signals are further used to construct dual advantage streams, enabling differentiated learning targets for the centralized critic and decentralized actors. In addition, we demonstrate how large language models (LLMs) can be leveraged to provide preference labels that enhance the quality of the learned reward model. Empirical evaluations on state-of-the-art benchmarks, including MAMuJoCo and SMACv2, show that our method achieves superior performance compared to existing baselines, highlighting its effectiveness in addressing sparse-reward challenges in online MARL.

Large language models (LLMs) transcend passive generation and act as goal-directed agents by invoking external tools. Reinforcement learning (RL) offers a principled framework for optimizing these emergent tool-use policies, yet the prevailing paradigm relies exclusively on sparse outcome rewards and lacks consideration of the particularity of tool-use tasks, inflating policy-gradient variance and resulting in inefficient training. To better understand and address these challenges, we first establish a theoretical link between policy entropy and training stability of tool-use tasks, which reveals that structured, low-entropy tokens are primary determinants of rewards. Motivated by this insight, we propose \textbf{Res}haped \textbf{T}oken-level policy gradients (\textbf{ResT}) for tool-use tasks. ResT reshapes the policy gradient through entropy-informed token reweighting, progressively upweighting reasoning tokens as training proceeds. This entropy-aware scheme enables a smooth shift from structural correctness to semantic reasoning and stabilizes convergence in multi-turn tool-use tasks. Evaluation on BFCL and API-Bank shows that ResT achieves state-of-the-art results, outperforming prior methods by up to $8.76\%$. When fine-tuned on a 4B base LLM, ResT further surpasses GPT-4o by $4.11\%$ on single-turn tasks and $1.50\%$ on multi-turn base tasks.