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) 2026-06-11

Bimanual Robot Manipulation via Multi-Agent In-Context Learning

arXiv:2604.20348v2 Announce Type: replace-cross Abstract: Language Models (LLMs) have emerged as powerful reasoning engines for embodied control. In particular, In-Context Learning (ICL) enables off-the-shelf, text-only LLMs to predict robot actions without any task-specific training while preserving their generalization capabilities. Applying ICL to bimanual manipulation remains challenging as the high-dimensional joint action space and tight inter-arm coordination constraints rapidly overwhelm standard context windows. To address this, we introduce BiCICLe (Bimanual Coordinated In-Context Learning), the first framework that enables standard LLMs to perform few-shot bimanual manipulation without fine-tuning. BiCICLe frames bimanual control as a multi-agent leader-follower problem, decoupling the action space into sequential, conditioned single-arm predictions. Evaluated on 13 tasks from the TWIN benchmark, BiCICLe achieves 70.5% average success rate, outperforming the best training-free baseline by 6.1 percentage points and surpassing most supervised methods. We also demonstrate superior real-world performance on 3 tasks without hardware-specific retraining.

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

FinBalance: A Multi-Document Accounting Reconciliation Benchmark

Existing financial-NLP benchmarks mostly evaluate prepared artifacts such as filings, tables, or extracted values. Real accounting begins earlier: source documents must be reconciled into cited journal entries, aggregated into a balance sheet, and checked for contradictions. We introduce FinBalance, a multi-document accounting reconciliation benchmark built from source-document bundles across eight industries, three period types, and five difficulty levels. Human-authored business scenarios, accounting policies, tax/FX treatments, document schemas, distractors, and inconsistency templates are composed by a deterministic generator whose ledger produces journal entries,balance sheets, and 23 inconsistency-code labels. On a 710-record evaluation split, six contemporary LLMs reach at most 46% exact final-balance-sheet accuracy. Four models show a 26-41 pp gap between BS_exact, the model's reported balance sheet, and BS_recon, the balance sheet obtained by replaying its entries through our ledger. Models often recover numerically plausible entries but fail to bind them to supporting documents and aggregate them consistently. Citation-pressure prompting barely changes document-linking errors, while ledger-feedback ablations substantially improve reported balance sheets and expose inconsistency-detection trade-offs. Expert finance reviewers validate the benchmark design and labels.

03.
arXiv (CS.LG) 2026-06-16

Understanding Latent Diffusability via Fisher Geometry

arXiv:2604.02751v2 Announce Type: replace Abstract: Diffusion models often degrade in latent spaces, yet the formal causes remain poorly understood. We quantify latent-space diffusability via the rate of change of the Minimum Mean Squared Error (MMSE) along the diffusion trajectory. Our framework decomposes this MMSE rate into contributions from Fisher Information (FI) and Fisher Information Rate (FIR). We demonstrate that while global isometry ensures FI alignment, FIR is governed by the interplay between encoder and data geometries. Our analysis decouples diffusion degradation into four penalties: dimensional compression, tangential distortion, high-frequency encoder curvature, and intrinsic data curvature. We derive theoretical conditions for FIR preservation to ensure stable diffusability. Experiments across diverse autoencoding architectures demonstrate the implications of our theoretical bounds. We establish FI and FIR as a comprehensive analytical framework for understanding latent diffusability.

04.
arXiv (CS.AI) 2026-06-12

The Theory of Mind Utility: Formal Specification of a Mentalizing Mechanism

arXiv:2606.12721v1 Announce Type: new Abstract: Inferring others' beliefs requires more than reading surface signals; it requires tracking who told them what, in what order, and how credibly. The Theory of Mind Utility (ToM-U) formalizes this epistemic state inference problem at the computational level of analysis, specifying what mentalizing computes and why without commitment to algorithmic or neural implementation. ToM-U achieves this by constructing Local Epistemic World Models (LEWMs) – directed typed graphs that represent agents, state nodes, and the epistemic relationships among them – and evaluating discrete candidate LEWMs against observed behavior until one achieves sufficient confidence. Five formal definitions specify the LEWM structure, agent node properties including ordered information access history, a bounded proliferation mechanism for recursive mentalizing, three inference procedures, and a residue function that captures the structured trace left by failed mentalizing attempts. ToM-U differs from Bayesian Theory of Mind and adjacent formal accounts, which presuppose rather than derive belief states, and from simulation theory and theory-theory, which lack a formal apparatus for epistemic state inference. The architecture generates directional, falsifiable predictions about mentalizing failure that follow from structural properties of the model rather than auxiliary assumptions, and positions ToM-U as a domain-agnostic mechanism upstream of goal inference and other downstream social cognitive processes.

