Papers

Generative models have found great success as data-driven methods of solving inverse problems. Two popular approaches work either by combining a pretrained generative prior with a known degradation model, or by training a conditional generative model directly from paired data. We target a setting that spans both regimes: unknown degradations can be learned from few paired examples, while known degradations can be learned from self-generated samples. We introduce Likelihood Score Approximation (LSA), a generative framework that keeps a pretrained unconditional model fixed and learns an observation-conditioned model that approximates the likelihood score from paired samples. Within a conditional stochastic-interpolant framework, LSA can be trained in either score or velocity coordinates, independently of the unconditional model's native parameterization, and supports both deterministic and stochastic sampling. We further show empirically that the prior model can be swapped post-training while keeping the same LSA model. Across speech and image inverse problems, LSA operates effectively even at roughly 0.01% of the full training dataset. On the ImageNet-256 benchmark it achieves competitive or better restoration quality than strong posterior-sampling baselines while requiring up to several orders of magnitude fewer network evaluations.

The logarithmic transform is essential in audio signal processing since human auditory perception is approximately logarithmic with respect to magnitude. However, directly incorporating prior knowledge about signals (e.g., harmonic structure) in the log-magnitude domain into optimization problems solved by standard proximal splitting algorithms remains challenging. To address this issue, this paper proposes a novel regularizer termed EPILOG (Exponential Penalty for Imposing priors on LOG-magnitude). EPILOG indirectly imposes prior knowledge on the log-magnitude of a complex-valued signal through regularization of an auxiliary variable that is shown to be linked with the log-magnitude. Furthermore, we derive its variable-wise proximity operators and develop a proximal splitting algorithm using these operators. Experiments on speech dereverberation demonstrate the effectiveness of the proposed regularizer, particularly in promoting cepstral-domain sparsity.

Language-model agents exchange messages to combine evidence, but their communication can also create echo chambers that reinforce shared errors. However, overall task performance does not explain how a message changes the receiving agent's internal activations and affects its decision. In this work, we introduce Social Circuits, a framework for tracing message effects through receiver activations. We compare the receiver's answers before and after changing a message. Then, we restore selected activations recorded under the original message to determine how much of the message effect these activations reproduce. Based on Social Circuits, we propose Circuit-Guided Deliberation (CGD), which learns to select useful messages using receiver activation changes. We establish when activation replacement preserves receiver decisions and bound the gap between CGD's task performance and the best achievable through message selection. Experiments show that receiver activation changes explain the message effects and guide message selection that improves the task performance. Across three models and four datasets, CGD achieves the highest or joint-highest average accuracy in our main comparisons while generating fewer tokens than multi-agent baselines.

In this article, we examine particle dynamics and Quasi-periodic oscillations within a charged Simpson-Visser-Anti-de Sitter (AdS) spacetime. We first analyze the behavior of the metric and black hole horizons under the competing impact of electric charge $Q$, regularization parameter $b$, and AdS radius $l$. The roots of the metric function reveal transitions between three distinct topological classes: non-extremal configurations featuring distinct inner and outer horizons, extremal limits with a single degenerate horizon, and strictly positive, horizonless wormhole geometries. By evaluating the effective potential, we determine how this modified geometry shifts the Innermost Stable Circular Orbit (ISCO) and alters the angular momentum and energy of test particles. Moreover, we derive the Keplerian, radial, and vertical frequencies to model quasi-periodic oscillations (QPOs). We show that the interplay between the regularization parameter $b$ and the AdS radius $l$ significantly modifies twin-peak QPOs. In particular, $b$ drives fundamental QPO frequencies higher than those predicted by standard general relativistic backgrounds. Twin-peak high-frequency QPO data from two black-hole binaries disfavor the Schwarzschild limit at $\sim\!4\sigma$, but constrain only a degenerate combination of charge and AdS curvature, leaving the regularization parameter entirely unconstrained by the data.

