Academic Intelligence · Curated Daily

Explore the Frontier of Global Academia

AcademicHub aggregates real-time literature from top journals and preprint platforms. Build your personal research radar and let large language models compile cross-disciplinary analysis briefings automatically.

01.
arXiv (CS.AI) 2026-06-19

Sovereign Execution Brokers: Enforcing Certificate-Bound Authority in Agentic Control Planes

arXiv:2606.20520v1 Announce Type: cross Abstract: Autonomous agents are increasingly connected to cloud, deployment, and data-control workflows, but production mutation authority should not reside inside non-deterministic reasoning processes. Existing access-control mechanisms authorize identities, while assurance layers certify proposed actions; neither alone provides a mandatory enforcement point for certified authority at the moment of mutation. This paper introduces the Sovereign Execution Broker (SEB), a runtime enforcement boundary for certificate-bound agentic infrastructure. SEB consumes certificates issued by the Sovereign Assurance Boundary (SAB), verifies that the requested mutation matches the certified execution contract, checks validity windows, policy epochs, revocation epochs, and live-state drift, mints scoped execution identity, invokes infrastructure APIs, and records signed decision and outcome records. By separating proposal, admission, and execution, SEB turns certified authority into a short-lived, revocable, auditable runtime capability, provided that production mutation APIs reject non-broker identities. We present the SEB execution model, certificate and replay-verification predicates, scoped identity semantics, bypass-prevention deployment patterns, failure behavior, and a concrete prototype implementation. We evaluate the prototype on AWS and Kubernetes clusters, measuring latency overheads, revocation propagation, drift detection, and security under fault injection.

02.
arXiv (CS.AI) 2026-06-17

Small Initialization Matters for Large Language Models

arXiv:2606.17945v1 Announce Type: new Abstract: Large language models provide a tractable system for asking how intelligence itself emerges, rather than only how LLMs can be engineered. Although progress is usually attributed to scale, data and architecture, we show that parameter initialization is a gene-like determinant of training and, in particular, of model capacity. Reducing the initialization scale consistently improves pretraining, with the largest gains on reasoning-demanding tasks. We identify two widely used empirical settings that restrain the advantage of small initialization, and show how relaxing them restores favorable scaling. We further uncover a critical initialization that balances the reasoning and training. Mechanistically, small initialization drives a distinct developmental trajectory: parameters first condense into low-complexity structures and later expand into richer representations, giving concrete form to the idea that compression is intelligence. Token-level analyses show that the gains concentrate on non-trivial, context-constrained predictions rather than all tokens uniformly. These results motivate a simple $\gamma$-initialization rule: expose initialization rage as an explicit knob and use small initialization by default, an almost cost-free intervention that improves pretraining and strengthens reasoning across model scales.

03.
arXiv (CS.CL) 2026-06-12

Structuring The Future: Diffusion LLM Speculative Decoding via Calibrated Draft Graphs

Diffusion LLMs (dLLMs) have recently emerged as a powerful alternative to autoregressive LLMs (AR-LLMs) with the potential to operate at significantly higher token-generation rates. To unlock this potential, we present Spiffy, a speculative decoding algorithm to accelerate dLLM inference while provably preserving the model's output distribution. This work addresses the unique challenges involved in applying ideas from speculative decoding of AR-LLMs to dLLMs. Spiffy performs auto-speculation to eliminate the overheads of an independent draft model, structuring draft states in the form of a novel directed draft graph to take advantage of the bidirectional, blockwise nature of dLLM generation. These draft graphs are calibrated offline to maximize acceptance rates and are dynamically pruned during inference for improved computational efficiency. We present a detailed formulation of Spiffy and demonstrate its ability to accelerate LLaDA, Dream, and SDAR models in combination with KV caching and threshold-based dynamic unmasking leading to up to $8.6\times$ reduction in model inferences and $6.3\times$ acceleration in token rate.