05.
arXiv (CS.LG) 2026-06-16

AREAL-DTA: Dynamic Tree Attention for Efficient Reinforcement Learning of Large Language Models

arXiv:2602.00482v2 Announce Type: replace Abstract: Reinforcement learning (RL)-based post-training for large language models (LLMs) is computationally expensive, as it generates many rollout sequences that frequently share long token prefixes. Existing RL frameworks usually process these sequences independently during policy training, i.e., repeatedly recomputing identical prefixes in both the forward and backward passes of policy gradient computation, leading to substantial inefficiencies in computation resources and memory usage. Although prefix sharing naturally induces a tree structure over rollouts, packed tree-mask approaches scale poorly in RL settings. In this paper, we introduce AReaL-DTA, which efficiently exploits prefix sharing in RL training. AReaL-DTA employs a depth-first search (DFS)-based execution strategy that dynamically traverses the rollout prefix tree during both forward and backward computation, materializing only a single root-to-leaf path at a time. To further improve scalability, AReaL-DTA incorporates a load-balanced distributed batching mechanism that dynamically constructs and processes prefix trees across multiple GPUs. On $\tau^2$-bench, AReaL-DTA improves training throughput by up to $8.31\times$ over dense training and up to $1.70\times$ over sparse training. Our code is available at https://github.com/areal-project/AReaL/tree/feat/dta.

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

Human-on-the-Loop Orchestration for AI-Assisted Legal Discovery

arXiv:2606.19812v1 Announce Type: new Abstract: Autonomous Large Language Model (LLM) agents are increasingly deployed in electronic discovery (e-discovery), where compounding errors across multi-step reasoning chains can constitute legal malpractice. Unlike single-turn retrieval, agentic workflows operating over privileged document corpora exhibit a class of failure we term "trajectory collapse": an early misclassification silently propagates, rendering an entire privilege review invalid. This paper makes three contributions. First, we propose a structured taxonomy of agentic failures in legal information retrieval, organized by functional stage. Second, we introduce a four-layer verification architecture – spanning planning, reasoning, execution, and uncertainty quantification – designed to intercept these failures before they compound. Third, we present a preliminary simulation study on a synthetic e-discovery corpus that demonstrates how mandatory Human-on-the-Loop (HOTL) escalation thresholds reduce privilege-waiver risk relative to fully autonomous baselines. Our results suggest that calibrated uncertainty thresholds can reduce privilege-waiver risk by up to 61% versus fully autonomous deployment, while routing fewer than one quarter of documents to attorney review.

07.
arXiv (CS.CV) 2026-06-16

CPS4: Class Prompt driven Semi-Supervised Spine Segmentation with Class-specific Consistency Constraint

Vision Language Model (VLM) has great potential to enhance the quality of pseudo labels in semi-supervised spine segmentation by leveraging textual class prompts to generate segmentation map, but no one has studied it yet. Although promising, it lacks explicit constraints to ensure consistency between spine class prompts and spine unit region, resulting in unsatisfactory performance in multi-class segmentation map generation. In this paper, we propose CPS4, the first text-guided semi-supervised spine segmentation network using class prompts to enhance the quality of spine pseudo labels. Specifically, CPS4 is implemented through two training stages. (i) Class-specific consistency constrained VLM pretraining stage: we propose token- and pixel-level attention loss to optimize the consistency between class prompts and spine units, forcing the textual class prompt to be closely coupled with the target spine unit in the semantic space. (ii) Class Prompt driven semi-supervised spine segmentation stage: using the pretrained vision-text encoder, we derive each class-specific binary segmentation map for the unlabeled spine image and integrate them into an unified multi-class segmentation map, improving the quality of the spine pseudo label generated by the semi-supervised spine segmentation network. Experimental results show that our CPS4 achieves superior spine segmentation performance with Dice of 80.44%, only using 5% labeled data on the public spine segmentation dataset, surpassing popular semi-supervised learning and VLM methods. Our code will be available.

