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.
PLOS Computational Biology 2026-06-15

Environmental “knees” and “wiggles” as strong stabilizers of species’ range limits set by interspecific competition

by Farshad Shirani, Benjamin G. Freeman Whether interspecific competition is a major contributing factor to setting species’ range limits has been debated for a long time. Theoretical studies have proposed that the interactions between interspecific competition and disruptive gene flow along an environmental gradient can halt range expansion of ecologically similar species where they meet. However, the stability of such range limits has not been well addressed. We use a deterministic mathematical model of adaptive range evolution over a continuous habitat to show that the range limits set by interspecific competition are unlikely to be evolutionarily stable if the environmental optima for fitness-related traits vary (almost) linearly in space. That is, in a linear environment without a dispersal barrier or a third (or more) species, the range borders formed between two competing species constantly move towards the weaker species. We demonstrate that environmental nonlinearities such as “knees” and “wiggles”—wherein an isolated sharp change or a step-like change occurs in the steepness of a trait optimum—can strongly stabilize competitively formed range limits. The stabilization mechanism relies on the contrast that such nonlinearities create in the level of disruptive gene flow to the peripheral population of each species, and succeeds when an additional process, such as Allee effects, prevents the establishment of an infinitesimal population in the presence of an abundant competitor. We show that the stability of the range limits at these nonlinearities is robust against moderate environmental disturbances. Whether strong disturbances such as rapid high-amplitude climate changes can destabilize such range limits depends on how the competitive dominance of the species changes across the nonlinearity. Therefore, our findings underscore the importance of assessing species’ competitive ability when predicting responses to climate change, and identify geographic regions where established range limits are likely to persist as well as regions where shifting limits may eventually stabilize.

02.
arXiv (CS.AI) 2026-06-16

Feature Attribution in Directed Acyclic Graphs Using Edge Intervention

arXiv:2606.15273v1 Announce Type: new Abstract: Shapley value-based feature attribution methods face challenges in scenarios involving complex feature interactions and causal relationships, even when a causal structure is provided. Existing methods typically adopt a node-centric view, attributing importance solely to individual features. Consequently, they often fail to simultaneously capture the externality and exogenous influence of features, leading to unreasonable interpretations. To overcome these limitations, we propose a novel feature attribution method called DAG-SHAP, which is based on edge intervention. DAG-SHAP treats each feature edge as an individual attribution object, ensuring that both externality and exogenous contributions of features are appropriately captured. Additionally, we introduce an approximation method for efficiently computing DAG-SHAP. Extensive experiments on both real and synthetic datasets validate the effectiveness of DAG-SHAP. Our code is available at https://github.com/ZJU-DIVER/DAG-SHAP.

03.
arXiv (CS.CV) 2026-06-18

The Market in the Model: Latent Diffusion as Neural Economy

Valuable critique of generative image models within visual culture and the humanities has emphasized the role of datasets in shaping the images they produce. Yet, close studies of the ideological positions embedded into the mechanism of the models have been neglected, leaving them imagined as "black boxes." In a bid to expand, rather than replace, dataset critique, this paper examines the mechanisms of the latent diffusion model in terms of the problems they were brought in to solve on behalf of computer vision engineers, and the decisions each component was tasked with automating. I interpret that ensemble through the histories of its parts and the theory of vision the system inscribes into every generated image. Drawing on Impett and Offert's notion of neural exchange value, I offer this analysis to argue that the model operates as a neural economy: a contained symbolic system that abstracts social communication into commensurable vectors as it transfers the social sphere into parcels for sale. Tracing the training and generation pipelines component by component reveals what each operation displaces, and how it further entrenches the logics of platform and attention economies over social communication. The paper warns that any critique fixated exclusively on copyright and commodity defenses risks reaffirming the very fetishism the model produces, and argues instead for centering social exchange.