04.
arXiv (quant-ph) 2026-06-24

Universality beyond the Kibble-Zurek mechanism in the condensation of coherently coupled Bose gases

arXiv:2606.24864v1 Announce Type: cross Abstract: We study the universal spatial statistics of point-like topological defects formed during the nonequilibrium condensation of a coherently coupled Bose gas using the stochastic projected Gross-Pitaevskii equation. The symmetry-breaking transition is driven by a linear quench of the chemical potential, leading to stochastic vortex nucleation in the individual condensate components. When the two components are considered together, these elementary defects may combine across components to emerge as composite topological defects known as full quantum vortices. Beyond the mean defect density predicted by the Kibble-Zurek mechanism (KZM), we investigate the spatial organization of both the elementary and composite defects and show that their positions are well described by a Poisson point process, revealing a universal stochastic geometry. This universality is further described through Voronoi tessellation, whose cell-area statistics follow Poisson-Voronoi predictions. We also introduce the spatial form factor for characterizing the vortex configurations and demonstrate the emergence of a characteristic dip-ramp-plateau structure. Our results establish universal stochastic geometry of topological defects beyond conventional Kibble-Zurek scaling and identify it as a fundamental feature of nonequilibrium condensation in coherently coupled Bose gases.

05.
arXiv (math.PR) 2026-06-11

On the $d$-rigidity phase transition in random graphs

Authors:

arXiv:2605.25711v2 Announce Type: replace-cross Abstract: We study generic $d$-dimensional rigidity in sparse random graphs. Our main result is that for every $d\ge 2$, the Erdős–Rényi random graph $G\sim G(n,c/n)$ undergoes a $d$-rigidity phase transition at the known, explicit, $d$-orientability threshold $c_d$: If $cc_d$, then $G$ is a.a.s. not independent in the generic $d$-rigidity matroid, and we give a sharp asymptotic estimate for its rank. In addition, the $d$-rigidity closure of $G$ has a giant clique of linear size, which contains all but at most $o(n)$ vertices of the $((d+1)+d)$-core of the graph. More generally, we compute, up to a $1+o(1)$ factor, the generic $d$-rigidity rank of random graphs with a given degree distribution. For example, we show that the uniform $n$-vertex $k$-regular graph a.a.s. has rank $\min(k/2,d)n+o(n).$ Our approach is to estimate the rigidity rank of a random graph from its Galton–Watson local weak limit, using a parameter that we call local flexibility.

06.
arXiv (CS.CL) 2026-06-15

DLawBench: Evaluating LLMs Through Multi-Turn Legal Consultation

Lawyer-client consultation is a critical starting point for legal services. Effective legal assistance hinges on eliciting sufficient and truthful information from clients in order to devise strategies that best protect their interests. This task requires Large Language Models (LLMs) not only to perform robust legal reasoning, but also to strategically elicit material facts through multi-turn interactions and effectively guide clients with diverse personalities. Yet existing legal benchmarks overlook this interactive capability. To fill this gap, we introduce DLawBench, a diagnostic benchmark for real-world legal consultation. Drawing on realistic client behavior, we characterize lawyer-client interactions into four types: Cooperative, Dependent, Withdrawn, and Adversarial. Using dialogues grounded in real cases, DLawBench evaluates whether LLMs can effectively conduct legal consultation under realistic conditions. DLawBench comprises 461 cases from Chinese and U.S. law, 5,532 paired fact entries, 3,411 inquiry rubrics, and 3,348 issue-resolution rubrics, and evaluates 26 representative LLMs. Systematic experiments show substantial headroom: the best-performing model, GPT-5.5, achieves only 0.562 on consultation-grounded legal reasoning. More importantly, DLawBench exposes both sycophancy in legal consultation and a paradox: models perform worse when clients need guidance most.

