Physical links are fundamental to the structure, function, and ultimately the understanding of many networked systems. Erroneously identified, such direct cause-effect relations undermine the understanding of the system’s dynamics and can impede effective intervention. To address the long-standing challenge of distinguishing direct from indirect links in observational data from complex dynamical networks, we introduce a general inference framework tailored to networked dynamical systems. Our approach combines cross mapping with a novel, physics-informed technique for incremental variable construction that leverages temporal delay effects to identify relevant variables and eliminate indirect links. We validate our method across a diverse set of noisy benchmark and real-world systems, including brain networks, gene regulation, synthetic gene circuits, and biosphere-atmosphere interactions. The method achieves particularly strong gains on mesoscale networks (n≈5–20), consistently outperforming existing approaches under diverse noise conditions. Although designed for small- to medium-sized systems, it continues to outperform competing methods even in larger networks, demonstrating both robustness and broad applicability.

Most multi-agent systems rely exclusively on autoregressive language models (ARMs) that are based on sequential generation. Although effective for fluent text, ARMs limit global reasoning and plan revision. On the other hand, Discrete Diffusion Language Models (DDLMs) enable non-sequential, globally revisable generation and have shown strong planning capabilities, but their limited text fluency hinders direct collaboration with ARMs. We introduce Latent-DARM, a latent-space communication framework bridging DDLM (planners) and ARM (executors), maximizing collaborative benefits. Across mathematical, scientific, and commonsense reasoning benchmarks, Latent-DARM outperforms text-based interfaces on average, improving accuracy from 27.0

Recent reinforcement learning (RL) methods have substantially enhanced the planning capabilities of Large Language Models (LLMs), yet the theoretical basis for their effectiveness remains elusive. In this work, we investigate RL's benefits and limitations through a tractable graph-based abstraction, focusing on policy gradient (PG) and Q-learning methods. Our theoretical analyses reveal that supervised fine-tuning (SFT) may introduce co-occurrence-based spurious solutions, whereas RL achieves correct planning primarily through exploration, underscoring exploration’s role in enabling better generalization. However, we also show that PG suffers from diversity collapse, where output diversity decreases during training and persists even after perfect accuracy is attained. By contrast, Q-learning provides two key advantages: off-policy learning and diversity preservation at convergence. We further demonstrate that careful reward design is necessary to prevent Q-value bias in Q-learning. Finally, applying our framework to the real-world planning benchmark Blocksworld, we confirm that these behaviors manifest in practice.

Beyond training-time optimization, scaling test-time computation has emerged as a key paradigm to extend the reasoning capabilities of Large Language Models (LLMs). However, most existing methods adopt a rigid Planning-before-Trial (PbT) policy, which inefficiently allocates test-time compute by incurring planning overhead even on directly solvable problems. We propose Planning-after-Trial (PaT), an adaptive policy for code generation that invokes a planner only upon verification failure. This adaptive policy naturally enables a heterogeneous model configuration: a cost-efficient model handles generation attempts, while a powerful model is reserved for targeted planning interventions. Empirically, across multiple benchmarks and model families, our approach significantly advances the cost-performance Pareto frontier. Notably, our heterogeneous configuration achieves performance comparable to a large homogeneous model while reducing inference cost by approximately 69%.

The multi-layered network exploration problem MuLaNE is a significant challenge derived from various practical applications. In MuLaNE, there are multiple layers of the network, where each node is assigned an importance weight, and each layer is explored through random walks. The objective is to allocate a total random walk budget B across the network layers to maximize the cumulative weights of the unique nodes visited. We systematically approach this problem by addressing both offline optimization and online learning scenarios. For the offline optimization case, where the network structure and node weights are known, we propose constant-ratio approximation algorithms based on greedy strategies for overlapping networks, and exact optimal solutions using greedy or dynamic programming for non-overlapping networks. In the online learning scenario, where neither the network structure nor the node weights are initially known, we adapt the combinatorial multi-armed bandit framework. We develop algorithms with two distinct confidence radius designs to simultaneously learn random walk parameters and node weights, while optimizing budget allocation across multiple rounds, achieving logarithmic regret bounds. Finally, experimental results on real-world social and computer networks validate the practical applicability of MuLaNE and our theoretical findings.