04.
arXiv (CS.CV) 2026-06-16

Gen-VCoT: Generative Visual Chain-of-Thought Reasoning via Diffusion-Based RGB Intermediate Representations

Multimodal large language models (MLLMs) excel at visual reasoning but rely on text-based chain-of-thought (CoT), lacking interpretable visual intermediates. Existing methods use opaque tokens or external tools, missing key properties. We propose Gen-VCoT, a framework using expert vision models to generate RGB images as reasoning intermediates. It has three stages: visual grounding (SAM segmentation), geometric reasoning (Marigold depth maps), and semantic reasoning (Qwen2-VL integration). An adaptive router selects reasoning depth. Evaluations show Gen-VCoT improves spatial (25% better) and depth (50% better) questions, but may hurt simple factual queries. Text CoT outperforms visual intermediates on CLEVR (91.2% vs 62.5%), showing task-dependent optimal representations. Gen-VCoT establishes a new paradigm for interpretable multimodal reasoning.

05.
bioRxiv (Bioinfo) 2026-06-11

SPARK: A Systems-level Computational Framework for Reconstructing Transcriptomic State Organisation in Lung Adenocarcinoma

Lung adenocarcinoma (LUAD) exhibits substantial molecular heterogeneity, which complicates tumour stratification and limits the ability of mutation-centric models to capture tumour behaviour and predict patient outcomes. This study investigates whether coordinated transcriptomic programs can provide a systems-level representation of tumour states. Bulk RNA-sequencing data from the TCGA-LUAD cohort were analysed to reconstruct pathway-level transcriptomic organisation using a stability-optimised network framework (SPARK). This analysis identified eight transcriptomic modules representing coordinated biological processes active across tumours. Module activity scores were subsequently used to derive a composite Transcriptomic Risk Score through elastic-net Cox proportional hazards modelling. The resulting risk score showed a significant association with overall survival in the discovery cohort and improved prognostic discrimination beyond clinical variables. An independent evaluation in the CPTAC-LUAD cohort confirmed the prognostic signal and preserved risk stratification across patient groups. Unsupervised clustering of module activity further revealed three transcriptomic patient groups characterised by distinct biological programs, genomic alteration patterns, and survival outcomes. Single-cell analysis also demonstrated that the identified transcriptomic modules reflect coordinated organisation of the tumour-immune-stromal ecosystem across cellular compartments. Together, these findings suggest that LUAD heterogeneity can be organised into coordinated transcriptomic programs with measurable clinical relevance, providing a systems-level framework for representing tumour molecular states.

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

Theorem-Grounded Execution Ontologies for Interpretable Machine Reasoning

arXiv:2606.16010v1 Announce Type: cross Abstract: Large language models have achieved impressive performance on reasoning tasks spanning mathematics, science, programming, and commonsense inference. Despite these advances, their reasoning processes remain largely latent, making them difficult to interpret, verify, replay, debug, and transfer across domains. Existing approaches such as chain-of-thought, tree-of-thoughts, graph-of-thoughts, and tool-augmented reasoning expose intermediate reasoning artifacts but typically lack explicit execution semantics, formal state representations, and verifiable reasoning structures. We introduce Theorem-Grounded Execution Ontologies (TGEO), a framework that models reasoning as an executable state-transition process rather than a sequence of generated tokens. Given an input problem, TGEO identifies relevant theorem families, binds the problem to a domain ontology, discovers semantic objects, instantiates states and operators, constructs predicates and contracts, and synthesizes an executable reasoning graph. The resulting graph provides an interpretable, replayable, and auditable representation of reasoning in which every state transition, operator application, and validation step is explicitly represented. TGEO integrates five architectural components: (1) theorem-grounded reasoning priors, (2) executable ontologies, (3) operator-mediated state transitions, (4) predicate and contract-based execution validation, and (5) architectural auditing and failure localization. We evaluate TGEO on theorem-intensive reasoning tasks derived from mathematical benchmark domains and a curated Golden Execution Suite. Our findings demonstrate the value of executable reasoning representations for interpretable, verifiable, and reproducible AI reasoning systems.