07.
arXiv (CS.AI) 2026-06-19

Controlled Comparison of Machine Learning Models for Fault Classification and Localization in Power System Protection

arXiv:2510.00831v2 Announce Type: replace Abstract: The increasing complexity of modern power systems, driven by the integration of inverter-based and distributed energy resources, challenges the reliability of conventional protection schemes and motivates the use of machine learning for protection tasks. However, published results are often difficult to compare because datasets, sensing assumptions, and decision horizons vary across studies. This paper presents a controlled comparison of machine learning models for fault classification (FC) and fault localization (FL) under identical sensing, timing, and validation conditions on a common electromagnetic transient dataset, using decision windows of 10-50 ms to reflect protection-relevant time scales. For FC, the best-performing nonlinear models achieve F1 scores above 0.98 already at 10 ms, while lower-capacity models degrade at shorter horizons but improve with longer windows, indicating that relevant fault-type information is already present in the earliest transient. For FL, the top-performing models reach a stable localization error of about 10 % of normalized line length across all evaluated horizons, while weaker models form a clearly separated second performance tier. Line-resolved analysis shows that localization accuracy varies across grid segments, indicating topology-dependent difficulty rather than insufficient temporal context alone. These findings provide a controlled reference for comparing machine learning models across two protection tasks with fundamentally different information requirements.

08.
arXiv (CS.LG) 2026-06-11

LakeFM: Toward a Foundation Model for Aquatic Ecosystems Using Irregular Multivariate Multi-depth Time Series Data

arXiv:2606.11268v1 Announce Type: new Abstract: Understanding and forecasting lake dynamics is critical for monitoring water quality and ecosystem health across lakes and reservoirs. While machine learning methods have been recently applied to ecological time-series data, existing works assume regular sampling in time and depth, and struggle to generalize across lakes with heterogeneous variables, depths, and observation patterns. To address these limitations, we introduce \textsc{LakeFM}, a foundation model for aquatic systems, pre-trained on large-scale ecological datasets comprising both simulated and observed lakes. Through extensive empirical evaluation, we show that \textsc{LakeFM} learns meaningful representations spanning broader lake-level characteristics, and achieves competitive or often superior-forecasting performance compared to existing time-series foundation and non-foundation models, while producing physically plausible predictions consistent with real-world lake dynamics.

09.
arXiv (CS.CV) 2026-06-16

OmniOPSD: Rationale-Privileged On-Policy Self-Distillation for Affective Computing

Reinforcement learning for multimodal large language models (MLLMs) is often hindered by severe reward sparsity in complex reasoning tasks. This challenge is particularly pronounced in human-centered scenarios involving states, emotions, intentions, and behaviors, where heterogeneous multimodal signals and subjective human factors make high-quality chain-of-thought (CoT) annotations expensive and difficult to obtain. Although many multimodal datasets provide expert-annotated ground-truth labels, directly using these labels for supervised fine-tuning may encourage shortcut learning in multimodal perception and provides limited transparency for safety-critical human–AI interaction. To address these limitations, we propose OmniOPSD, a Rationale-Privileged On-Policy Self-Distillation framework that uses frontier-generated rationales as teacher-side privileged evidence rather than student imitation targets. OmniOPSD uses frontier-generated evidence-aware rationales only as training-time privileged evidence context for a local teacher. The student samples its own rollout from the original multimodal input, while the rationale-privileged teacher scores the same tokens and provides dense token-level supervision. Thus, the student learns on its own trajectory distribution without directly imitating frontier-model completions, and inference requires no labels, rationales, CoT annotations, or closed-source model access. Experiments on MER-UniBench show that OmniOPSD achieves state-of-the-art performance with an average score of $84.19$, and ablations further support the value of rationale-privileged teacher guidance.