08.
arXiv (CS.CV) 2026-06-16

Detect Before You Leap: Mirage Detection in Vision-Language Models

Vision-language models (VLMs) can produce confident visual answers even when the required visual evidence is missing, blank, or unrelated to the question. This failure mode, recently described as mirage (mirage2026), is especially concerning in medical and document VQA, where a plausible but visually ungrounded answer may be mistaken for image-based evidence. We study the complementary problem of pre-release mirage detection: given an image-question pair, determine whether the VLM should answer or abstain before generation. To that end, we propose a novel model-agnostic Text-Conditioned Layer-wise Internal Alignment (TC-LIA) method that probes patch-token representations across the layers of a CLIP ViT-H/14 vision encoder. The key idea is to project layer-wise image patch tokens into the final CLIP embedding space and measure their similarity with the question embedding, thereby tracking whether question-relevant visual evidence emerges across vision layers. TC-LIA summarizes this alignment trajectory using final image-text cosine similarity, late-layer top-k patch-text alignment, early-to-late gain, and layer-wise slope. These features are combined with pixel-statistic based blank/noise detection, zero-shot domain routing, and structured VLM self-assessment in an ensemble. Across five VQA domains with related, unrelated-real, and blank/noise inputs, and across twelve VLM backbones, Qwen2.5-VL-32B achieves the highest three-class detection accuracy of 94.7% with a 3.0% mirage rate, while Qwen2.5-VL-72B achieves 94.6% accuracy with a lower 2.8% mirage rate. Baseline mirage rates span 21.7-66.6%.

09.
arXiv (CS.CL) 2026-06-11

Models That Know How Evaluations Are Designed Score Safer

The validity of AI safety evaluations depends on models behaving consistently across controlled and deployment settings. Prior work has identified test-time contextual cues, such as hypothetical scenarios, as a source of verbalized evaluation awareness and subsequent behavioral shift. In this paper, we investigate a potential explanation of this phenomenon: evaluation meta-knowledge, defined as parametric knowledge about the structural traits that characterize evaluations. Similar to dataset contamination, where benchmark exposure leads to higher performance through memorization, we hypothesize that models trained on texts describing evaluation practices may implicitly learn to recognize and respond to evaluation-like contexts, for instance, through exposure to scientific articles or social media posts about AI benchmarking. To test this, we fine-tune models on synthetic documents describing evaluation traits such as verifiable structures or moral dilemmas. Evaluating this fine-tuned model on six safety benchmarks, we find that it is significantly safer than the base model and control model. This behavioral shift persists even when restricting the analysis to responses lacking explicit verbalization of evaluation awareness. Our results demonstrate that evaluation meta-knowledge may inflate safety benchmark performance, introducing a novel confounder that is independent of explicit memorization or verbalized evaluation awareness, thus, challenging to detect. These findings have important implications for the design and interpretation of AI safety evaluations. Our code and models are available at https://github.com/compass-group-tue/arxiv2026_evaluation_meta_knowledge.

10.
arXiv (CS.CV) 2026-06-11

Ouroboros-Spatial: Closing the Data-Model Loop for Spatial Reasoning

Spatial reasoning remains a persistent challenge for multimodal large language models (MLLMs). Existing approaches largely rely on large-scale, statically curated datasets, where all training samples are treated uniformly regardless of the model's evolving capabilities. This static paradigm is inherently data-inefficient: training capacity is often spent on samples that are either trivial or overly difficult for the model at its current stage. To address this limitation, we propose Ouroboros-Spatial, a self-evolving training framework in which the model plays dual roles as a proposer and a solver. In each iteration, a frozen proposer generates spatial question-answer (QA) pairs from 3D scene metadata and raw video frames, together with executable code for deriving reliable ground truth. A learnable solver is then fine-tuned on the accepted samples, and its per-sample prediction confidence is used as a difficulty signal. This signal is fed back to the proposer in the next iteration, guiding it to generate questions better matched to the solver's current capabilities. Through this closed-loop design, the training distribution co-evolves with model ability, reducing redundant trivial examples while filtering out ambiguous or uninformative samples with limited learning value. Across six spatial reasoning benchmarks, Ouroboros-Spatial substantially improves Qwen3-VL-4B and Qwen3-VL-8B while using an order of magnitude fewer training examples than recent large-scale curated datasets. On VSI-Bench, it yields absolute gains of 9.9 and 6.8 points for the 4B and 8B models, respectively, enabling both to outperform a wide range of strong open-source and proprietary baselines.

