arXiv:2605.17669v1 Announce Type: new
Abstract: The preservation and interpretation of cultural heritage increasingly rely on digital technologies, among which Knowledge Graphs (KGs) stand out for their ability to structure vast amounts of data. However, the construction and expansion of these KGs often face challenges due to the diverse and complex nature of cultural heritage information. In this paper, we propose a novel approach for extending KG resources in the domain of cultural heritage, which we applied to French data. First, we introduce a new knowledge graph in the domain of French cultural heritage, WJoconde, which is distinguished by its multimodality as it integrates both textual and image information of the entities. We further introduce three variants of WJoconde to facilitate downstream research, such as Knowledge Graph Completion (KGC). We also built a comprehensive benchmark for KGC methods on our dataset. Second, we propose a new framework for extending cultural heritage KGs using multi-modal approaches leveraging Large Language Models (LLMs) and Vision-Language Models (VLMs), which includes automated data extraction from unstructured resources combined with a special validation pipeline for grounding the output of both models, to further extend WJoconde. Our results show that by integrating the rich text and image information in cultural heritage data, we can efficiently enhance KGs with high reliability. We open-source all code and benchmark datasets with text and images, as well as the original data with an interactive access point
Science Journals
arXiv:2605.13877v2 Announce Type: replace
Abstract: We present ARES-LSHADE, a memetic differential-evolution variant submitted to the GECCO 2026 competition on LLM-designed evolutionary algorithms for the Generalized Numerical Benchmark Generator (GNBG). The algorithm builds on the LLM-LSHADE 2025 winner, contributing two new components: (a) a scout-augmented mutation operator with adaptive CMA-ES integration, produced by an autonomous research loop across approximately thirty LLM-driven design experiments, and (b) a multi-start L-BFGS-B polish phase that respects strict blackbox treatment of the benchmark. On the official 31-run-per-function evaluation with the competition-specified function-evaluation budgets, ARES-LSHADE obtains 510 of 744 wins (per-function gap below 1e-8), reaching machine precision on 18 of 24 functions. The remaining six functions exhibit characteristic plateau signatures consistent with GNBG's compositional structure, and were independently identified by the autoresearch loop as the hardest of the suite. Beyond the result itself, this report documents two methodological observations: (i) an LLM-driven research loop with operator-only edit surface and fitness-only observation space converges to a characteristic plateau on this benchmark; (ii) when we initially widened the observation space to include the benchmark's compositional metadata, the resulting algorithm trivially solved all 24 functions but violated the competition's blackbox rule, which we identified before submission. We discuss this tension between LLM capability and benchmark integrity as a design consideration for future LLM-driven optimization-algorithm research. Code and reproducibility artifacts are available at https://github.com/anaeem1/ARES-LSHADE.
arXiv:2605.17601v1 Announce Type: new
Abstract: We present a Learning from Demonstration (LfD) framework that achieves one-shot generalization in multi-stage, contact-rich manipulation tasks. Central to our approach is the utilization of environmental constraints as the inductive bias. By representing a demonstration as a sequence of behaviors that exploit environmental constraints, the robot separates task-general structure -- the constraint types and their transitions -- from instance-specific details such as exact demonstration trajectories, poses, and local geometries. Our four-stage pipeline builds a complete policy on this representation: the robot first abstracts a single demonstration into environmental-constraint primitives, then disambiguates them through self-guided exploration, next assimilates targeted human corrections that handle out-of-distribution variations, and finally recovers the abstracted-away details online through compliant interaction. Because the resulting policy follows constraints rather than mimics trajectories, it generalizes across object poses, local geometries, and unmodeled contact dynamics. We validate our approach on seven real-world multi-stage contact-rich manipulation tasks and achieve over 90% success. These extensive experimental results establish environmental constraints as fundamental building blocks for efficient generalization in learning from demonstration.