10.
bioRxiv (Bioinfo) 2026-06-17

MetaHarmonizer: robust biomedical metadata harmonization and a contamination control for inflated LLM performance on public benchmarks

Public biomedical repositories hold substantial reuse potential, but inconsistent metadata routinely blocks integration across studies. Recent LLM-based harmonization approaches address scale but suffer from non-determinism, hallucinated ontology terms, and, in their highest-accuracy configurations, dependence on proprietary APIs or labeled fine-tuning data. A more fundamental concern is that LLM accuracies on widely-used public benchmarks may substantially inflate transferable capability: under a contamination-controlled evaluation protocol we developed, the apparent LLM-only advantage on the GDC schema-mapping benchmark is inverted, and three out of five LLMs recover 80 -100% of GDC identifiers from zero-schema context, suggesting direct memorization. Building on this insight, we present MetaHarmonizer, an automated metadata harmonization system designed to be robust by construction: SchemaMapper aligns attribute names across schemas, and OntologyMapper standardizes values to controlled vocabularies. Both modules implement a multi-stage cascade that escalates to more resource-intensive methods only when earlier stages fall short, with all candidates grounded in pre-defined controlled vocabularies to preclude hallucinated outputs and LLMs used only as bounded preprocessing components rather than inference-time dependencies. On the GDC schema-matching benchmark, SchemaMapper with the deployment-optimized LLM-generated alias dictionary achieved 71.6% Top-1 accuracy and the higher Recall@GT than Magneto bipartite variants, recovering significantly more ground-truth mappings; with the best performing alias dictionary, it reached the highest Top-1/Top-5/Recall@GT, and also matched the best Magneto reranker (fine-tuned LLM-reranker) on MRR; and it also outperforms LLM-only performance under contamination-controlled conditions. On four EFO benchmarks, OntologyMapper achieved 77.9 - 95.5% Top-1 accuracy, outperforming text2term by up to 16.4 pp and direct LLM inference (against the smaller corpus) by 19.2 pp because memorization is not a viable shortcut for this task. Across both modules, calibrated confidence scores separate correct from incorrect predictions (AUC 0.73 - 0.94), enabling principled human-in-the-loop triage. Inference is fully local, deterministic, and computationally efficient - seconds on schema mapping and under a minute for ontology mapping of up to ~7,000 terms against the pre-indexed 33,230-term corpus. Released as a Python package with a domain-agnostic architecture, MetaHarmonizer provides a scalable foundation for improving the FAIRness of biomedical data and enabling cross-study integration, alongside an evaluation methodology applicable to any LLM-augmented bioinformatics benchmark built on public benchmarks.

11.
arXiv (math.PR) 2026-06-11

Large deviations for marked sparse random graphs with applications to interacting diffusions

arXiv:2204.08789v2 Announce Type: replace Abstract: We consider the empirical neighborhood distribution of marked sparse Erdős-Rényi random graphs, obtained by decorating edges and vertices of a sparse Erdős-Rényi random graph with i.i.d. random elements taking values on Polish spaces. We prove that the empirical neighborhood distribution of this model satisfies a large deviation principle in the framework of local weak convergence. We rely on the concept of BC-entropy introduced by Delgosha and Anantharam~(2019) which is inspired on the previous work by Bordenave and Caputo~(2015). Our main technical contribution is an approximation result that allows one to pass from graph with marks in discrete spaces to marks in general Polish spaces. As an application of the results developed here, we prove a large deviation principle for interacting diffusions driven by gradient evolution and defined on top of sparse Erdős-Rényi random graphs. In particular, our results apply for the stochastic Kuramoto model. We obtain analogous results for the sparse uniform random graph with given number of edges.