While an unlearned language model may no longer recall a fact directly, the fact often remains recoverable through multi-hop reasoning over related knowledge. Most existing unlearning techniques overlook this vulnerability, targeting facts in isolation while leaving their supporting knowledge intact. To achieve true forgetting, we propose a general deep unlearning framework compatible with existing unlearning algorithms. Our approach adaptively explores both explicit responses and latent internal representations to discover valid reasoning paths, compiles them into a confidence-aware supporting subgraph, and we apply a graph minimum cut to sever all recovery paths while preserving unrelated knowledge. To rigorously evaluate deep unlearning, we introduce a model-specific pipeline that extracts and completes knowledge graphs from raw text, filtering them by calibrated model confidence to reflect what the model genuinely retains. Comprehensive experiments demonstrate that selectively unlearning supporting knowledge yields substantially deeper forgetting than superficial methods while preserving model utility, highlighting that genuine unlearning requires breaking the relational structures that enable factual reconstruction.

The $\triangle$-convexity space on a finite, simple graph $G = (V,E)$ is the collection $\mathcal{C}$ of subsets $S \subseteq V(G)$ such that whenever $x \in V(G)$ forms a triangle with two vertices in $S$, we have $x \in S$. The members of $\mathcal{C}$ are called convex sets, and the convex hull of a set $S \subseteq V(G)$, denoted $\operatorname{Hull}(S)$, is the smallest member of $\mathcal{C}$ that contains $S$. The Carathéodory (resp., exchange) number, $c_{\triangle}(G)$ (resp., $e_{\triangle}(G)$), is the size of a largest Carathéodory (resp., exchange) independent subset of $V(G)$. It was shown by Anand et al. (JCMCC 126, 2025, 11–27) that $c_{\triangle}(G) \leq t(G)+1$ and $e_{\triangle}(G) \leq t(G) + 2$, where $t(G)$ is the number of triangles in $G$, and that these bounds are tight. They also computed $c_{\triangle}(G)$ and $e_{\triangle}(G)$ for a block graph $G$ in terms of the number and arrangement of non-$K_2$ blocks in $G$. In this paper, we point out a gap in the proof in Anand et al. of the first inequality, $c_{\triangle}(G) \leq t(G)+1$, which has consequences for the proof of the second inequality, $e_{\triangle}(G) \leq t(G) + 2$, as well. Moreover, the tightness results are inadvertently applied as characterizations of the extremal graphs, leading to incorrect computations of the Carathéodory and exchange numbers of block graphs in certain cases. We fix these gaps by giving a full proof of the first inequality via a different route from that in Anand et al. Together with the argument in Anand et al., this also completes the proof of the second inequality. Our proof also leads to a characterization of the extremal graphs for each bound, which we use to compute the Carathéodory and exchange numbers of block graphs and to identify the extremal block graphs. We also determine $e_{\triangle}(G)$ exactly for $k$-trees for every $k \geq 2$.

Driving benchmark scores are increasingly used not only for evaluation but also as optimization targets. This raises a fundamental question: do score gains remain reliable evidence of driving improvement once the score itself is optimized? We address this question by examining how the scoring process responds to changes in driving behavior and whether the resulting gains persist under repeated execution and replanning. We decompose the process into execution, measurement, subscore mapping, and aggregation. Controlled interventions reveal substantial behavioral changes that receive little score response because distinctions are omitted, thresholded, or attenuated between requested and executed motion. Closed-loop comparisons further show that optimization gains can reverse when the execution interface changes, demonstrating their dependence on how requests are executed and returned as feedback. Together, these findings connect the behavioral distinctions preserved by a metric to the conditions under which its gains transfer. Metric validity under optimization therefore requires examining both what the scoring process measures and how the optimized behavior is executed.

Medical vision-language pre-training largely follows a visit-level image-report matching paradigm, aligning paired images and reports at individual visits. While effective for static cross-modal correspondence, this paradigm provides limited supervision for longitudinal clinical change, such as whether abnormalities improve, remain stable, or worsen over time. Learning such change is challenging because temporal semantics are implicit in free-text reports, and different abnormalities within the same patient may evolve asynchronously or even in opposite directions. We propose MedCTA, which reframes medical vision-language pre-training from visit-level cross-modal matching to learning clinical change. Rather than compressing a patient history into a single temporal representation, MedCTA models clinical change at two complementary scopes. At the abnormality scope, clinically grounded queries construct abnormality-conditioned visual and textual trajectories to capture heterogeneous abnormality evolution. At the patient-course scope, global image and report sequences are modeled to capture overall clinical progression beyond any individual abnormality. Structured trend supervision is extracted from longitudinal reports by an offline LLM parser, removing the need for manual temporal annotations. Combined with static image-report alignment, MedCTA learns representations that preserve visit-level cross-modal correspondence while encoding longitudinal change semantics. Experiments on temporal image classification, image-text retrieval, and zero-shot classification show consistent gains over strong medical vision-language baselines.