arXiv:2605.18078v1 Announce Type: new
Abstract: Multi-agent policy-gradient methods have been shown to converge locally near stable Nash equilibria. Local convergence, however, does not determine which equilibrium is reached. We study this question through basin-entry probability with respect to a target set of equilibria selected by an external criterion, such as payoff dominance. For finite-unroll Meta-MAPG, we show that the update decomposes into ordinary policy gradient plus own-learning and peer-learning corrections, with controlled sampling noise and finite-unroll bias. We identify the peer-learning correction as the main equilibrium-selection mechanism: under a local alignment condition, the probability of entering the certified attraction region of the target stable-Nash set increases, relative to ordinary policy gradient. Because persistent correction may shift zero-update points of the original game, annealing the correction after entering the basin recovers ordinary policy-gradient dynamics and inherits local stable-Nash convergence guarantees. Experiments in Stag Hunt, iterated Prisoner's Dilemma, and preliminary neural-policy coordination environments support this basin-entry view, showing increased entry into cooperative basins under peer-aware updates.
arXiv:2605.10923v2 Announce Type: replace
Abstract: Large language model agents increasingly rely on external skills to solve complex tasks, where skills act as modular units that extend their capabilities beyond what parametric memory alone supports. Existing methods assume external skills either accumulate as persistent guidance or internalized into the policy, eventually leading to zero-skill inference. We argue this assumption is overly restrictive, since with limited parametric capacity and uneven marginal contribution across skills, the optimal active skill set is non-monotonic, task- and stage-dependent. In this work, we propose SLIM, a framework of dynamic Skill LIfecycle Management for agentic reinforcement learning (RL), which treats the active external skill set as a dynamic optimization variable jointly updated with policy learning. Specifically, SLIM estimates each active skill's marginal external contribution through leave-one-skill-out validation, then applies three lifecycle operations: retaining high-value skills, retiring skills whose contribution becomes negligible after sufficient exposure, and expanding the skill bank when persistent failures reveal missing capability coverage. Experiments show that SLIM outperforms the best baselines by an average of 7.1% points across ALFWorld and SearchQA. Results further indicate that policy learning and external skill retention are not mutually exclusive: some skills are absorbed into the policy, while others continue to provide external value, supporting SLIM as a more general paradigm for skill-based agentic RL.
Multi-task learning on partially labeled datasets via invariant/equivariant semi-supervised learning
arXiv:2605.17624v1 Announce Type: new
Abstract: We investigate the potential of invariant and equivariant semi-supervised learning for addressing the challenges of training multi-task models on partially labeled datasets with differently structured output tasks. Specifically, we use the popular FixMatch method for invariant semi-supervised learning and its equivariant extension Dense FixMatch. We evaluate their performance on the Cityscapes and BDD100K datasets in the context of the prevalent object detection and semantic segmentation tasks in computer vision. We consider varying sizes of the subsets annotated for each task and different overlaps among them. Our results for both invariant and equivariant semi-supervised learning outperform supervised baselines in most situations, with the most significant improvements observed when fewer labeled samples are available for a task and generally better results for the latter approach. Our study suggests that invariant/equivariant learning is a promising general direction for multi-task learning from limited labeled data.
arXiv:2605.11654v2 Announce Type: replace
Abstract: Cross-view geo-localization (CVGL), which matches an oblique drone view to a geo-referenced satellite tile, has emerged as a key alternative for autonomous drone navigation when GNSS signals are jammed, spoofed, or unavailable. Despite strong recent progress, three limitations persist: (1) global-descriptor designs compress the patch grid into a single vector without separating layout from texture across the view gap; (2) altitude-related scale variation is retained in the learned embedding rather than marginalized; and (3) multi-objective training relies on hand-tuned scalars over losses on incompatible gradient scales. We propose SkyPart, a lightweight swappable head for patch-based vision transformers (ViTs) that institutes explicit part grouping over the patch grid. SkyPart has four theory-grounded components: (i) learnable prototypes competing for patch tokens via single-pass cosine assignment; (ii) altitude-conditioned linear modulation applied only during training, making the retrieval embedding altitude-free at inference; (iii) a graph-attention readout over active prototypes; and (iv) a Kendall uncertainty-weighted multi-objective loss whose stationary points are Pareto-stationary. At 26.95M parameters and 22.14 GFLOPs, SkyPart is the smallest among top-performing methods and sets a new state of the art on SUES-200, University-1652, and DenseUAV under a single-pass, no-re-ranking, no-TTA protocol. Its advantage over the strongest baseline widens under the ten-condition WeatherPrompt corruption benchmark.