12.
arXiv (CS.AI) 2026-06-16

JADE: Expert-Grounded Dynamic Evaluation for Open-Ended Professional Tasks

arXiv:2602.06486v2 Announce Type: replace Abstract: Evaluating agentic AI on open-ended professional tasks faces a fundamental dilemma between rigor and flexibility. Static rubrics provide rigorous, reproducible assessment but fail to accommodate diverse valid response strategies, while LLM-as-a-judge approaches adapt to individual responses yet suffer from instability and bias. Human experts address this dilemma by combining domain-grounded principles with dynamic, claim-level assessment. Inspired by this process, we propose JADE, a two-layer evaluation framework. Layer 1 encodes expert knowledge as a predefined set of evaluation skills, providing stable evaluation criteria. Layer 2 performs report-specific, claim-level evaluation to flexibly assess diverse reasoning strategies, with evidence-dependency gating to invalidate conclusions built on refuted claims. Experiments on BizBench show that JADE improves evaluation stability and reveals critical agent failure modes missed by holistic LLM-based evaluators. We further demonstrate strong alignment with expert-authored rubrics and effective transfer to HealthBench and DR.BENCH, covering medical and 10-domain professional evaluation settings. Code and data are available at https://github.com/smiling-world/JADE.

13.
arXiv (CS.CV) 2026-06-24

Does it matter which Gaussians you pick in 4D Gaussian streaming?

Anchor-driven 4D Gaussian streaming methods such as Instant Gaussian Stream (IGS) update a dynamic scene each frame from a compact set of Gaussian anchors, chosen by default with Farthest Point Sampling (FPS) at a fixed budget of $8{,}192$. Because these anchors act as control points that drive the whole scene through linear blend skinning, the rule used to choose them ought to affect reconstruction quality. We test this by holding the IGS pipeline fixed and changing only the sampler, comparing FPS, random, uniform, an opacity-scale heuristic, and a learned policy across budgets and refinement settings on N3DV and MeetingRoom. At deployment budgets the sampler has no measurable effect: a cheap random or uniform sampler at $4{,}096$ anchors matches FPS@8192 within measurement error, the default budget is over-provisioned, and the result holds on a second backbone (3DGStream). The learned policy is mixed rather than consistently better: it can improve the N3DV validation set at tight budgets, but does not give a stable cross-dataset rule, and selection is never the bottleneck because refinement dominates runtime. We will release our full sweep and evaluation protocol as a sampler benchmark.

14.
arXiv (CS.LG) 2026-06-18

Reliable Neural-Codec Text-to-Speech by ASR Self-Verification and Distillation: Near-Zero Catastrophic Failures Across Models and Codecs

arXiv:2606.18323v1 Announce Type: cross Abstract: Open autoregressive neural-codec text-to-speech (TTS) models sound excellent on typical inputs yet suffer stochastic catastrophic failures: on a meaningful fraction of utterances they emit silence, terminate early, or collapse into repetitive or hallucinated content. We show this failure mode is cheap to remove. Under a single format-robust metric (a catastrophic-failure rate via an ASR round-trip), best-of-N ASR self-verification drives failures to near-zero: no observed failures remain by N=2 on a standard corpus (LibriSpeech) and by N=4 on a hard prompt set. This is not an artifact of one model: the reduction replicates across four open codec-TTS systems and three neural codecs (XCodec2, SNAC, Mimi), reaching the near-zero floor by N=2 on three of the four. We then make the fix free at inference time by distilling the self-verified behaviour into the model, which recovers much of the robustness in single-shot decoding, closing ~52-58% of the failure mass on hard inputs at no test-time cost. The distillation gain concentrates where it is needed (hard inputs); on already-reliable prose there is no headroom and no detectable change. A controlled comparison adds a clean negative: offline direct preference optimization (DPO/IPO) does not beat plain supervised distillation, and an online iterative variant is promising but not statistically separable at our evaluation size. We report honestly the one model that resists (a larger Llasa where scale did not obviously help) and a rare-word capability ceiling that no self-distillation method overcomes

15.
Science (Express) 2026-04-16

Protein-templated synthesis of dinucleotide repeat DNA by an antiphage reverse transcriptase | Science