11.
medRxiv (Medicine) 2026-06-15

A controlled human infection model for symptomatic pertussis in North America using the pertactin-producing clinical isolate D420

Background Despite widespread vaccination, pertussis remains a poorly controlled disease globally and results in substantial annual morbidity and mortality, particularly in young children. Controlled human infection models (CHIMs) using the causative agent Bordetella pertussis are promising systems to enable the study of pertussis disease pathogenesis and immunology and to rapidly assess vaccines and therapeutics. While a pertussis CHIM that produces asymptomatic infection has been established in Europe, the development of a CHIM that leads to symptomatic illness would be advantageous for evaluating vaccine efficacy against both infection and disease. Methods Healthy participants 18-40 years of age were inoculated intranasally with one of eight doses (ranging from 104 to 108 colony forming units (CFU)) of the pertactin-producing B. pertussis isolate D420 at the challenge facility within the Canadian Center for Vaccinology (Nova Scotia, Canada). The study occurred in two stages. In stage one, the B. pertussis dose was escalated in cohort groups of five to six participants until reaching an endpoint where 70-90% of participants exhibited mild (non-severe, Grade 1 or 2) symptomatic infection, defined as the Human Infectious Dose 70-90 (HID70-90). In stage two, additional challenges were conducted for doses below, at, and above the identified HID70-90 to characterize the emerging pertussis model. For all challenge doses, participants were closely monitored during an inpatient stay of up to 24 days and post-discharge for laboratory-confirmed infection, pertussis symptoms, safety, and IgG antibody responses to four B. pertussis antigens including pertussis toxin, filamentous hemagglutinin, fimbriae, and pertactin. All participants received a five-day course of azithromycin, where timing of initiation depended on B. pertussis testing and symptoms. The study was conducted between July 4, 2022 and March 19, 2025. Findings Seventy-five participants were inoculated with one of the eight B. pertussis D420 challenge doses and completed the inpatient stay. From the stage-one dose escalation, we found that 107 CFU of B. pertussis D420 was the lowest dose that achieved the HID70-90, where 9 of 12 participants (75.0%) exhibited mild symptomatic infection. Following stage-two challenges, 16 of 22 total participants at 107 CFU (72.7%) developed mild symptomatic infection, thus verifying the HID70-90. The symptomatic infection rate below the HID70-90 at 5x106 CFU of D420 was 20.0% and above the HID70-90 at 5x107 and 108 CFU were 58.3% and 55.6%, respectively. Symptoms with elevated frequency for symptomatic infection (relative to background symptoms in non-infected) included nasal congestion, runny nose, fatigue, malaise, and cough. At the HID70-90, 50% of symptomatic infections included cough. Serological analyses of the four highest (stage-two) challenge doses (5x106, 107, 5x107, 108 CFU) revealed that antibody titres increased over time post-challenge. Seroconversion for at least one of the four studied antibodies was nearly twice as common for symptomatic (70.0%) than asymptomatic (35.7%) infection and was absent (0%) for non-infected. All infections were cleared following azithromycin treatment (100%) and there were no study-related serious adverse events. Interpretation A safe and reproducible symptomatic pertussis CHIM was achieved, providing a model for research on pertussis disease pathogenesis and immunology and for assessing vaccines and therapeutics. (Clinicaltrials.gov, NCT05136599).

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

ReCal: Reward Calibration for RL-based LLM Routing

arXiv:2606.12479v1 Announce Type: cross Abstract: Large language model (LLM) routing has emerged as an effective paradigm for leveraging the complementary strengths of multiple LLMs through dynamic model and reasoning-strategy selection. Recent reinforcement learning (RL)-based routing methods further improve routing quality by optimizing routing policies from interaction feedback. However, they still struggle to provide informative and comparable learning signals under heterogeneous tasks with varying difficulty. In practice, multiple objectives (e.g., correctness, format behavior) are aggregated into a single scalar reward, leading to ambiguous credit assignment and conflicting optimization signals. Moreover, reward signals exhibit significant variability across instances, where some instances produce higher or more variable rewards, introducing optimization bias that favors trivial samples over informative ones. To address these issues, we propose ReCal, a \underline{Re}ward \underline{Cal}ibration framework for RL-based LLM routing. We first introduce a hierarchical reward decomposition mechanism with component-wise advantage estimation. We further propose a distribution-aware optimization strategy that calibrates optimization variability through variance-aware reweighting and per-dataset normalization. Experiments on seven datasets demonstrate that ReCal consistently improves routing performance, and training stability over baselines. Code is available at https://anonymous.4open.science/r/ReCal.