Long-term memory is essential for language agents to maintain coherent and effective behavior over extended, multi-session interactions. Existing memory systems mainly use retrieval at read time, while write-time memory formation still relies on direct extraction or compression. However, when future information needs are unknown, compressing an entire interaction in one pass can overlook locally important details that may matter later. To this end, we introduce RIME, a retrieval-induced memory framework that shifts memory construction from monolithic compression toward evidence-centered integration. RIME uses generic self-questions to retrieve focused dialogue evidence and grounds memory formation in both the retrieved evidence and relevant historical memories, which are jointly reconciled into an evolving memory bank with temporal and provenance information. At inference time, compressed memory serves as the primary rather than the sole source of evidence: when it cannot support an answer, RIME retrieves relevant source dialogue together with its local context to recover information omitted during memory formation, without resorting to full-history processing. Extensive experiments on LoCoMo with Qwen3-235B-A22B and GPT-5.6 Sol show that RIME consistently achieves the best performance across all three quality metrics among the compared methods, while requiring substantially fewer query-time LLM tokens.

In this largely expository article we give an introduction to higher dimensional local class field theory, more precisely class field theory for smooth projective schemes over local fields. We begin with a quick summary of some of the main results of classical local class field theory. Then we explain the main definitions and statements in the higher dimensional setting. We give a new and simple proof of the $\ell$-part, which is originally due to Jannsen–Saito and Forré. Inspired by this proof, we propose a conjecture which would imply the remaining open cases with $p$-coefficients and which improves our understanding of the $p$-part. A particular goal of the exposition is to outline analogies between the classical and the higher dimensional theory and on how the unit filtration and ramification, i.e. the $p$- and the $\ell$-part, can be interpreted in the higher dimensional setting.

We give an explicit hitting set of size $\operatorname{poly}(n,d,r)$ for the class of $n$-variate degree-$d$ polynomials whose partial derivative space is bounded by $r$, over any field $\mathbb{F}$ of characteristic zero (or large enough). In particular, this yields a polynomial sized hitting set for the class of depth-$3$ powering circuits. The proof extends to the setting of noncommutative ABPs and provides a polynomial sized hitting set (of matrices) for the class of noncommututive polynomials that can be computed by polynomial sized noncommutative ABPs (this was also one of the results reported by Open AI~\cite{OpenAI26} in their release on 6 Oct 2026). The main technical insight is the construction of a “formal derivation” and properties of the associated Wronskian with respect to this derivation, which was previously studied by Moura \cite{Moura_2004} in a very different context. The proofs in this paper are elementary and completely self-contained. AI disclosure: A subset of the stewards obtained a weaker statement (hitting sets for polynomials with bounded partial derivative space) during conversations \cite{astra_proof} with OpenAI GPT-6 Astra. Extensions of this idea to the broader class of noncommutative ABPs were obtained by the other stewards (via further conversations with Astra), prior to OpenAI's dump on 6 Oct 2026. The proofs presented in this writeup is a rewriting (in the authors' words) of the proofs for both settings obtained by the models in a form that we believe is understandable to researchers. Although the proof is technically identical to OpenAI's release, we believe this exposition is self-contained, simpler, and more valuable to the community. The absurdity of the situation is not lost on us.