The congestion game is a powerful model that encompasses a range of engineering systems such as traffic networks and resource allocation. It describes the behavior of a group of agents who share a common set of F facilities and take actions as subsets of k facilities. In this work, we study the online formulation of congestion games, where agents participate in the game repeatedly and observe feedback with randomness. We note that this paper is the result of the merging of [22] arXiv:2306.13673 and [17] arXiv:2401.09628 . In [22], we propose CongestEXP, a decentralized algorithm that is based on the classic exponential weights method. By maintaining weights on the facility level, the regret bound of CongestEXP avoids the exponential dependence on the size of possible facility sets, i.e., ( [ F; k ]) ≈ F^k , and scales only linearly with F. Specifically, we show that CongestEXP attains a regret upper bound of O(kF√(T)) for every individual player, where T is the time horizon. If a strict Nash equilibrium exists, we show that CongestEXP can converge to the strict Nash policy almost exponentially fast in O(Fexp (-t^1-α)) , where t is the number of iterations and α∈ (1/2, 1) . In [17], we present an online learning algorithm in the bandit feedback model that, once adopted by all agents of a congestion game, results in game-dynamics that converge to an ϵ -approximate Nash Equilibrium in a polynomial number of rounds with respect to 1/ϵ , the number of players and the number of available resources. The proposed algorithm also guarantees sublinear regret to any agent adopting it. As a result, our work answers an open question from [16] and extends the recent results of [34] to the bandit feedback model.

Mixture-of-Experts (MoE) language models route each token to a small subset of experts, but whether the routes selected by a trained top-k router are good ones is rarely evaluated directly. Holding the model fixed, we compare each standard route against sampled equal-compute alternatives for the same token and score each by the next-token probability it assigns to the realized token in a verified reasoning trajectory. The result is sharply token-conditional: the standard router is well-aligned with route utility on confident tokens but uninformative on the fragile tokens that drive hard reasoning, where lower-loss equal-compute routes consistently exist inside the frozen model but are not selected. The same pattern holds across Qwen3-30B-A3B, GPT-OSS-20B, DeepSeek-V2-Lite, and OLMoE-1B-7B, and follows structurally from how standard top-k training evaluates routing decisions: the language modeling loss scores only the executed route, and load balancing depends only on aggregate routing statistics. A minimal router-only update to the final-layer router, leaving every expert and every other router frozen, is sufficient to shift pass@K on AIME 2024+2025 and HMMT 2025 for both Qwen3-30B-A3B and GPT-OSS-20B, suggesting that at least part of the failure reflects router-reachable misallocation rather than expert capacity alone.

Offline-to-online learning aims to improve online decision-making by leveraging offline logged data. A central challenge in this setting is the distribution shift between offline and online environments. While some existing works attempt to leverage shifted offline data, they largely rely on UCB-type algorithms. Thompson sampling (TS) represents another canonical class of bandit algorithms, well known for its strong empirical performance and naturally suited to offline-to-online learning through its Bayesian formulation. However, unlike UCB indices, posterior samples in TS are not guaranteed to be optimistic with respect to the true arm means. This makes indices constructed from purely online and hybrid data difficult to compare and complicates their use. To address this issue, we propose sample-mean anchored TS (Anchor-TS), which introduces a novel median-based anchoring rule that defines the arm index as the median of an online posterior sample, a hybrid posterior sample, and the online sample mean. The median anchoring systematically corrects bias induced by distribution shift by mitigating over-estimation for suboptimal arms and under-estimation for optimal arms, while exploiting offline information to obtain more accurate estimates when the shift is small. We establish theoretical guarantees showing that the proposed algorithm safely leverages offline data to accelerate online learning, and quantifying how the degree of distribution shift and the size of offline data affect the resulting regret reduction. Extensive experiments demonstrate consistent improvements of our algorithm over baselines.

Whether Large Language Models (LLMs) develop coherent internal world models remains a core debate. While conventional Next-Token Prediction (NTP) focuses on one-step-ahead supervision, Multi-Token Prediction (MTP) has shown promise in learning more structured representations. In this work, we provide a theoretical perspective analyzing the gradient inductive bias of MTP, supported by empirical evidence, showing that MTP promotes the convergence toward internal belief states by inducing representational contractivity via gradient coupling. However, we reveal that standard MTP often suffers from structural hallucinations, where discrete token supervision encourages illegal shortcuts in latent space that violate environmental constraints. To address this, we propose a novel method Latent Semantic Enhancement MTP (LSE-MTP), which anchors predictions to ground-truth hidden state trajectories. Experiments on synthetic graphs and real-world Manhattan Taxi Ride show that LSE-MTP effectively bridges the gap between discrete tokens and continuous state representations, enhancing representation alignment, reducing structural hallucinations, and improving robustness to perturbations.