07.
arXiv (quant-ph) 2026-06-15

Optimal Decoding of Small Codes by Density Matrix Propagation

arXiv:2606.14455v1 Announce Type: new Abstract: Accurate and efficient decoding is a crucial component for achieving fault-tolerant quantum computing. Realistic circuit-level noise introduces temporal correlations and degeneracy, making optimal (maximum-likelihood) decoding computationally intractable in general. As a result, practical decoders rely on heuristic approximations, and it is generally difficult to quantify how suboptimal they are, as this strongly depends on the code and noise model considered. In this work, we study the accuracy of practical decoding algorithms under circuit-level noise by comparing them against a maximum likelihood decoding benchmark. Our approach propagates the density matrix through the full memory experiment and computes the optimal decoding decision for each syndrome history. We introduce pruning techniques with rigorous bounds, allowing us to access larger numbers of syndrome-extraction rounds. We apply this framework to small instances of the repetition code and a cellular automaton code, and benchmark minimum-weight perfect matching (MWPM), belief propagation with ordered statistics decoding (BP+OSD), Tesseract, and Planar decoders against optimal decoding. While standard decoders remain close to optimal for the repetition code, we find significant deviations for the cellular automaton code, with BP+OSD deteriorating already in experimentally relevant noise regimes. Moreover, the pruning method developed here highlights that, at low physical error rates, only a narrow fraction of syndrome histories contributes significantly to the logical error rate.

08.
arXiv (CS.AI) 2026-06-19

The Hidden Evolution of Disguised Visual Context inside the VLM

arXiv:2606.20077v1 Announce Type: cross Abstract: Visual tokens enter Large Language Models (LLMs) as raw, foreign signals. How they are transformed into meaningful representations and interact with the language space depends entirely on the integration architecture. Whether by treating visual tokens as in-context prompts within the input sequence or injecting them directly into the LLM's intermediate layers. A controlled comparison and understanding of how these architectural choices affect visual information and its internal transformation to integrate with the LLM remains underexplored. We provide a fair comparison by evaluating in-context and layer-wise injection VLM integration paradigms under identical training conditions across single image, multi-image, and video benchmarks. In doing so, we uncover a hidden evolution where visual tokens enter the LLM as disguised visual context, raw representations lacking linguistic structure, but are progressively reshaped depending on the integration paradigm, each capturing fundamentally different frequency characteristics of the visual signal. We show that this evolution inside the LLM determines what visual features the VLM can utilize effectively, how visual representations align with the language space, and ultimately how each paradigm performs across different tasks. We further demonstrate that attention allocation alone is insufficient, and that performance is driven by the quality of visual representations at each layer.

09.
arXiv (CS.AI) 2026-06-19

VERITAS: Verifier-Guided Proof Search for Zero-Shot Formal Theorem Proving

arXiv:2606.19399v1 Announce Type: cross Abstract: LLM-based formal provers often collapse rich verifier signals (syntax errors, type mismatches, partial goal progress) into a binary pass/fail bit. We present VERITAS, a zero-shot framework that routes every verifier signal back into proof search through a two-phase protocol: Best-of-N sampling first, then a critic-guided MCTS pass that ingests Phase 1 failures as explicit negative examples. The protocol preserves every theorem solved by its own Phase 1 sweep, so Phase 2's additional solves are attributable to feedback-driven exploration. VERITAS reaches 40.6% on miniF2F (vs. an independently run Best-of-5 at 36.9%, Portfolio 26.2%) and 7.3% on VERITAS-CombiBench, a 55-theorem combinatorics benchmark we release on which Best-of-5 (1.8%) falls below Portfolio (3.6%), exposing that unguided sampling hurts when correct lemma names must be recovered iteratively from verifier feedback. Artifacts are available on GitHub.

10.
arXiv (CS.AI) 2026-06-12

Functional Cache Grafting: Robust and Rapid Code-Policy Synthesis for Embodied Agents

arXiv:2606.13097v1 Announce Type: cross Abstract: Code-writing large language models (CodeLLMs) generate executable code policies for embodied agents by translating natural language goals and environmental constraints into structured control programs. However, policy generation in open-domain embodied environments suffers from two fundamental limitations: (i) delayed decoding caused by repetitive prefill computation over long prompts, and (ii) limited robustness due to fully generative decoding, which often produces API mismatches, missing safety guards, and unstable control logic. To address these limitations, we present FCGraft, a Functional Cache Grafting framework. FCGraft maintains a library of function-level validated code skeletons and their associated prompt-level Transformer key-value (KV) caches, and synthesizes new policies by retrieving relevant functions and grafting their KV caches when a new task is provided. Given retrieved function caches, FCGraft performs cache grafting via stitching, which composes cached function segments into a composite policy, and patching, which locally adapts only the necessary code regions to satisfy task-specific parameters and constraints with minimal additional decoding. By eliminating redundant prefill computation, this approach reduces generation latency, while reusing validated control structures improves robustness over prompt-level caching methods RAGCache, achieving 18.31% higher task success rate and 2.3x faster policy synthesis.