We prove a rigidity theorem: a globally symmetric complex Finsler space $(M, J, F)$ is necessarily a Kähler-Berwald space, namely $F$ must be a Kähler-Berwald metric. We also obtain a Schwarz lemma for holomorphic mappings $f$ from an arbitrary bounded symmetric domain $\mathfrak{D}$ into itself whenever $\mathfrak{D}$ is endowed with an $\mbox{Aut}(\mathfrak{D})$-invariant Kähler-Berwald metric $F$ such that its holomorphic sectional curvature is bounded below and above by negative constants $-K_1<0$ and $-K_2<0$, respectively. The novelty of this Schwarz lemma is that the Lu constant of $(\mathfrak{D},F)$ is optimal for each given $F$ whenever rank$(\mathfrak{D})\geq 2$.

In this paper, we introduce and investigate the notion of phi-semi contractibility for Banach algebras. We establish a complete characterization of phi-semi contractibility through the existence of specific elements satisfying particular algebraic conditions. We thoroughly examine the hereditary properties of this concept, including its behavior under closed ideals, continuous homomorphisms with dense range, and projective tensor products. Notably, we prove that the projective tensor product A tensor B is phi tensor psi-semi contractible precisely when both A and B are semi contractible with respect to their respective characters, where phi belongs to the character space of A and psi belongs to the character space of B. Furthermore, we introduce and analyze the approximate version of phi-semi contractibility, providing equivalent characterizations.

Generating a plausible clinical trajectory does not establish what would happen under a different treatment. We present ADMIT, a framework combining irregular multimodal representations, treatment-conditioned latent diffusion and explicit constraints on generated states or actions. We formulate its interventional target through sequential g-computation and distinguish causal assumptions from constraint satisfaction. Its admissibility mechanism translates physiological prior knowledge into explicit constraints on generated states and proposed actions. Treatment-exposure dynamics condition latent transitions, while state projection or action gating applies the constraints during rollout so that they influence subsequent trajectory generation. In our preliminary experiments, multimodal inputs improved supervised hidden-state recovery and reduced treatment-contrast error. In a simulated dosing-schedule experiment with leak-free history encoding, ADMIT predicted most of the tumor-volume change caused by redistributing a fixed total dose. An exposure input improved these predictions around a temporary dose reduction whether or not the assumed clearance rate was correct, but reduced the predicted size of a dose effect, and a deterministic recurrent baseline matched ADMIT's average predictions. Exposure projection reduced constraint violations, although enforcement remained incomplete. Semi-synthetic experiments using eICU context illustrated treatment-response generation under fixed and adaptive policies. Observational examples further characterize model treatment sensitivity. ADMIT provides a framework for testing whether complementary observations and physiological restrictions improve intervention trajectories, with representation recovery, effect accuracy and rule enforcement assessed separately.

Large language model (LLM) agents execute applications through a workflow of inference requests with tool calls and user interactions. Serving these applications at production scale requires understanding how application behavior shapes inference demand and for guiding efficient execution. Recent characterization studies provide request-level workload measurements and agent execution analysis. However, an end-to-end view connecting task initiation, workflow execution, and inference infrastructure remains unexplored. In this paper, we analyze a two-week trace of 11.7 million requests from a large-scale production platform for general-purpose agents, backed by inference infrastructure comprising over 10k GPUs. We characterize the platform at three connected levels: task-level initiation semantics, workflow-level execution patterns, and infrastructure level serving demands. Our measurements reveal workload patterns such as highly skewed request volumes across sessions, rare execution overlap among logical sibling requests, and context reuse across task boundaries. Building on these observations, we analyze deployment implications and identify open problems to guide future research on agent serving systems.

Conditional Flow Matching (CFM) models for text-to-speech (TTS) suffer from incoherent frequency evolution during inference. While similar spectral imbalances are addressed in diffusion models for other domains, those generic solutions fail to generalize to the inherently uncoordinated acoustic dynamics of CFM. We demonstrate that this issue can be effectively mitigated by introducing a novel training-free frequency-selective boosting strategy. Using the Discrete Wavelet Transform (DWT), our method dynamically modulates mel-spectrogram sub-bands during ODE integration, synchronizing spectral development by penalizing aggressive low-frequency growth and boosting lagging high-frequency details. Validated across diverse architectures (Matcha-TTS, F5-TTS, IndicF5), our approach reduces the required Number of Function Evaluations (NFE) from 32 to 26 and improves Frechet Audio Distance (FAD) by up to 61%, all without compromising mean opinion scores, speaker similarity, and speech intelligibility.