Temporal reasoning is a fundamental capability for large language models (LLMs) to understand real-world dynamics. Existing research on temporal reasoning has predominantly focused on the Gregorian calendar. However, as many countries and regions concurrently adopt multiple calendar systems, temporal reasoning across calendars becomes crucial for LLMs in global and multicultural contexts. Unfortunately, cross-calendar temporal reasoning remains underexplored, with no dedicated benchmark available to evaluate this capability. To bridge this gap, we introduce SPAN, a cross-calendar temporal reasoning benchmark, which requires LLMs to perform intra-calendar temporal reasoning and inter-calendar temporal conversion. SPAN features ten cross-calendar temporal reasoning directions, two reasoning types, and two question formats across six calendars. To enable time-variant and contamination-free evaluation, we propose a template-driven protocol for dynamic instance generation that enables assessment on a user-specified Gregorian date. We conduct extensive experiments on both open- and closed-source state-of-the-art (SOTA) LLMs over a range of dates spanning 100 years from 1960 to 2060. Our evaluations show that these LLMs achieve an average accuracy of only 34.5%, with none exceeding 80%, indicating that this task remains challenging. Through in-depth analysis of reasoning types, question formats, and temporal reasoning directions, we identify two key obstacles for LLMs: Future-Date Degradation and Calendar Asymmetry Bias. To strengthen LLMs' cross-calendar temporal reasoning capability, we further develop an LLM-powered Time Agent that leverages tool-augmented code generation. Empirical results show that Time Agent achieves an average accuracy of 95.31%, outperforming several competitive baselines, highlighting the potential of tool-augmented code generation to advance cross-calendar temporal reasoning. We hope this work will inspire further efforts toward more temporally and culturally adaptive LLMs.

In this article, we introduce the Time-Decaying Independent Cascade (TDIC) model—a novel extension of the classical Independent Cascade model that integrates temporal dynamics inspired by Ebbinghaus’ forgetting curve. By incorporating a time decay factor, our model captures the natural decline in the probability of information spread over time. Uniquely, we are the first to combine user-friendship network and item-sharing bipartite graph to address the influence maximization problem, offering a dual perspective that enriches our understanding of information diffusion. We formalize the influence maximization with time decay (IMTD) problem and propose an effective seed selection strategy that leverages these temporal dynamics. Both our theoretical analysis and experimental evaluations on real-world datasets demonstrate that the TDIC model not only replicates key network properties, such as degree distributions, but also significantly enhances influence spread compared to static models. Our findings underscore the importance of incorporating temporal factors and multinetwork insights to develop more realistic and effective approaches for influence propagation in social networks.

The Rising Multi-Armed Bandit (RMAB) framework models environments where expected rewards of arms increase with plays, which models practical scenarios where performance of each option improves with the repeated usage, such as in robotics and hyperparameter tuning. For instance, in hyperparameter tuning, the validation accuracy of a model configuration (arm) typically increases with each training epoch. A defining characteristic of RMAB is em horizon-dependent optimality: unlike standard settings, the optimal strategy here shifts dramatically depending on the available budget T. This implies that knowledge of T yields significantly greater utility in RMAB, empowering the learner to align its decision-making with this shifting optimality. However, the horizon-aware setting remains underexplored. To address this, we propose a novel CUmulative Reward Estimation UCB (CURE-UCB) that explicitly integrates the horizon. We provide a rigorous analysis establishing a new regret upper bound and prove that our method strictly outperforms horizon-agnostic strategies in structured environments like “linear-then-flat” instances. Extensive experiments demonstrate its significant superiority over baselines.

This paper examines how political networks influence insider trading in China. Using biographical data to construct chairman-politician social networks, we find that firms with stronger political networks engage in significantly less insider trading. The effect is stronger for non-stateowned enterprises (non-SOEs) and for long-standing or high-ranking connections. The muted trading persists during periods when insiders possess valuable private information, including prior to M&A announcements and major policy events. The evidence suggests that personal political networks function as informal governance mechanisms that discipline managerial opportunism when formal governance through state ownership is absent.

Large Language Models (LLMs) are revolutionizing how users interact with information systems, yet their high inference cost poses serious scalability and sustainability challenges. Caching inference responses, allowing them to be retrieved without another forward pass through the LLM, has emerged as one possible solution. Traditional exact-match caching, however, overlooks the semantic similarity between queries, leading to unnecessary recomputation. Semantic caching addresses this by retrieving responses based on semantic similarity, but introduces a fundamentally different cache eviction problem: one must account for mismatch costs between incoming queries and cached responses. Moreover, key system parameters, such as query arrival probabilities and serving costs, are often unknown and must be learned over time. Existing semantic caching methods are largely ad-hoc, lacking theoretical foundations and unable to adapt to real-world uncertainty. In this paper, we present a principled, learning-based framework for semantic cache eviction under unknown query and cost distributions. We formulate both offline optimization and online learning variants of the problem, and develop provably efficient algorithms with state-of-the-art guarantees. We also evaluate our framework on a synthetic dataset, showing that our proposed algorithms perform matching or superior performance compared with baselines.