arXiv:2605.10930v2 Announce Type: replace
Abstract: Large Language Models (LLMs) and Large Reasoning Models (LRMs) are increasingly used for critical tasks, yet they provide no guarantees about the correctness of their solutions. Users must decide whether to trust the model's answer, aided by reasoning traces, their summaries, or post-hoc generated explanations. These reasoning traces, despite evidence that they are neither faithful representations of the model's computations nor necessarily semantically meaningful, are often interpreted as provenance explanations. It is unclear whether explanations or reasoning traces help users identify when the AI is incorrect, or whether they simply persuade users to trust the AI regardless. In this paper, we take a user-centered approach and develop an evaluation protocol to study how different explanation types affect users' ability to judge the correctness of AI-generated answers and engender false trust in the users. We conduct a between-subject user study, simulating a setting where users do not have the means to verify the solution and analyze the false trust engendered by commonly used LLM explanations - reasoning traces, their summaries and post-hoc explanations. We also test a contrastive dual explanation setting where we present arguments for and against the AI's answer. We find that reasoning traces and post-hoc explanations are persuasive but not informative: they increase user acceptance of LLM predictions regardless of their correctness. In contrast, dual explanation is the only condition that genuinely improves users' ability to distinguish correct from incorrect AI outputs.
arXiv:2605.18157v1 Announce Type: new
Abstract: We introduce a class of cooperative games induced by weighted directed graphs. Specifically, the coalitional value combines an internal interaction term given by the induced subgraph game with an external component based on minimal incoming edges from outside the coalition. The resulting game has a convenient representation in terms of unanimity games. This representation enables closed-form polynomial-time formulas for the Shapley and Banzhaf values. We further establish that the game has a nonempty core and is totally balanced. The class of such games therefore provides an analytically and computationally tractable example of structured network- induced cooperative games in which stability-based allocations and fairness-based solution concepts do not coincide.
arXiv:2505.16831v3 Announce Type: replace
Abstract: Unlearning in large language models (LLMs) aims to remove specified data, but its efficacy is typically assessed with task-level metrics like accuracy and perplexity. We show that these metrics can be misleading, as models can appear to forget while their original behavior is easily restored through minimal fine-tuning. This \emph{reversibility} suggests that information is merely suppressed, not genuinely erased. To address this critical evaluation gap, we introduce a \emph{representation-level analysis framework}. Our toolkit comprises PCA similarity and shift, centered kernel alignment (CKA), and Fisher information, complemented by a summary metric, the mean PCA distance, to measure representational drift. Applying this framework across multiple unlearning methods, data domains, and LLMs, we identify four distinct forgetting regimes based on their \emph{reversibility} and \emph{catastrophicity}. We compare recovery strategies and show that relearning efficiency relies on the data source. We also find that irreversible, non-catastrophic forgetting is exceptionally challenging. By probing unlearning limits, we identify a case of seemingly irreversible, targeted forgetting, offering insights for more robust erasure algorithms. Overall, our findings expose a gap in current evaluation and establish a representation-level foundation for trustworthy unlearning.
arXiv:2605.17572v1 Announce Type: new
Abstract: Motivated by the controller placement problems in software-defined networks and the fair division principles of classical "cake cutting", we investigate the following two-player zero-sum game. In our model, a defender places a limited number of controllers on graph vertices, while an attacker deletes a limited number of vertices. The defender score is the total number of surviving vertices reachable from any remaining controller. We formalize the computational problems associated with various game dynamics (defender plays first; attacker plays first; players play simultaneously; pure or mixed strategies).
We show that these natural problems are $\mathsf{NP}$-complete or $\Sigma^\mathsf{P}_2$-complete, depending on the specific variant. These hardness results provide limitations for optimal controller placement algorithms under different notions of quality of a solution. Finally, we present structural insights that yield efficient algorithms for restricted graph classes (namely interval graphs and graphs of bounded treewidth).