11.
arXiv (CS.LG) 2026-06-17

RadSEM: A Finding-by-Finding Metric for Clinical Consistency in Radiology Reports

arXiv:2606.17062v1 Announce Type: cross Abstract: Radiology report evaluation must distinguish clinical compatibility from surface similarity, because negation, laterality, or normal-abnormal polarity can reverse a finding. We propose RadSEM (Radiology Sentence-Level Evaluation Metric), a constrained LLM-assisted metric for reference-based evaluation of radiology Findings. RadSEM rewrites reference and generated reports into ordered atomic finding sentences, each expressing one site-finding proposition. It then performs contradiction-constrained many-to-many matching: incompatible pairs such as "effusion" and "no effusion" receive no credit, while compatible granularity differences can receive partial credit. A deterministic stage weights pairs by part-whole and abnormal-detail relationships, counts unmatched findings, and produces an abnormal-focused weighted F1 score. Thus, the LLM supports structured rewriting and local alignment rather than acting as an opaque judge. We evaluate RadSEM with SSREE, a controlled monotonicity stress test built from 2,448 de-identified reports expanded into five graded corruption levels. RadSEM achieves Kendall tau_b of 0.957, all-pairs concordance of 97.8%, adjacent concordance of 95.0%, and strict five-level ordering for 81.9% of reports, outperforming radiology-specific and general text metrics while avoiding the failure in which polarity-inverted reports regain lexical overlap. On the same SSREE set, RadSEM outperforms the Ref-anchored RadSEM-Alt policy, improving adjacent concordance from 90.7% to 95.0% and strict ordering from 67.2% to 81.9%. On a 599-triplet synonym/antonym subset, RadSEM prefers synonyms in 597 cases (99.67%). These results suggest that explicit finding units, contradiction-aware matching, and abnormal-focused deterministic scoring make report scoring more interpretable and sensitive to clinically meaningful errors. Code is available at https://github.com/jdh-algo/RadSEM.

12.
medRxiv (Medicine) 2026-06-17

Non-Medical COVID-19 Impacts and Hearing Status: A Global Study of Differential Health Impact Among Deaf, Hard of Hearing, and Hearing Populations

Background: Deaf and hard of hearing (HoH) experienced complex challenges during the COVID19 pandemic, including obscured visual communication from mask mandates, inaccessible public health messaging, and inadequate interpreter availability. We examined whether hearing status predicted nonmedical COVID19 impact on a global level. Methods: We conducted a nested cross-sectional analysis within a global study collecting data across two waves (April to May 2020 and July to August 2022) from 184 countries. Participants (N=7,998) were categorized as Deaf (n=304), Hard of Hearing (HoH; n=951), or Hearing (n=6,743). The primary outcome was a composite COVID-related non-medical Personal Impact TScore derived from 14 items across employment, resource access, and healthcare domains. Multinomial logistic regression models progressively adjusted for demographic, structural, and psychosocial variables. Results: Deaf participants reported substantially higher rates of pandemic-related job loss (28.9% vs. 9.6% hearing), healthcare cancellations (39.9% vs. 24.6%), and inability to obtain basic supplies. Over half (55.9%) of Deaf participants scored above the median composite impact index, compared to 39.2% of hearing participants. In the fully adjusted model, Deaf status remained an independent predictor of high non-medical impact (aOR=1.6, 95% CI: 1.1 to 2.4). HoH status showed no statistically significant difference from hearing participants in any model. Conclusions: People identifying as Deaf experienced significant disparities during COVID19 when compared with HoH or hearing people, driven by language access barriers and institutional exclusion rather than hearing loss per se. These experiences underscore the importance for systemic interventions centering on accessible communication, Deaf-centered needs, and reducing audism in Deaf-hearing interaction.