Combinatorial online learning is a fundamental task for selecting the optimal action (or super arm) as a combination of base arms in sequential interactions with systems providing stochastic rewards. It is applicable to diverse domains such as robotics, social advertising, network routing, and recommendation systems. In many real-world scenarios, we often encounter rising rewards, where playing a base arm not only provides an instantaneous reward but also contributes to the enhancement of future rewards, e.g., robots improving through practice and social influence strengthening in the history of successful recommendations. Crucially, these enhancements may propagate to multiple super arms that share the same base arms, introducing dependencies beyond the scope of existing bandit models. To address this gap, we introduce the Combinatorial Rising Bandit (CRB) framework and propose a provably efficient and empirically effective algorithm, Combinatorial Rising Upper Confidence Bound (CRUCB). We empirically demonstrate the effectiveness of CRUCB in realistic deep reinforcement learning environments and synthetic settings, while our theoretical analysis establishes tight regret bounds. Together, they underscore the practical impact and theoretical rigor of our approach.

As Large Language Models (LLMs) become increasingly popular, caching responses so that they can be reused by users with semantically similar queries has become a vital strategy for reducing inference costs and latency. Existing caching frameworks have proposed to decide which query responses to cache by assuming a finite, known universe of discrete queries and learning their serving costs and arrival probabilities. As LLMs' pool of users and queries expands, however, such an assumption becomes increasingly untenable: real-world LLM queries reside in an infinite, continuous embedding space. In this paper, we establish the first rigorous theoretical framework for semantic LLM response caching in continuous query space under uncertainty. To bridge the gap between discrete optimization and continuous representation spaces, we introduce dynamic ε-net discretization coupled with Kernel Ridge Regression. This design enables the system to formally quantify estimation uncertainty and generalize partial feedback on LLM query costs across continuous semantic query neighborhoods. We develop both offline learning and online adaptive algorithms optimized to reduce switching costs incurred by changing the cached responses. We prove that our online algorithm achieves a sublinear regret bound against an optimal continuous oracle, which reduces to existing bounds for discrete query models. Extensive empirical evaluations demonstrate that our framework approximates the continuous optimal cache well while also reducing computational and switching overhead compared to existing methods.

Current autoregressive language models (ARMs) achieve high accuracy but require long token sequences, making them costly. Discrete diffusion language models (DDLMs) enable parallel and flexible generation within a fixed number of steps and have recently emerged for their strong performance in complex reasoning and long-term planning tasks. We present a study exploring hybrid architectures that couple DDLMs with ARMs to assess whether their collaboration can yield complementary benefits. We first examine collaboration in text space, where one model plans the reasoning process and another executes the final answer based on that plan. We then extend this setup to latent-space communication, introducing a learned projector that maps DDLM latents into the ARM's embedding space, potentially bypassing some of the text-generation limitations of diffusion models. We find that shifting DDLM –> ARM communication from text space to latent space yields significant accuracy gains, for example increasing from 27.0

>抑郁症（major depressive disorder, MDD）是最为常见的人类精神疾病之一 [1] ,在全球人类疾病负担排名中位居第二 [2] .由于其发病原因和致病机制目前尚不清楚,目前对抑郁症的治疗存在缺陷 [3] .最近,东南大学附属中大医院和中国科学院深圳先进技术研究院/深圳理工大学张志珺教授领导的科研团队发表在Science Bulletin的成果发现一种抑郁症相关环状RNA circ FKBP8（5S,6）,

Fair Influence Maximization (FIM) seeks to mitigate disparities in influence across different groups and has recently garnered increasing attention. A widely adopted notion of fairness in FIM is the maximin constraint, which directly requires maximizing the utility (influenced ratio within a group) of the worst-off group. Despite its intuitive formulation, designing efficient algorithms with strong theoretical guarantees remains challenging, as the maximin objective does not satisfy submodularity, a key property for designing approximate algorithms in traditional influence maximization settings. In this paper, we address this challenge by proposing a two-step optimization framework consisting of Inner-group Maximization (IGM) and Across-group Maximization (AGM). We first prove that the influence spread within any individual group remains submodular, enabling effective optimization within groups. Based on this, IGM applies a greedy approach to pick high-quality seeds for each group. In the second step, AGM coordinates seed selection across groups by introducing two strategies: Uniform Selection (US) and Greedy Selection (GS). We prove that AGM-GS holds a (1-1/e-ε) approximation to the optimal solution when groups are completely disconnected, while AGM-US guarantees a roughly 1/m(1-1/e-ε) lower bound regardless of the group structure, with m denoting the number of groups.