Authors: Unknown Author

Defense-associated reverse transcriptases (DRTs) are widespread bacterial anti-phage systems that use unconventional mechanisms of polynucleotide synthesis. We show that DRT3, which comprises two distinct RTs (Drt3a and Drt3b) and a noncoding RNA (ncRNA), synthesizes alternating poly(GT/AC) double-stranded DNA. Cryo–electron microscopy structures at 2.6 Å resolution reveal a D3-symmetric 6:6:6 complex of Drt3a, Drt3b, and ncRNA. Drt3a produces the poly(GT) strand using a conserved ACACAC template within the ncRNA. Notably, Drt3b synthesizes a complementary, protein-primed poly(AC) strand in the complete absence of a nucleic acid template, using conserved active site residues specific to Drt3b to enforce precise base alternation. These findings expand the functional landscape of nucleic acid polymerases, revealing a protein-templated mechanism for sequence-specific DNA synthesis.

16.
arXiv (CS.CV) 2026-06-18

Reasoning as Intersection: Consensus-Frame Alignment for Visual Focus in Video-MLLMs

Reinforcement learning has improved the reasoning ability of large language models, but applying outcome-only rewards to video multimodal large language models (Video-MLLMs) provides limited guidance on which visual evidence should support the answer. Inspired by multisensory integration, where consistent cues can enhance the salience and reliability of perceptual estimates, we introduce Consensus Frame GRPO (CF-GRPO), a temporal-annotation-free process-level reward framework for evidence-aware video reasoning. CF-GRPO constructs a consensus frame prior from intrinsic video cues, including temporal coverage, scene-transition cues, and query-conditioned visual relevance. It then computes a model-side frame-use score from visual and response representations and optimizes their agreement through the Consensus Frame Reward (CFR). With salience-aware sparse aggregation and distribution sharpening, CFR provides a high-contrast reward signal without requiring human temporal annotations. Experiments show that VideoCFR achieves competitive performance across complex video reasoning benchmarks and improves several metrics over representative Video-MLLM and RL baselines, while the consensus prior provides an interpretable view of the evidence frames emphasized during training. The implementation is available at https://github.com/1Pansy/VideoCFR.

17.
arXiv (CS.LG) 2026-06-16

An Integrable Token Mixing Layer from the Generalized Yang Baxter Equation

arXiv:2606.15085v1 Announce Type: new Abstract: The YB Mixer is a sequence token mixing layer derived from free fermion and generalized Yang Baxter structures. It applies a core principle from integrable systems where a local algebraic constraint guarantees global computational stability. By using the Ising exchange algebra the mixer creates a free fermionic structure that acts as an exactly norm preserving orthogonal map. This algebra also produces commuting transfer matrices which allow inference to be order free and adaptable to any variable budget. To ensure the model can generalize to longer sequence lengths it uses a spectral circulant generator. This generator maintains the crucial orthogonal and commuting properties of the system. The result is a highly stable and mathematically grounded architecture for sequence processing.

18.
arXiv (CS.AI) 2026-06-15

Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models

arXiv:2606.07157v2 Announce Type: replace Abstract: Many efforts to ensure frontier AI models are safe rely on monitoring their chain-of-thought (CoT) reasoning. If models become able to perform sufficiently complex reasoning internally, without explicit thinking tokens, this would undermine such oversight. We measure how well frontier models reason without CoT across a suite of over 30,000 questions spanning 43 benchmarks in domains including math, coding, puzzles, causality, theory-of-mind, and strategic reasoning. To compare models against humans, we estimate the $50\%$-task-completion time horizon (TH): the human time required for tasks a model completes with $50\%$ success rate. We complement this with a $50\%$ reasoning token horizon: the minimum number of o3-mini reasoning tokens needed for tasks a model solves with $50\%$ success rate. We find that the no-CoT $50\%$ TH of frontier models has been doubling roughly every year over the past six years, with GPT-5.5's TH reaching over 3 minutes and reasoning token horizon exceeding 1,500 tokens. Our median estimates predict that frontier no-CoT THs could exceed 7 minutes by 2028, and 25 minutes by 2030, though these projections carry substantial uncertainty. We recommend frontier developers track this explicitly.