13.
medRxiv (Medicine) 2026-06-16

Utilising Artificial Intelligence to Identify Ventricular Tachycardia Ablation Targets in Sinus Rhythm

Background and Aims: Machine learning has shown potential in predicting ablation targets for ventricular tachycardia (VT) in an animal model. This study progresses to externally validating deep learning approaches for human data. Methods: The development and external validation dataset included 21 and 13 patients, respectively, with structural VT undergoing catheter ablation. In the development datasets, electrophysiological studies were conducted using the AdvisorTM HD grid (EnsiteTM X), while both CARTO and Ensite Precision were used in the validation dataset. In each patient, VT ablation targets were defined as mapping points within 8 mm of VT isthmuses. Three advanced machine learning models were trained using cardiac mapping data acquired in both omnipolar and unipolar configurations during sinus rhythm and ventricular pacing. Discrimination was evaluated using nested leave-one-out cross-validation at patient level. Results: Overall, graph convolutional networks (GCNs), which integrate intracardiac signal waveforms with three-dimensional electroanatomical geometries, achieved the highest performance, with optimal results obtained from unipolar electrograms acquired in sinus rhythm (median AUC 0.793, sensitivity 83.6%, specificity 69.0%). This may be partly explained by the inclusion of repolarization dynamics in unipolar electrograms and the higher point density of sinus rhythm maps. Comparable performance was observed in the external dataset. Conclusion: This study demonstrates that graph convolutional networks applied to sinus rhythm EGM waveforms collected during substrate mapping can localise critical components of VT re-entry circuits. This approach has potential to provide fast and accurate ablation guidance without the need to induce and map VT, improving safety and efficacy of VT catheter ablation.

14.
arXiv (CS.CV) 2026-06-11

TextHOI-3D: Text-to-3D Hand-Object Interaction via Discrete Multi-View Generation and Joint Mesh Optimization

Text-conditioned 3D generation has progressed rapidly for images and isolated objects, but producing a hand-object mesh remains challenging: the output must preserve language semantics, cross-view consistency, object geometry, articulated hand shape, and physically plausible contact. We present TextHOI-3D, a staged framework that uses generated multi-view observations as an explicit interface between text-conditioned visual generation and geometry-aware hand-object recovery. TextHOI-3D learns a compact VQ token space for fixed-camera hand-object observations, predicts multi-view visual tokens from text with a CLIP-conditioned visual autoregressive model, and recovers a unified hand-object mesh through prior initialization, multi-view joint optimization, and anti-penetration refinement. The design separates semantic generation from geometric recovery while keeping both stages connected by a discrete multi-view representation. On HO3D-derived evaluations, the multi-view setting reduces object CD from 17.26 mm to 4.92 mm and penetration volume from 5.3721 cm^3 to 0.2193 cm^3 compared with a single-view counterpart, while improving hand errors and surface F-scores. These results support multi-view visual tokens as an effective intermediate representation for text-driven 3D hand-object mesh creation.

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

BoRAD: Bootstrap your Own Representations for Multi-class Anomaly Detection

Reconstruction-based anomaly detection is attractive for industrial inspection, but scaling it from category-specific training to a one-for-all setting is challenging. A single model must reconstruct diverse normal appearances without copying abnormal details, which exposes two coupled failure modes: identical shortcut, where anomalies pass through the reconstruction path, and mis-reconstruction, where normal categories are confused with one another. We propose BoRAD, a label-free training framework that treats this as a representation-capacity allocation problem. BoRAD uses a shared learnable prototype bank to impose two complementary regularizers: spatial prototype alignment contracts local within-prototype variation to suppress anomaly copying, while prototype-relative global alignment preserves between-prototype structure and improves sensitivity to abnormal angular deviations. The prototype bank and prediction heads are used only during training; inference remains a standard teacher-student feature discrepancy pass, with no class labels, negative pairs, memory retrieval, or prototype lookup. BoRAD achieves competitive one-for-all anomaly detection performance, including 86.2\% mAD on MVTec AD, 80.7\% mAD on VisA and 73.1\% mAD on Real-IAD. Diagnostic analyses further show reduced anomaly leakage, improved normal-category separability, and stronger anomaly-normal score separation.