arXiv:2605.11365v2 Announce Type: replace
Abstract: Automated systems built on artificial intelligence (AI) are increasingly deployed across high-stakes domains, raising critical concerns about fairness and the perpetuation of demographic disparities that exist in the world. In this context, causal inference provides a principled framework for reasoning about fairness, as it links observed disparities to underlying mechanisms and aligns naturally with human intuition and legal notions of discrimination. Prior work on causal fairness primarily focuses on the standard machine learning setting, where a decision-maker constructs a single predictive mechanism $f_{\widehat Y}$ for an outcome variable $Y$, while inheriting the causal mechanisms of all other covariates from the real world. The generative AI setting, however, is markedly more complex: generative models can sample from arbitrary conditionals over any set of variables, implicitly constructing their own beliefs about all causal mechanisms rather than learning a single predictive function. This fundamental difference requires new developments in causal fairness methodology. We formalize the problem of causal fairness in generative AI and unify it with the standard ML setting under a common theoretical framework. We then derive new causal decomposition results that enable granular quantification of fairness impacts along both (a) different causal pathways and (b) the replacement of real-world mechanisms by the generative model's mechanisms. We establish identification conditions and introduce efficient estimators for causal quantities of interest, and demonstrate the value of our methodology by analyzing race and gender bias in large language models across different datasets.
arXiv:2605.17556v1 Announce Type: new
Abstract: Clay sculpting is a nuanced, artistic task involving dexterous manipulation with long-horizon planning to achieve high-level goals. As a robotics problem, we formulate clay sculpting as a shape-to-shape matching challenge. Prior deformable object manipulation work either requires retraining a policy per goal or relies on dynamics models which represent state as sparse point clouds which do not capture important clay features, such as textures, well. We present a method for modeling the dynamics of deformable materials and planning for robotic sculpting in a representation that is visually-aligned, capturing lighting and texture features. With three different deformable materials and various end-effectors, we demonstrate that our dynamics model is comparable in performance to the state-of-the-art with the added benefit of being compatible with visual planning. Our actions are represented as parametrized pushes into clay with a single end-effector, which proved to be suitable for long-horizon (>100 actions) clay relief sculptures. Lastly, we show the benefits of planning in a visually-aligned representation, but also provide analysis providing evidence as to why this representation is challenging to plan in compared to 3D representations.
arXiv:2605.18710v1 Announce Type: new
Abstract: With the wide adoption of Multimodal Models (MMs) in real-world scenarios, it is significant to efficiently train emerging MMs that exhibit increasingly complex module architectures. For MM deployment, existing works allocate a GPU to only one MM module in a temporal-multiplexing manner; this compromises training efficiency because a single module often fails to achieve high GPU utilization. To improve GPU utilization and enable efficient MM training, we propose deploying MMs in a temporal-spatial multiplexing manner, allowing multiple MM modules to colocate on a GPU with well-controlled resource quotas. In this paper, we propose Apollo, an efficient MM training system that applies temporal-spatial multiplexing. We first develop a flexible and lightweight execution engine that supports MM training with arbitrary resource quotas, and then build a comprehensive and accurate performance model to estimate module execution time under different allocation plans. With the performance model, we further adopt effective heuristics to derive high-quality MM deployment plans efficiently. Testbed experiments confirm that Apollo effectively improves the training efficiency of popular MMs, with a training speedup of up to 1.31x.
arXiv:2605.10759v2 Announce Type: replace
Abstract: Diffusion and flow-matching models scale because pretraining is supervised regression: a clean sample is noised analytically, and a model regresses against a closed-form target. RL post-training aligns the model with a reward. In image generation, this makes samples compose objects correctly, render text legibly, and match human preferences. Existing methods rely on costly SDE rollouts, reward gradients, or surrogate losses, sacrificing pretraining's regression structure. We show that the structure extends to RL post-training. Under KL-regularized reward maximization, the optimal generative process tilts the clean-endpoint distribution towards samples with higher reward and leaves the noising law unchanged. Combining this with the adjoint-matching optimality condition and a REINFORCE identity, we derive Reinforce Adjoint Matching (RAM): a consistency loss that corrects the pretraining target with the reward. At each step, we draw a clean endpoint from the current model, evaluate its reward, noise it as in pretraining, and regress. No SDE rollouts, backward adjoint sweeps, or reward gradients are required. Like the pretraining objective, RAM is simple and scales. On Stable Diffusion 3.5M, RAM achieves the highest reward on composability, text rendering, and human preference, reaching Flow-GRPO's peak reward in up to $50\times$ fewer training steps.