Accurate retrieval of human motions is a crucial first step in text-guided human motion modeling and synthesis, as it selects semantically relevant sequences from large datasets and provides grounded references for downstream tasks. Retrieving motions from natural language descriptions remains challenging because sentences can describe multiple actions, overlapping movements, and intricate dependencies between body parts. Existing methods often focus on simple, single-action descriptions and typically process body parts independently or by merely concatenating features, without explicitly modeling how torso movements influence other parts. In addition, their processing pipelines often rely on computationally heavy models, introducing considerable overhead, particularly when modeling longer or more complex motion sequences. This limits learning discriminative motion-pattern representations, reducing retrieval accuracy, interpretability, and efficiency in practical applications. To address these limitations, we propose HUMAN-TCI, a Hierarchical Multi-Stream Motion-Aware Network for text-guided human motion retrieval. HUMAN-TCI employs a three-stream architecture that separately models upper-body, lower-body, and torso motions while explicitly capturing their interactions, allowing torso-related movements to influence the positioning and dynamics of other body parts. By incorporating tailored torso attention, our model effectively recognizes complex human motion patterns, captures fine-grained motion relationships and handles complex multi-action descriptions. Our framework supports retrieval for both simple, single-action sentences and long, compositional descriptions containing sequential or overlapping actions without relying on complex models.

We study the wall crossing for the moduli space of unmarked cubic surfaces at the log Calabi-Yau threshold. We prove the existence of a good moduli space for the corresponding moduli stack of boundary polarized Calabi-Yau pairs with Gorenstein underlying surfaces and describe the wall crossing morphisms explicitly. More precisely, the morphism from the K-moduli space to the log Calabi-Yau moduli space is an isomorphism, whereas the morphism from the KSBA moduli space contracts a divisor isomorphic to $\mathbb{P}^3$ to a point and is an isomorphism away from this divisor.

Agentic tasks require a large language model to interact with the world, navigating information and gathering evidence across multiple steps with restricted resources. Due to this complexity, agentic task failures arise from various sources, and pinpointing these failure causes is essential to diagnose and improve agentic systems. Existing benchmarks, however, tend to focus on a single leaderboard score, leaving the underlying failure modes opaque. To fill this gap, we introduce AgentHop, a diagnostic benchmark of 1,011 multiple-choice questions paired with a controlled seven-tool sandbox under fixed token, turn, and tool-call constraints. AgentHop reveals model vulnerabilities by dissecting a single accuracy score along four axes of agent operation: retrieval, synthesis, tool-call, and resource management. Across 19 models, we find that behavior clusters by model family, with tool-call signatures revealing distinct family fingerprints: GPT models commit early, Anthropic and GLM checkpoints verify before committing, DeepSeek and Kimi over-search, and Gemini-3 Pro stays balanced. Decomposed axes further expose within-family structure: Claude Opus 4.6 and Sonnet 4.6 land within one accuracy point yet diverge on retrieval-versus-synthesis emphasis, with Opus retrieving more and Sonnet synthesizing better. We release the full benchmark set and the harness to support diagnostic agent benchmarking.

Scaling LLM-based optimization from textbook-scale instances to real-world, industrial tasks remains a critical open challenge. Existing approaches are predominantly evaluated on small, self-contained textual problems and often commit to a solver-integrated paradigm, limiting their ability to handle the scale and structural diversity of practical optimization workloads. In this work, we propose a practical framework for training open-source LLMs to tackle real-world, industrial-scale optimization. We first show empirically that solver-integrated reasoning, exact combinatorial algorithm, and heuristic search exhibit complementary strengths across different problem structures and scales. Motivated by this, we introduce Strategy-Diverse Reinforcement Learning (SDRL), which trains LLMs as adaptive optimization meta-solvers. SDRL leverages this complementarity through a correctness-gated hierarchical diversity reward that promotes robust exploration across varying strategies and within each strategy, effectively preventing premature strategy collapse. We further introduce a mixed-format training scheme that jointly supports both self-contained textual problems and file-grounded instances. Across comprehensive evaluations, our framework outperforms existing fine-tuned methods and frontier models including DeepSeek-V4-Pro and GPT-5.5, both on average across benchmarks and on industrial-scale optimization tasks. We release our code and training data to facilitate future research https://github.com/tokentideAI/LLMMetaSolver.