13.
arXiv (CS.LG) 2026-06-16

Exploding and vanishing gradients in deep neural networks: the effect of residual connections

arXiv:2606.17013v1 Announce Type: cross Abstract: The well known phenomenon of exploding and vanishing gradients in deep neural networks is analyzed using multiplicative ergodic theory. The effect of adding a residual connection is explained in this context. Specifically, a characterization of Liapunov exponents due to Furstenberg and Kifer is exploited in order to make a precise statement about the Liapunov spectrum and the effect of residual connections on it.

14.
arXiv (CS.CV) 2026-06-17

Adversarial Attacks Leverage Interference Between Features in Superposition

Why do adversarial examples exist, and why do they transfer between models? Existing explanations appeal to high-dimensional geometry, non-robust patterns in the input, and decision boundary structure, but none provides a representation-level mechanism that explains why specific perturbations succeed and why attacks transfer between models. In this paper, we show that adversarial vulnerability can stem from efficient information encoding in neural networks. Specifically, vulnerability can arise from superposition - the phenomenon where networks represent more concepts than they have dimensions, forcing non-orthogonal representation and thus interference. This interference causes perturbations targeting one representation to affect others, creating vulnerabilities determined by interference patterns. In synthetic settings with precisely controlled superposition, we establish that superposition suffices to create adversarial vulnerability. The resulting attacks are predictable: PGD-discovered perturbations align with theoretically optimal perturbations derived from the interference geometry. Models trained on similar data develop similar interference patterns, explaining attack transferability. We then show that successful attacks on image classifiers exhibit the structure predicted by our proposed mechanism. These findings reveal that adversarial vulnerability can be a byproduct of networks' representational compression, complementing existing explanations based on data properties or architectural factors.

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

Contract-Based Compositional Shielding for Safe Multi-Agent Reinforcement Learning

arXiv:2606.14130v1 Announce Type: new Abstract: Safe coordination problems surface in multi-agent reinforcement learning when global safety cannot be enforced by any agent unilaterally: the admissibility of one agent's action may depend on the dynamics of other agents. Decentralised shields can enforce safety at runtime, but purely factorised permissions often exclude optimal team behaviour that is safe only through coordination. We study deterministic safety guarantees for agents trained and deployed under decentralised execution, recovering team-optimal safe behaviour without centralised runtime control. Agents have a shared global specification $\phi$ in the safety fragment of Linear Temporal Logic ($\mathsf{LTL}_{\mathsf{safe}}$ ), and select among tuples of local $\mathsf{LTL}_{\mathsf{safe}}$ obligations whose conjunction implies the global specification $\phi$. Each agent may rely on the other agents' local obligations as assumptions because the whole contract tuple is certified simultaneously and allows projection into local action masks. At learning time, a non-stationary multi-armed bandit chooses among a library of local $\mathsf{LTL}_{\mathsf{safe}}$ obligations to select the tuple that optimises team reward, all without forgoing end-to-end safety. We evaluate the approach across 6 environments and 15 algorithmic variants.

16.
medRxiv (Medicine) 2026-06-18

Hospital-Level Variation in Antenatal Corticosteroids for Late Preterm Births

Objective: To determine whether and to what extent hospitals across the United States vary in their use of late-preterm steroids using a novel data set in which the timing of steroid administration relative to delivery can be observed. Methods: This was a retrospective cohort study of singleton births with known gestational ages identified in the Premier Healthcare Database from 2015 to 2022. The primary variable of interest was hospital-level adoption of antenatal corticosteroids for late-preterm singleton deliveries, calculated as the proportion of late-preterm singleton births (34-36 completed weeks of gestation) with any betamethasone exposure during the same late-preterm period. Hospital adoption was defined as the weighted average rate of ALPS administration among late-preterm infants across the entire post-period. Hospitals were ranked by their late-preterm steroid adoption rates and categorized by quartile based on the empirical distribution. Temporal trends were assessed using annual hospital-level adoption rates and visualized using time-series plots and distributional plots. A logistic regression model was constructed to determine hospital characteristics associated with being a highest-quartile adopting hospital. Results: The analysis cohort included 728 hospitals and 5,452,791 births, of which 361,006 (6.6%) were singleton late preterm births. Hospital steroid exposure rates ranged from 0 to 82% and were categorized into quartiles based on overall exposure rate, with cutoffs at 20.6%, 29.8%, and 40.1%. Median exposure rates increased progressively across quartiles from 14.1% (IQR 9.3-17.4%) in the lowest adopting hospitals (Q1) to 47.6% (IQR 43.7-53.2%) in the highest adopting hospitals (Q4), with substantial within-quartile variation. In the multivariable model, urban location was a strong predictor of high adoption after adjustment (aOR 2.05; 95% CI 1.11-3.83, p=0.02). Compared to Midwest hospitals, Southern hospitals had significantly lower odds of being high adopters (aOR 0.37; 95% CI 0.20-0.69, p