arXiv:2601.19300v2 Announce Type: replace
Abstract: We introduce contextual queueing bandits, a new context-aware framework for scheduling while simultaneously learning unknown service rates. Individual jobs carry heterogeneous contextual features, based on which the agent chooses a job and matches it with a server to maximize the departure rate. The service/departure rate is governed by a logistic model of the contextual feature with an unknown server-specific parameter. To evaluate the performance of a policy, we consider queue length regret, defined as the difference in queue length between the policy and the optimal policy. The main challenge in the analysis is that the lists of remaining job features in the queue may differ under our policy versus the optimal policy for a given time step, since they may process jobs in different orders. To address this, we propose the idea of policy-switching queues equipped with a sophisticated coupling argument. This leads to a novel queue length regret decomposition framework, allowing us to understand the short-term effect of choosing a suboptimal job-server pair and its long-term effect on queue state differences. We show that our algorithm, CQB-$\varepsilon$, achieves a regret upper bound of $\widetilde{\mathcal{O}}(T^{-1/4})$. We also consider the setting of adversarially chosen contexts, for which our second algorithm, CQB-Opt, achieves a regret upper bound of $\mathcal{O}(\log^2 T)$. Lastly, we provide experimental results that validate our theoretical findings.
arXiv:2605.18626v1 Announce Type: new
Abstract: Man-made and natural disruptions such as planned constructions on roads, suspensions of bridges, and blocked roads by trees/mudslides/floods can often create obstacles that separate two connected regions. As a result, the traveling and reachability of agents from their respective regions to other regions can be affected. To minimize the impact of the obstacles and maintain agent accessibility, we initiate the problem of constructing a new pathway (e.g., a detour or new bridge) connecting the regions disconnected by obstacles from the mechanism design perspective. In the problem, each agent in their region has a private location and is required to access the other region. The cost of an agent is the distance from their location to the other region via the pathway. Our goal is to design strategyproof mechanisms that elicit truthful locations from the agents and approximately optimize the social or maximum cost of agents by determining locations in the regions for building a pathway. We provide a characterization of all strategyproof and anonymous mechanisms. For the social and maximum costs, we provide upper and lower bounds on the approximation ratios of strategyproof mechanisms.
arXiv:2601.09722v2 Announce Type: replace
Abstract: In this work, we present an annotation framework that demonstrates how a multilingual LLM pretrained on a large corpus can be used as a teacher model to distill the expert knowledge needed for tagging medical texts in Polish. This work is part of a larger project called ADMEDVOICE, within which we collected an extensive corpus of medical texts representing five clinical categories - Radiology, Oncology, Cardiology, Hypertension, and Pathology. Using this data, we had to develop a multi-class classifier, but the fundamental problem turned out to be the lack of resources for annotating an adequate number of texts. Therefore, in our solution, we used the multilingual Llama3.1 model to annotate an extensive corpus of medical texts in Polish. Using our limited annotation resources, we verified only a portion of these labels, creating a test set from them. The data annotated in this way were then used for training and validation of 3 different types of classifiers based on the BERT architecture - the distilled DistilBERT model, BioBERT fine-tuned on medical data, and HerBERT fine-tuned on the Polish language corpus. Among the models we trained, the DistilBERT model achieved the best results, reaching an F1 score > 0.80 for each clinical category and an F1 score > 0.93 for 3 of them. In this way, we obtained a series of highly effective classifiers that represent an alternative to large language models, due to their nearly 500 times smaller size, 300 times lower GPU VRAM consumption, and several hundred times faster inference.
arXiv:2605.18654v1 Announce Type: new
Abstract: A fraud scorer needs to answer in under 2 ms. The best tabular foundation models (TFMs) take 151-1,275 ms on GPU. We close this gap by distilling the TFM offline into an XGBoost or CatBoost student that runs natively on CPU. The central obstacle is specific to in-context learning (ICL) teachers: they leak labels when scoring their own training set, so the soft targets collapse to near-one-hot vectors with no inter-class structure left to distill. Stratified out-of-fold (OOF) teacher labeling prevents this. Across 153 classification datasets drawn from TALENT, OpenML-CC18, TabZilla, and TabArena, distilling TabICLv2 into XGBoost gives 0.882 macro-mean AUC (96.5% of teacher AUC) at 1.9 ms on CPU, a 38x to 860x speedup across teacher-student pairs with a statistically significant edge over a tuned CatBoost baseline (Wilcoxon p = 0.0008; 51% win rate). Four further findings: teacher rank transfers exactly to student rank; gains concentrate on low-dimensional data (< 21 features: +0.011 over CatBoost vs. >21 features: +0.001); multi-teacher averaging helps MLP students (+0.006, p = 0.003) but adds less than 0.001 for tree students; and on high-dimensional tasks where the teacher itself trails CatBoost, distillation makes things worse rather than better. The full pipeline is open-sourced as part of the TabTune library.