19.
arXiv (math.PR) 2026-06-12

Quenched and Annealed CLTs for the one-periodic Aztec diamond in random environment

arXiv:2510.11846v2 Announce Type: replace Abstract: We study the asymptotic behavior of random dimer coverings of the one-periodic Aztec diamond in random environment. We investigate quenched limit theorems for the height function and we extend annealed limit theorems that were recently studied in [arXiv:2507.08560]. We consider more general choices of random edge weights (independence is not assumed) and we distinguish two cases where the random edge weights satisfy the Central Limit Theorem (CLT) under different scalings. For both cases, we prove convergence to the Gaussian Free Field for the quenched fluctuations. For the annealed version, it had been shown in [arXiv:2507.08560], that Gaussian Free Field fluctuations can be dominated by the much larger fluctuations of the random environment. To access quenched fluctuations we analyze the Schur process with random parameters in a way that allows to prove the annealed CLT for the height function for non i.i.d. weights. We consider specific examples where we determine the asymptotic fluctuations.

20.
arXiv (CS.CV) 2026-06-16

Efficient Reinforcement for Visual-Textual Thinking with Discrete Diffusion Model

RL-based post-training has been widely adopted to enable interleaved visual and textual reasoning in unified multimodal models capable of both text and image generation. However, most existing approaches are built upon autoregressive (AR) unified models, which require full image regeneration during visual reasoning. In this work, we demonstrate that multimodal discrete diffusion models are effective alternatives to AR models for reinforcement learning in interleaved reasoning, owing to their ability to perform efficient visual rollouts via localized visual editing rather than full image-token regeneration. This reduces rollout computation during GRPO by 26.9\% compared to AR baselines, with minimal performance drop. Despite the improved efficiency, we find that joint reward assignment, which employs a shared reward signal across modalities, introduces cross-modal interference between unrelated image and text token sequences during RL updates. To address this issue, we propose factorized reward assignment, a strategy that assigns rewards independently to text and vision segments. With factorized reward assignment, our RL approach achieves an 11.2% improvement over joint reward assignment and a 38.04% improvement over the base model.

21.
arXiv (CS.CL) 2026-06-18

SFT Overtraining Predicts Rank Inversion via Entropy Collapse Under RLVR

The standard heuristic of selecting the SFT checkpoint with the highest pass@1 for GRPO can fail when SFT compresses the rollout distribution. For binary rewards, the expected within group advantage variance is $p(1{-}p)(g{-}1)/g$; when early GRPO drives $p$ below $p^*(g)$, most groups have identical rewards and provide no group relative signal. We study SFT depth ladders for Qwen2.5-Coder-3B and DeepSeek-Coder-6.7B. We test Qwen2.5-Coder-3B across five depths and three seeds, and DeepSeek-Coder-6.7B across four matched depths and three seeds. On Qwen, pre RL pass@1 rises with SFT depth, but peak GRPO pass@10 falls from $0.806$ to $0.481$ (3 seed mean, $n{=}20$); pre RL entropy is positively associated with the GRPO outcome ($\rho{=}{+}0.69$). On DeepSeek, pass@1 remains far above $p^*(8){=}0.083$, and GRPO outcomes compress rather than invert. A two stage diagnostic, combining pre RL entropy triage with an early GRPO entropy monitor, flags high risk checkpoints and can stop failing runs early. Simple KL to reference regularisation and label smoothing variants do not rescue the collapsed Qwen checkpoint in our setting, suggesting the failure is not a trivial GRPO hyperparameter artefact.

22.
arXiv (math.PR) 2026-06-24

Critical Erd{\H o}s-Rényi digraph: all eigenvectors away from zero are delocalized

arXiv:2606.24887v1 Announce Type: new Abstract: We consider the adjacency matrix of the directed Erd{\H o}s-Rényi graph. As long as the expected degree is larger than the logarithm of the number of vertices, the graph is connected, we show that all eigenvectors are completely delocalized. Below this critical scale, we prove eigenvector delocalization if the corresponding eigenvalue is away from zero. This contrasts the undirected or Hermitian setting, where large eigenvalues have localized eigenvectors [arXiv:2005.14180]. Our results also hold for sparse random matrices with independent entries, which can be viewed as weighted Erd{\H o}s-Rényi digraphs.