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

SkillChain-Gym: A Benchmark for Reskilling-Aware Production-Inventory Control under Disruptions

arXiv:2606.17266v1 Announce Type: new Abstract: Production planning increasingly has to treat workforce capability as a decision variable: certifications lapse when skills are not maintained, new products require skills the current workforce does not hold, and reskilling competes for the same worker hours needed for production. Existing operations benchmarks usually treat labor as exogenous, while workforce-planning models with skills and learning are rarely released as reusable testbeds. We introduce SkillChain-Gym, a benchmark specification for reskilling-aware production-inventory control: a single-site environment with stylized worker skill-state dynamics, hard threshold certification, forgetting, and capacity-consuming training actions constrained by the same per-worker time budget as production. The benchmark includes seed-controlled disruption scenarios, three feasibility modes with projection diagnostics, deterministic replay, and metrics covering operations, resilience, capability growth, and training-access distribution. We evaluate production-only, reactive adaptive, water-filling adaptive, and static-insurance policies with budget variants over 60-shift horizons with paired statistical tests. The results are regime-dependent rather than a ranking. Training-capable policies dominate the production-only baseline, and maintenance training is necessary under forgetting even without disruptions. Among training-capable classes, adaptive training helps when bottlenecks are visible in the forecast, while a lean static cross-training plan, a deliberately favorable comparator whose structure encodes relevant skill contingencies, acts as strong insurance under surprise shocks and absenteeism. Capacity slack and the forgetting rate govern the boundary between these regimes. No policy class dominates across regimes, motivating forecast-driven controllers that decide when to buy skill insurance and when to react.

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

How to Detect and Measure the AI Dangers to Democracy

arXiv:2606.16054v1 Announce Type: cross Abstract: Research on artificial intelligence and democracy has grown quickly over the last decade. A shared conclusion in this literature is that AI does not create new democratic problems so much as it makes old ones worse. We now see this across information ecosystems, in elections, and in public administration. However, despite growing evidence, we lack a clear way to prioritize risks in this area, compare them across domains, and identify where democratic control is most likely to break down. So, our problem is: How can we systematize the problems that AI systems pose to democratic processes? This paper argues that principal agent theory may fit the task. In many phases of democratic systems, principals delegate key functions to AI systems and their providers without really being able to monitor how these systems operate or the outputs they produce. Treating AI as a delegation problem helps identify accountability gaps and other governance failures. Most importantly, as we shall illustrate, it provides metrics for empirical assessments of AI impact on democracy. As a second analytical element, we draw on the NIST AI Risk Management Framework and its seven characteristics of trustworthy AI, which supply substantive criteria for evaluating delegated tasks. Operationalized across the three domains through measurable indicators and domain specific trustworthiness criteria, we propose an analytical framework that centers on institutional assessability as the central condition for democratic control over AI. However, we stress that how severe a harm is, and how much risk is acceptable, are evaluative judgments that current methodologies neither acknowledge nor operationalize. This becomes acute when such evaluative judgments are (silently) delegated to private vendors. We identify this as a strong limitation left for future work.

19.
arXiv (CS.AI) 2026-06-15

MA-ProofBench: A Two-Tiered Evaluation of LLMs for Theorem Proving in Mathematical Analysis

arXiv:2606.13782v1 Announce Type: new Abstract: Large Language Models (LLMs) have made notable progress in automated theorem proving, yet existing formal benchmarks remain limited in both mathematical coverage and difficulty. Most are concentrated in areas that are easier to formalize, such as algebra and elementary number theory, and provide limited coverage of subfields that require deeper reasoning, including mathematical analysis. To address this gap, we introduce MA-ProofBench, to the best of our knowledge, the first formal theorem-proving benchmark dedicated to Mathematical Analysis. The benchmark contains 200 formalized theorems covering 6 core topics and 27 subcategories, including measure and integration theory, complex analysis, and functional analysis. The problems are divided into two difficulty levels, an undergraduate level (Level I, 100 problems) and a Ph.D. qualifying level (Level II, 100 problems), to evaluate how well LLMs perform formal reasoning at different mathematical depths. Each problem is constructed through a human-led, LLM-assisted formalization pipeline followed by independent expert review, ensuring that the formal statements remain faithful to the original mathematics. We evaluate a range of recent general-purpose reasoning models and formal theorem provers on MA-ProofBench. However, most models perform poorly: even the best-performing model, GPT-5.5, achieves only 16% Pass@8 on Level I and 5% on Level II, while most models stay close to 0% on Level II. Further analysis identifies Mathlib hallucinations and incomplete proofs as the two dominant failure modes, while an evaluation on the natural-language version of the benchmark exposes a clear gap between informal and formal reasoning. MA-ProofBench is intended to serve as a reliable reference for tracking progress in formal mathematical reasoning in advanced domains.