arXiv:2605.17648v1 Announce Type: new
Abstract: Generative recommendation treats next-item prediction as autoregressive item-identifier generation. Specifically, items are encoded as semantic identifiers (SIDs), which are short coarse-to-fine token sequences whose early tokens capture broad semantics and later tokens refine them. Recent work augments this paradigm with reasoning traces and optimizes them via reinforcement learning with verifiable rewards, typically outcome-reward algorithm with exact-match feedback on the generated SID. However, in large-catalog recommendation, exact-match feedback on the generated SID only reports whether the final item is correct; when a generated SID mismatches, outcome-reward cannot identify which SID-token prediction caused the mismatch and may penalize matched SID-token positions together with the mismatched position. We identify that the natural unit of credit assignment in this setting is a single reasoning step (one thinking block paired with one SID token). We instantiate this idea in SAPO (Step-Aligned Policy Optimization): rather than broadcasting one advantage to the whole response, SAPO computes a separate group-relative advantage for each reasoning step and applies it only to the corresponding thinking block and SID token. Across three real-world recommendation datasets, SAPO stabilizes reinforcement-learning training and consistently improves over existing generative recommendation baselines, with the largest gains where sparse exact-match feedback makes reasoning-step credit assignment important. Our results suggest that reinforcement-learning objectives for structured generation should mirror the decoder's own decomposition of the output.
arXiv:2605.18750v1 Announce Type: new
Abstract: Pipeline parallelism is a key technique for scaling large-model training, but modern workloads exhibit runtime variability in computation and communication. Existing pipeline systems typically consume static, profiled, or adaptively generated schedules as pre-committed execution orders. When realized task readiness diverges from the pre-committed order, stages may wait for not-yet-ready work even though other executable work is available, creating stage misalignment, idle bubbles, and reduced utilization.
We present Runtime-Readiness-First Pipeline (RRFP), a readiness-driven runtime for pipeline-parallel training. RRFP changes how schedules are consumed at runtime: instead of treating a schedule as a sequence that stages must wait to follow, it treats the schedule as a non-binding hint order for ranking currently ready work. To support this model, RRFP combines message-driven asynchronous communication, lightweight tensor-parallel coordination for collective consistency, and ready-set arbitration for low-overhead dispatch.
We implement RRFP in a Megatron-based training framework and evaluate it on language-only and multimodal workloads at up to 128 GPUs. RRFP improves over fixed-order pipeline baselines across all settings. Using the BFW hint, RRFP achieves up to 1.77$\times$ speedup on language-only workloads and up to 2.77$\times$ on multimodal workloads. In cross-framework comparisons, RRFP with the default BF hint outperforms the faster available external system by up to 1.84$\times$ while preserving training correctness.
arXiv:2605.16357v1 Announce Type: cross
Abstract: WiFi fingerprint-based indoor localization has been widely studied, but most existing approaches focus on absolute positioning and rely on dense coordinate annotations, which are costly to obtain at scale. In this paper, we study a fundamentally different problem: relative localization, where the goal is to directly estimate the displacement between two WiFi fingerprint traces without predicting their absolute positions. To reduce annotation overhead, we adopt weak supervision in the form of stepwise motion vectors obtained from inertial sensing. We propose Intersection Pathway (IP), a cross-modal learning framework that aligns fingerprint traces (f-traces) and displacement traces (d-traces) in a shared latent space. The key idea is to enforce an additive structure in the latent space, such that latent addition and subtraction correspond to physical motion composition, enabling direct relative-displacement inference. Experiments on a synthesized dataset derived from real measurements demonstrate that the proposed method learns displacement-aware WiFi representations and achieves accurate relative localization across varying displacement ranges. Furthermore, the learned model can be extended to few-shot absolute localization with sparse anchors.