17.
arXiv (CS.AI) 2026-06-18

R2BC: Multi-Agent Imitation Learning from Single-Agent Demonstrations

arXiv:2510.18085v2 Announce Type: replace-cross Abstract: Imitation Learning (IL) is a natural way for humans to teach robots, particularly when high-quality demonstrations are easy to obtain. While IL has been widely applied to single-robot settings, relatively few studies have addressed the extension of these methods to multi-agent systems, especially in settings where a single human must provide demonstrations to a team of collaborating robots. In this paper, we introduce and study Round-Robin Behavior Cloning (R2BC), a method that enables a single human operator to effectively train multi-robot systems through sequential, single-agent demonstrations. Our approach allows the human to teleoperate one agent at a time and incrementally teach multi-agent behavior to the entire system, without requiring demonstrations in the joint multi-agent action space. We show that R2BC methods match, and in some cases surpass, the performance of an oracle behavior cloning approach trained on privileged synchronized demonstrations across four multi-agent simulated tasks. Finally, we deploy R2BC on two physical robot tasks trained using real human demonstrations.

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

SwitchBraidNet: Quantisation-Aware Lightweight Architecture for Hybrid Brain-Computer Interface

arXiv:2606.18816v1 Announce Type: cross Abstract: Hybrid brain-computer interfaces (BCIs) that integrate motor imagery (MI) and steady-state visual evoked potentials (SSVEP) provide high-dimensional neural decoding but typically exceed the computational limits of embedded hardware. To address this, we propose SwitchBraidNet, a compact EEG classification architecture designed for low-power deployment. The model employs a dual-path temporal braid to extract multiscale oscillatory features, an adaptive squeeze-and-excitation spatial switch for electrode gating, and a log-variance readout layer for direct band-power encoding. Furthermore, through systematic quantisation-aware training on the OpenBMI dataset, we compared SwitchBraidNet against four established baselines across FP32, FP16, and INT8 precisions. Experimental results demonstrate superior efficiency and performance, achieving MI accuracy of 69.49% (FP16), SSVEP accuracy of 93.48% (FP32), and a hybrid information transfer rate of 64.82 bits/min (FP16). With an INT8 footprint of only 3.03 KB, SwitchBraidNet maintains high accuracy across varying numerical precisions, demonstrating its suitability for low-power embedded BCI deployment.

19.
arXiv (CS.LG) 2026-06-16

Pre-Training for Simulation-Based Science: A Study on Jet Foundation Model Training Objectives

arXiv:2606.14870v1 Announce Type: cross Abstract: Foundation models (FMs) trained on large datasets and fine-tuned on downstream tasks have emerged as a powerful paradigm in AI for science. Industrial FMs are typically trained using self-supervision with masking due to the lack of labels. In many scientific domains, accurate simulations are plentiful and facilitate large, labeled datasets. This opens up new possibilities for pre-training. We present a systematic comparison of pre-training methods using the OmniLearned High Energy Physics FM framework. We test supervised classification, flow-matching generation, and self-supervised masked particle modeling. All models are pre-trained on the JetClass dataset and fine-tuned on two representative downstream tasks, top jet classification and JetNet conditional generation. Among other observations, for classification tasks, we find that pure classifier pre-training is optimal when downstream labels and model capacity are plentiful, but combining it with self-supervised masked particle modeling (MPM) is uniquely powerful in the low-finetuning label regime. Flow matching-based generative pre-training seems to provide little benefit for downstream classification, and interestingly, for downstream generation, we find that flow matching must be in the pre-training objective to see a significant finetuning advantage, hinting at the orthogonality of classification and generation tasks. That is, for a model to transfer to both generative and classification downstream tasks, it must be pre-trained on both. This study provides a template for controlled scaling analysis of pre-training objectives for foundation models in simulation-based sciences.