This paper addresses the problem of safe offline reinforcement learning, which involves training a policy to satisfy safety constraints using an offline dataset. This problem is inherently challenging as it requires balancing three highly interconnected and competing objectives: satisfying safety constraints, maximizing rewards, and adhering to the behavior regularization imposed by the offline dataset. To tackle this trilogy challenge, we propose Q-learning Penalized Transformer policy (QPT), a training–inference consistent framework that bridges conditional sequence modeling with constraint-aware value estimation. QPT trains a Transformer policy that generates actions conditioned on trajectory context and target return/cost, retaining strong behavior regularization. To inject explicit safety semantics during learning, we augment sequence-model training with a Q-shaped penalty using learned reward and cost Q-functions to favor high return under low constraint violation. At inference, the same Q-functions enforce the cost threshold and choose the highest-reward feasible action, closing the loop between training and deployment. We provide a principled analysis under stylized near-deterministic CMDPs, characterizing how Q-penalized conditional generation improve safety and performance. Empirically, QPT consistently outperforms strong safe offline RL baselines across 38 tasks on the DSRL benchmark, and exhibits robust zero-shot adaptation to different constraint thresholds.

Topic segmentation structures spoken documents into coherent sections, facilitating navigation and downstream understanding. The appropriate granularity can vary substantially, ranging from broad thematic shifts to fine-grained subtopics. Existing LLM-based segmenters, however, often struggle to adapt to this variation, causing them to either merge distinct subtopics or over-segment coherent themes. To address this, we introduce Cue-Grounded Segmentation (CGS), a training-free framework that operates without any task-specific supervision. CGS first identifies phrases that explicitly signal the start of a new topic and uses their sentence positions as segment boundaries. When such cues are insufficient, it falls back to semantic segmentation, guided by the document structure inferred during cue extraction. Across six benchmarks and six LLM backbones, CGS consistently outperforms existing baselines, remains robust to noisy ASR transcripts, and achieves these gains with low API cost on proprietary models.

Repeated threshold queries may reuse selected identities without reusing stale reports, but cheaper certificates need not shorten the complete response. We study selected-lower, atomic-upper (SLA) certificates for fixed-rank conjunctive screening with explicit missing-information semantics. An endpoint characterization and a counterexample separate same-source containment from policy-dependent online behavior. The original 320-session experiment reduces summed construction medians by 31.18% against an exclusion-cover certificate, yet its Cover/SLA full-API geometric time ratio is 0.9682 (95% conditional blocked interval 0.9593-0.9769), and SLA takes 12.94% more summed time than uncached Bitmap. Three separately launched complete repeats preserve this adverse ordering, with Cover/SLA ratios of 0.9665-0.9718. An additional 720-configuration exploration varies catalogue size, construction period, requested count and query locality on empirically resampled tables. SLA is faster in 295 configurations against Cover and 271 against Bitmap, descriptive counts that do not establish universal superiority. Separately instrumented additive costs distinguish construction savings from retrieval and report costs. A plane-stress component case adds an independent analytical displacement check and 4,608 boundary-challenging queries: five implementations agree exactly, while medium- and fine-mesh selections differ at 459 positions. A state-stratified public bolt-record exercise preserves 185 incomplete positions among 1,479 requests. Raw timings, complete configuration results and a tested clean-environment package support reproducibility. The SLA construction was explored and refined through the self-evolving AI system ZiYor; the named authors specified, implemented and evaluated it. This is a bounded mechanics-to-query study, not physical joint qualification or universal speedup.