arXiv:2605.11480v2 Announce Type: replace
Abstract: Reward fine-tuning has become a common approach for aligning pretrained diffusion and flow models with human preferences in text-to-image generation. Among reward-gradient-based methods, Adjoint Matching (AM) provides a principled formulation by casting reward fine-tuning as a stochastic optimal control (SOC) problem. However, AM inevitably requires a substantial computational cost: it requires (i) stochastic simulation of full generative trajectories under memoryless dynamics, resulting in a large number of function evaluations, and (ii) backward ODE simulation of the adjoint state along each sampled trajectory. In this work, we observe that both bottlenecks are closely tied to the \textit{non-trivial base drift} inherited from the pretrained model. Motivated by this observation, we propose \textbf{Efficient Adjoint Matching (EAM)}, which substantially improves training efficiency by reformulating the SOC problem with a \textit{linear base drift} and a correspondingly modified \textit{terminal cost}. This reformulation removes both sources of inefficiency; it enables training-time sampling with a few-step deterministic ODE solver and yields a closed-form adjoint solution that eliminates backward adjoint simulation. On standard text-to-image reward fine-tuning benchmarks, EAM converges up to 4x faster than AM and matches or surpasses it across various metrics including PickScore, ImageReward, HPSv2.1, CLIPScore and Aesthetics.
arXiv:2605.11518v2 Announce Type: replace
Abstract: Effectively configuring scalable large language model (LLM) experiments, spanning architecture design, hyperparameter tuning, and beyond, is crucial for advancing LLM research, as poor configuration choices can waste substantial computational resources and prevent models from realizing their full potential. Prior automated methods are designed for low-cost settings where repeated trial and error is feasible, but scalable LLM experiments are too expensive for such extensive iteration. To our knowledge, no work has addressed the automation of high-cost LLM experiment configurations, leaving this problem labor-intensive and dependent on expert intuition. Motivated by this gap, we propose AutoLLMResearch, an agentic framework that mimics how human researchers learn generalizable principles from low-fidelity experiments and extrapolate to efficiently identify promising configurations in expensive LLM settings. The core challenge is how to enable an agent to learn, through interaction with a multi-fidelity experimental environment that captures the structure of the LLM configuration landscape. To achieve this, we propose a systematic framework with two key components: 1) LLMConfig-Gym, a multi-fidelity environment encompassing four critical LLM experiment tasks, supported by over one million GPU hours of verifiable experiment outcomes; 2) A structured training pipeline that formulates configuration research as a long-horizon Markov Decision Process and accordingly incentivizes cross-fidelity extrapolation reasoning. Extensive evaluation against diverse strong baselines on held-out experiments demonstrates the effectiveness, generalization, and interpretability of our framework, supporting its potential as a practical and general solution for scalable real-world LLM experiment automation.
arXiv:2605.18323v1 Announce Type: new
Abstract: The finite-length performance of spatially-coupled low-density parity-check (SC-LDPC) codes is strongly affected by short cycle configurations and the harmful structures induced by them. This paper studies SC-LDPC code design directly at the protograph level, where the design variables are the edge-spreading assignments specified by the partition matrix. In contrast to CLLL/Moser--Tardos based constructive frameworks for QC-SC-LDPC codes, we focus on sharper nonconstructive existence and counting bounds. By encoding cycle-activation conditions as polynomial vanishing constraints over finite grids, we apply the Combinatorial Nullstellensatz to derive sufficient memory conditions for eliminating prescribed cycle-induced harmful structures. For fully connected $(\gamma,\kappa)$ base graphs, the resulting bounds explicitly characterize the memory required to destroy all $4$-cycles as well as all $4$- and $6$-cycles, and for fixed $\gamma$, they are asymptotically tight up to a constant factor compared with known lower bounds. We further apply the Alon--F\"uredi theorem to obtain lower bounds on the number of feasible edge-spreading assignments, including an explicit counting bound for assignments that eliminate all $4$-cycles and hence yield girth at least six. These results provide a refined algebraic-combinatorial characterization of the feasible design space for high-memory SC-LDPC codes, although no corresponding construction algorithm is provided.