20.
arXiv (CS.LG) 2026-06-11

UniIntervene: Agentic Intervention for Efficient Real-World Reinforcement Learning

arXiv:2606.12372v1 Announce Type: cross Abstract: Human-in-the-loop reinforcement learning (HiL-RL) has emerged as an effective paradigm for real-world robotic manipulation, enabling online policy improvement with human guidance. However, current HiL-RL frameworks remain intervention-intensive, relying on frequent human corrections to redirect the policy out of unproductive exploration, which incurs high labor cost and limits real-world scalability. To address this, we propose UniIntervene, an agentic intervention model that detects unproductive exploration and autonomously recovers the policy toward high-value states, taking over the bulk of interventions from human operators. Specifically, UniIntervene first performs future-conditioned action-value estimation, predicting the latent consequence of the current action and evaluating its induced value, which provides a more stable progress signal. Building on this, a temporal value-risk critic aggregates recent value dynamics and triggers intervention when the estimated value exhibits sustained stagnation or degradation. When intervention is required, UniIntervene retrieves a high-value recovery target from a memory of past intervention episodes and produces executable corrective actions through a goal-conditioned recovery policy. In this way, UniIntervene turns intervention from passive human correction into a value-aware recovery process for efficient real-world RL. Extensive experiments on diverse real-world manipulation tasks demonstrate that UniIntervene improves the average success rate by 8.6% while reducing human interventions by 57% relative to state-of-the-art HiL-RL baselines.

21.
bioRxiv (Bioinfo) 2026-06-18

Benchmarking gene expression reconstruction from single-cell latent representations

Single-cell transcriptomics is typically modeled in low-dimensional latent representations that improve the signal-to-noise ratio of the data. Such representations underpin data integration, cell state discovery, and perturbation prediction, with applications ranging from large-scale organ atlases to latent trajectory modeling. Recent virtual cell approaches further leverage these representations to predict cellular responses as distributional shifts in latent space. Each of these applications ultimately requires faithful gene expression reconstruction from latent spaces for biological interpretation, enabling gene-level analysis of predicted perturbed or batch-corrected cells. Yet representation choice is typically treated as an implementation detail rather than a primary modeling decision, with no systematic evaluation of how well latent representations support gene expression reconstruction. Here, we introduce ReconEval, a benchmark for evaluating gene expression reconstruction from single-cell latent spaces. We benchmark two classes of latent representations: end-to-end trained models such as PCA, autoencoders, and variational autoencoders, and pretrained single-cell foundation model embeddings coupled to newly trained decoders. Reconstruction is evaluated both directly and after latent-space perturbation prediction. Across perturbational and observational datasets totaling over 100 million cells, our metric suite quantifies statistical fidelity; biological signal preservation, including differential expression, coexpression, cell-cycle structure, cytokine response and pathway activity; and perturbation-specific effects. We find that autoencoders achieve the strongest stand-alone reconstruction at low dimensionality, while variational regularization does not improve generalization in reconstruction. Frozen foundation model embeddings retain recoverable gene-level information, with reconstruction quality depending strongly on decoder architecture and pretraining objective. In latent perturbation modeling, high-dimensional PCA matches foundation model embeddings, while low-dimensional AE embeddings are optimal for flow-based generative models. Overall, reconstruction depends critically on the interplay between representation and downstream model, and simpler representations can outperform complex alternatives given appropriate capacity. Our benchmark establishes reconstruction as a critical evaluation axis for single-cell foundation models. We envision it improving the biological interpretability of latent-space modeling, a prerequisite for future virtual cell models to be validated by domain experts and grounded in biology.

22.
bioRxiv (Bioinfo) 2026-06-11

Pillbox: A Leakage-Aware Foundation-Model Predictor and Lineage-Ceiling Diagnostic for Cancer Drug Response