20.
arXiv (CS.AI) 2026-06-12

CMI-RewardBench: Evaluating Music Reward Models with Compositional Multimodal Instruction

arXiv:2603.00610v3 Announce Type: replace-cross Abstract: While music generation models have evolved to handle complex multimodal inputs mixing text, lyrics, and reference audio, evaluation mechanisms have lagged behind. In this paper, we bridge this critical gap by establishing a comprehensive ecosystem for music reward modeling under Compositional Multimodal Instruction (CMI), where the generated music may be conditioned on text descriptions, lyrics, and audio prompts. We first introduce CMI-Pref-Pseudo, a large-scale preference dataset comprising 110k pseudo-labeled samples, and CMI-Pref, a high-quality, human-annotated corpus tailored for fine-grained alignment tasks. To unify the evaluation landscape, we propose CMI-RewardBench, a unified benchmark that evaluates music reward models on heterogeneous samples across musicality, text-music alignment, and compositional instruction alignment. Leveraging these resources, we develop CMI reward models (CMI-RMs), a parameter-efficient reward model family capable of processing heterogeneous inputs. We evaluate their correlation with human judgment scores on musicality and alignment on CMI-Pref along with previous datasets. Further experiments demonstrate that CMI-RM not only correlates strongly with human judgments, but also enables effective inference-time scaling via top-k filtering. Code is available at GitHub (https://github.com/Haiwen-Xia/CMI-RewardBench). Model weights: CMI-RM (https://huggingface.co/HaiwenXia/CMI-RM). Datasets: CMI-Pref-Pseudo (https://huggingface.co/datasets/HaiwenXia/cmi-pref-pseudo) and CMI-Pref (https://huggingface.co/datasets/HaiwenXia/cmi-pref)

21.
arXiv (CS.CL) 2026-06-16

GRACE-DS: a Guarded Reward-guided Agent Correction Environment in Data Science

We introduce GRACE-DS, a Guarded Reward-guided Agent Correction Environment in Data Science for pre-deployment evaluation of LLM-powered AutoML agents. GRACE-DS is a set of evaluation metrics in an isolated environment that can be applied to tabular ML tasks specific to a particular organization. It exposes agents to realistic workflow stages, from planning and data inspection through feature engineering, model development, validation, and code repair to final submission, while hidden executable validators measure not only final predictive performance but also leakage avoidance, reproducibility, protocol validity, correction behavior, and reward alignment. The strongest structured regime, flexible iterative interaction (our approach), achieves higher end-to-end normalized hidden-test quality than single-shot generation, unstructured interaction, and restart-based baselines, while also improving protocol-valid completion. Validated across more than 7,000 episodes, these results establish GRACE-DS as a robust platform for assessing the capacity of LLM-based AutoML agents to execute machine learning workflows under production-like conditions and in accordance with organization-specific requirements.

22.
bioRxiv (Bioinfo) 2026-06-16

Phylogenetic tree inference using generative models

Accurate inference of phylogenetic trees is fundamental to evolutionary biology, yet existing methods rely on complex pipelines involving multiple sequence alignment, explicit evolutionary models, and computationally intensive tree search procedures. Here, we present BetaInfer, a generative framework that reformulates phylogenetic tree inference as a sequence transduction problem. BetaInfer leverages hybrid transformer-based architectures to directly map sets of unaligned sequences to phylogenetic trees represented in Newick format. Trained on large-scale simulated evolutionary data with known ground truth, BetaInfer learns to capture complex evolutionary signals directly from sequence data. Ensemble-based generation of multiple candidate trees further improves robustness, reducing reconstruction error by over 30% relative to single predictions. Across extensive evaluations on both simulated and empirical datasets, BetaInfer achieves competitive performance relative to state-of-the-art phylogenetic pipelines, matching, and in some cases exceeding, the accuracy of established likelihood-based and distance-based methods under a wide range of conditions. Interpretability analyses reveal that BetaInfer leverages internal pairwise-distance computations to synthesize evolutionary relationships into an integrated, global representation that supports direct tree generation. Together, these results demonstrate that generative models can serve as a viable and scalable alternative to standard phylogenetic pipelines.