Non-stationary online learning has attracted much attention in recent years, as static regret is insufficient to guide algorithm design in changing environments. To address this limitation, interval regret and dynamic regret have been introduced as two representative performance metrics that strengthen static regret in complementary directions. Interval regret requires an online algorithm to achieve competitive static regret over every local time interval, whereas dynamic regret evaluates performance against an arbitrary sequence of time-varying comparators. Despite their importance, the relation between these metrics has long remained unclear. Prior work has often regarded interval regret as the stronger notion, based on the intuition that local guarantees should naturally induce global guarantees. Consequently, it is widely conjectured that an algorithm with optimal interval regret should automatically attain optimal dynamic regret. In this paper, we first establish a negative result that refutes this intuition of a metric-level implication. Specifically, for both convex and curved functions (including exp-concave and strongly convex functions), we show that there exist instances in which an algorithm with optimal interval regret nevertheless fails to achieve optimal dynamic regret. We then show how to leverage local adaptivity to obtain optimal dynamic regret. In particular, optimal dynamic regret can be attained by invoking an interval regret minimization process over an enlarged Euclidean ball containing the original convex feasible domain and using a suitable domain-converted surrogate loss. This reduction applies to both convex and curved functions. As a byproduct, we obtain the first proper and efficient algorithm with optimal dynamic regret for exp-concave functions, improving prior results while significantly simplifying the analysis.

Language-model agents increasingly tackle long-horizon tasks whose interaction histories exceed the model's active context. Recent work has begun to use reinforcement learning to make memory control part of the policy, often relying on predefined memory tools within domain-specific training environments of relatively short horizons. This setup ties learned memory behavior to environment-specific interfaces that lie outside the base model's pre-training and must be learned from scratch, so even after post-training, agents struggle to use memory in long-horizon tasks. To address these limitations, we introduce Coding Agent Memory Gym (CAMG), a suite of long-horizon agentic-RL environments spanning Shop, Coding, DeepResearch, and AutoResearch. Alongside each environment's native task interface, CAMG provides executable shell access and an episode-persistent workspace, enabling agents to create, revise, search, and reuse files as memory throughout an episode. We also introduce CAMG-RL, which trains a single policy jointly across all four environments with fully asynchronous PPO, learning this file-based memory behavior directly from downstream task reward, and we train CAMG-RL-4B and CAMG-RL-9B from Qwen3.5 models of matching size. On SWE-bench Verified and MLE-bench Lite, CAMG-RL-4B and CAMG-RL-9B are competitive with Qwen3.5-35B-A3B and Qwen3.5-122B-A10B, respectively.

We investigated the scaling relation of the magnetohydrodynamic turbulence in the inner and outer shells of the supernova remnant HB 9 using multi-tracers. For the eastern inner shell section, one-dimensional structure function and power spectrum analyses of the radio total intensity, H-alpha emission, and polarized intensity consistently show a Kolmogorov-like scaling of l^2/3 over the inertial range. The two-dimensional power spectrum for the entire inner region follows k^-8/3 of all tracers, including the polarization gradient. It indicates that fully developed, trans-Alfvenic Kolmogorov turbulence is present in the inner shell of HB 9. We estimated the strength of random magnetic field therein from the low fractional polarization (5%), and obtained a upper limit delta B of 18 micro G. The turbulence in the inner shock shell is most likely driven by the Richtmyer-Meshkov instability.

Graph Transformers (GTs) offer superior representation capabilities by overcoming the depth limitations and over-smoothing issues of traditional Graph Neural Networks (GNNs). However, scaling GTs to large graphs poses critical bottlenecks. Specifically, the attention score matrix and its associated topology-aware bias matrix jointly incur significant per-layer memory overhead, and heavy graph embedding layers result in severe workload imbalances. These characteristics are unique to GT training and are not addressed by parallelism techniques designed for either conventional GNNs or Transformers, making a dedicated solution necessary. This paper introduces MegaGraph, the first automated hybrid parallelism framework designed for efficient GT training. MegaGraph designs three specialized strategies, namely graph-aware context parallelism, heterogeneous pipeline parallelism, and hybrid data parallelism, to support efficient training on large-scale graphs. However, coordinating these three parallelism strategies yields an exponentially large configuration space. To address this complexity, an automatic search engine leverages precise cost models via a Profile - Model - Search workflow to identify the optimal parallelism configuration. Evaluations demonstrate that MegaGraph enables training on large-scale graphs where state-of-the-art baselines fail due to out-of-memory (OOM) errors. The framework reduces per-device peak memory by up to 77.8% and achieves up to 4.51$\times$ training speedup while maintaining model accuracy.