23.
arXiv (CS.CV) 2026-06-25

ASSCG: Just-Right Gating over Chattering for Fast-Slow LLM Planning in Autonomous Driving

Large language models (LLMs) can improve autonomous driving planning but are costly to query online, and existing fast-slow planners often rely on hand-designed triggering rules that either over-call the slow system or call it at the wrong times. We formulate slow-system invocation as a resource-aware sequential decision problem and propose the Adaptive Slow-System Control Gate (ASSCG), which makes frame-level Query/Cache/Drop decisions to refresh, reuse, or suppress slow guidance. ASSCG uses an RWKV backbone for efficient long-horizon gating and is trained with supervised fine-tuning followed by GRPO-style compute-aware reinforcement fine-tuning. We apply ASSCG to two different fast-slow architectures: (i) AsyncDriver on nuPlan Hard20 closed-loop evaluation, where ASSCG improves score to 67.28 (+2.28) while reducing average end-to-end inference latency by 60%; and (ii) a RecogDrive-based dual system that we build by replacing its original VLM-2B module with a lightweight ViT-based fast planner and adding an LLM slow planner, evaluated on NAVSIM, where ASSCG achieves 91.4 PDMS (+0.6) and increases average speed by 25%. The project page, including video visualizations and additional results, is available at https://williamxuanyu.github.io/asscg/.

24.
arXiv (CS.CL) 2026-06-15

Graph-based Target Back-Propagation for Context Adaptation in Multi-LLM Agentic Systems

Context adaptation automates prompt engineering in LLM-based systems by iteratively revising tunable prompts from task feedback, without modifying model weights. Extending this paradigm to multi-LLM agentic systems is crucial: existing methods suffer from inaccurate credit assignment and lack convergence guarantees. We propose Graph-based Target Back-Propagation (GTBP), a context adaptation framework for agentic workflows modeled as directed acyclic graphs. GTBP propagates local target outputs backward through the workflow graph and uses target–output discrepancies to guide a stage-wise prompt update mechanism. Theoretically, we show that GTBP's stage-wise prompt updates become stable over iterations, and that a sufficiently capable LLM optimizer can decrease the overall objective. Empirically, GTBP consistently outperforms strong baselines across three benchmarks while maintaining comparable computational cost.

25.
arXiv (math.PR) 2026-06-17

Optional Stopping for Superhedging Supermartingales

arXiv:2606.17452v1 Announce Type: new Abstract: Superhedging supermartingales, introduced by the authors in previous work, are non-probabilistic processes defined via subadditive outer integrals that carry a purely financial interpretation in terms of superhedging cost. Building on the Leinert-König theory of non-lattice integration, the present paper establishes several results that are classical in probability theory but whose non-probabilistic proofs require fundamentally new arguments: (i) a tower inequality for the conditional outer integral \overline{\sigma}_j applied at stopping times, reducing to equality when the integrand is conditionally integrable; (ii) three versions of Doob's optional stopping theorem, organised by the class of supermartingale and the range of the stopping times; and (iii) Dubins' upcrossing inequality in both finite- and infinite-time horizons. A key structural result, property (K)-a.e., identifies conditions under which the two superhedging operators \overline{\sigma}_j and \overline{I}_j coincide on non-negative functions, extending the scope of all preceding results to the positive operator \overline{I}_j. None of the proofs invoke classical measure-theoretic tools; in particular, (classical) integrability and measurability are not assumed. The analogues of classical stochastic results acquire a purely financial interpretation and, in this way, gain depth and generality by providing a context that is independent of any a priori probabilistic structure.