23.
arXiv (CS.CL) 2026-06-16

When the Chain of Thought Knows Better: Failure Modes in Multi-Turn Reasoning Models

Failures in multi-turn reasoning models are largely invisible to terminal-score evaluation. A model can lock onto an unsafe stance early in a long dialogue, yet its final-turn refusal rate may appear indistinguishable from a robustly aligned baseline. To expose these hidden temporal dynamics, we propose a trace-level diagnostic - the CoT-Output 2x2 safety matrix. This framework labels every turn along two independent axes (internal reasoning and visible output), yielding four operationally defined failure cells: robust alignment, alignment faking, overt jailbreak, and a distinct failure mode we term context-injection failure (where the CoT maintains safe reasoning, but the visible output produces harm, highlighting a multi-turn manifestation of reasoning unfaithfulness). We evaluate three distilled reasoning targets against a fixed attacker across five oversight conditions, collecting 6750 turn-level observations on the Information-Hazard scenario. Our analysis reveals two reproducible vulnerabilities: an oversight paradox where explicit monitoring cues paradoxically increase alignment-faking rates rather than suppress them, and a context-injection failure where models lock onto unsafe external outputs despite safe internal states. We release the full dataset of multi-turn dialogues and CoT traces to support follow-up trace-diagnostic research.

24.
arXiv (CS.CL) 2026-06-11

Unstable Features, Reproducible Subspaces: Understanding Seed Dependence in Sparse Autoencoders

Sparse autoencoders (SAEs) are widely used to interpret neural network representations, but their utility depends on whether the learned features are reproducible across training runs. We study this question through feature stability: for each SAE feature, we estimate the probability that a similar feature reappears in an independently trained SAE. This yields a scalable per-feature signal that separates stable from unstable features. In a large-scale study across seeds, models, layers, dictionary sizes, and SAE variants, we find a pronounced functional asymmetry: stable features carry most of the reconstruction- and prediction-relevant signal, while unstable features have weak marginal impact and are dominated by low-frequency surface-form triggers in both activation statistics and automatic explanations. Geometrically, unstable features are individually non-reproducible but concentrate in reproducible lower-rank subspaces, suggesting that seed dependence often reflects basis ambiguity within a shared region of activation space rather than pure noise. A controlled synthetic model makes this mechanism explicit, showing that low-rank ground-truth features can be recovered at the subspace level while remaining non-identifiable as individual SAE latents across seeds. Finally, by pooling unique cross-seed features, we construct more stable SAEs while preserving explained variance in this setting. Together, these results show that unstable features are not merely failed or noisy latents: they have weak individual functional impact, but reflect reproducible low-dimensional structure that standard SAEs resolve differently across seeds.

25.
arXiv (CS.CV) 2026-06-16

Stepwise Token Selection for Efficient Multimodal Large Language Models

In multimodal large language models (MLLMs), inference cost is largely dominated by the visual token prefix rather than the language backbone, making token reduction a key factor for improving efficiency. Existing approaches typically assign independent importance scores to visual tokens and retain a fixed number of top-ranked tokens, implicitly assuming token independence and a uniform compression ratio across inputs. In this work, we reformulate visual token pruning as a sequential decision-making process. Specifically, we introduce a pointer-style selection mechanism that iteratively chooses informative tokens, conditioning each decision on previously selected ones, and dynamically determines when to stop via a learned termination action. This enables joint optimization of both the selected subset and its size. To enable end-to-end training under standard language modeling objectives, we design a differentiable relaxation based on a variance-preserving noise interpolation scheme, allowing gradients to propagate through the discrete selection process. Extensive experiments on LLaVA-v1.5-7B and Qwen2.5-VL-7B demonstrate that our approach consistently outperforms fixed-ratio baselines across different compression levels. Under aggressive pruning that removes 88.9% of visual tokens, our method preserves 94.6% of the original accuracy while achieving a 1.88x speed-up in prefill latency.