Automatic human-like facial reaction generation (FRG) is essential for building intelligent systems that can engage in human-computer interaction (HCI). While diverse and context-appropriate facial reactions can reflect latent appraisal and affective processes in human interaction, most existing FRG methods rely on end-to-end architectures that directly map speaker behaviours to listener expressions without an explicit intermediate internal state. We reformulate FRG as generation mediated by a structured internal-state process and propose the Intentional Agent, which shifts FRG from direct stimulus-response mapping to stimulus-grounded generation through explicit intermediate states. To represent temporal internal-state evolution, we propose an internal dynamics model that integrates emotional drives with an iterative Inner Thought Flow (ITF) within a structured intermediate state used for subsequent generation. This state can continue to update during conversational silences. Furthermore, to bridge abstract internal states with physiological actions, we formulate FRG as a downstream affective mapping from this latent thought flow to facial expressions. Experiments on the REACT 2025 dataset show an FRDist of 72.39 and an FRDiv of 0.5057; perceptual plausibility is evaluated separately through blinded human ratings. A blinded human evaluation of 96 reactions found no significant difference in mean score between Full and ground truth ($5.527$ vs.\ $5.195$, $p_{\mathrm{Holm}}=.076$), while Full significantly outperformed Event-Triggered and Heuristic-Only (both $p_{\mathrm{Holm}}<.001$). The Reaction Quality Scorer (RQS) correlated strongly with human judgements (Pearson $r=.855$; Spearman $\rho=.821$, both $p<.05$), supporting its use as an automatic metric. These results underscore the immense potential of endogenous dynamics in building highly autonomous, human-like agents.

Maintaining persona consistency across multi-turn dialogues remains a core challenge for role-playing language models. Off-policy distillation from external teachers incurs distribution mismatch that compounds across dialogue turns, while reinforcement learning struggles with reward ambiguity inherent in subjective persona fidelity. We propose OSPD, an on-policy self-distillation framework where the same model serves as both teacher and student under asymmetric information: the teacher receives a complete character profile while the student sees only a brief summary, and the student generates trajectories from its own policy. We find that teacher confidence in role-playing dialogue exhibits a bimodal structure—sharply peaked at character-critical tokens yet diffuse at generic utterances—and introduce role-aware divergence switching to match this structure. A progressive trait masking curriculum further forces staged internalization of character knowledge along semantic dimensions. Experiments on CharacterBench, CharacterEval, and SocialBench show that OSPD substantially improves persona consistency over supervised fine-tuning and multi-turn RL baselines, without requiring any external teacher or reward model.

Context. Abundance ratios between elements synthesized on different nucleosynthetic timescales provide empirical chemical clocks for estimating stellar ages, but robust calibrations require homogeneous samples with precise stellar parameters and elemental abundances. Aims. We aim to investigate whether the presence of planets modifies empirical chemical clock relations. Methods. We present high-precision stellar parameters and elemental abundances for 52 new planet-hosting targets observed with Keck/HIRES, LBT/PEPSI, and Subaru/HDS. Combined with PASTA I and II, the full sample comprises 94 stars, of which 81 are identified as planet-hosting solar twins and analogs, and is used to characterize Galactic chemical evolution and derive empirical chemical clock relations. Results. We recover the expected Galactic chemical evolution trends, with $\alpha$-elements such as Mg and S increasing with stellar age, while s-process elements such as Y, Ba, and Sr show negative age correlations. In contrast, Fe-peak elements and the r-process element Eu show little or no significant age dependence. These trends give rise to tight chemical clock relations based on [Y/Mg], [Y/Al], [Ba/Mg], [Ba/Al], and [Sr/Mg], with [Y/Mg] and [Y/Al] providing the strongest correlations. [Ba/Al] shows the steepest age dependence, and the Ba-based ratios are generally more sensitive to age than their Y-based counterparts. Conclusions. Planet-hosting solar analogs follow chemical-clock relations consistent with those of nearby solar twins, with no evidence for a significant planet-related effect. The derived relations are primarily applicable to local solar twins and analogs in the Galactic thin disk.