We present Pillbox, a predictor whose pipeline is audited against the six Asiaee leakage modes with the one residual pathway shown by per-fold ablation to be non-load-bearing on hard splits. Our model combines CpGPT methylation embeddings, CLAMP drug embeddings, and per-fold-fit gene-expression principal components which are fused by Feature-wise Linear Modulation (FiLM)-conditioned graph attention on the STRING v12 protein-protein interaction graph. Then we alpha-ensemble the model against a histogram-based gradient boosting regressor baseline. On GDSC GSE68379 (987 cell lines, 375 drugs) across seeds 42, 7, and 123, the ensemble reaches test R-Squared of 0.78, 0.77, and 0.76 on random, histology-blind, and site-blind splits respectively, with cell-aware lifts above the drug-mean floor of +0.054, +0.060, and +0.037. As a quantitative diagnostic for feature-stack saturation we propose the cross-architecture residual correlation, calibrated against a same-architecture-different-initialization control. On histology-blind splits the cross-architecture value of 0.939 falls short of the same-architecture ceiling of 0.974 by approximately 0.03 in residual correlation, a gap we interpret as the headroom available to architecture choice on top of the current foundation-model representation and consistent with the long-established observation that tissue lineage dominates cell-line drug response. We integrated curated mutation, methylation, and drug-target-expression channels, but these do not improve prediction once foundation-model embeddings are in place. Cross-screen validation against PRISM matches the GDSC-to-PRISM measurement reproducibility ceiling within 0.01 Spearman.

23.
arXiv (CS.CL) 2026-06-18

Structured Inference with Large Language Gibbs

The knowledge encoded in large language models (LLMs) can serve as a substrate for structured reasoning over variables describing a complex world, but accessing this knowledge in a probabilistically coherent manner poses a difficult inference problem. We propose Large Language Gibbs, a scheme for structured probabilistic inference that uses conditional distributions of an LLM as transition operators. Rather than sampling structured objects through single-pass autoregressive generation, we iteratively resample individual variables conditioned on others using an LLM's next-token conditionals. This approach avoids order-dependent biases and produces a stationary distribution that reflects a compromise between all local conditionals. We apply this approach to sampling from synthetic distributions, consistent reasoning tasks, and Bayesian structure learning. The results suggest that the use of LLM conditionals in MCMC is a practical alternative to one-pass generation for structured probabilistic inference under a world prior accessible through noisy LLM conditionals.

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

LiAuto-GeoX: Efficient Grounded Driving Transformer

Dense 3D reconstruction has demonstrated immense potential for spatial understanding, yet its viability as a real-time, onboard representation for autonomous driving remains an open challenge. Existing large-scale visual geometry models typically require substantial computational resources and lack the long-range geometric fidelity, surround-view consistency, and real-time efficiency demanded by dynamic driving environments. To bridge this gap, we present LiAuto-GeoX, an efficient grounded driving transformer designed for deployable, ego-centric 3D scene understanding. Our approach begins by learning a high-capacity driving geometry model from large-scale surround-view data, utilizing sparse LiDAR priors to provide robust geometric grounding in distant, ambiguous, or structure-sparse regions. We then instantiate this capability into a highly compact 155M-parameter onboard model through a novel geometry-preserving distillation framework. This framework employs mask-guided depth-aware distillation to retain fine-grained metric structures by emphasizing geometrically informative regions, and relative-pose relational distillation to enforce cross-view spatial consistency through pose-induced geometric relations. Extensive evaluations reveal that LiAuto-GeoX runs at 220 FPS on KITTI while maintaining high-fidelity dense reconstruction, enabling real-time deployment. The learned geometry transfers seamlessly to downstream autonomy tasks, achieving 90.6 PDMS in trajectory prediction, 24.63 mIoU in occupancy prediction, and 47.67 IoU in future-frame prediction. These all demonstrate that efficient dense 3D reconstruction can transcend its traditional role as a perception target to serve as a scalable, foundational geometric representation for next-generation autonomous driving.

25.
arXiv (CS.AI) 2026-06-12

Evaluation Sovereignty in Metadata-Driven Classification: A Multi-Track Framework for Weakly Supervised Information Systems

arXiv:2606.13436v1 Announce Type: new Abstract: Evaluation in machine learning is typically treated as a neutral measurement process. However, in operational information systems, evaluation outcomes are often conditioned by the processes used to generate labels. This paper does not seek to improve classification performance. Instead, it examines the validity of performance measurement under differing label-authority regimes. This issue is particularly relevant in large-scale metadata-driven systems, where labels are often incomplete, inconsistent, or weakly supervised. We introduce evaluation sovereignty, defined as the degree to which performance metrics are independent of label authority and supervision regime, and propose a multi-track evaluation framework that systematically varies training and evaluation label sources. Using hierarchical multi-label classification on large-scale scientific metadata, we demonstrate that models exhibiting strong performance under operational ("silver") evaluation degrade substantially under independent ("gold") evaluation, particularly for fine-grained classification. For example, Micro-F1 decreases from approximately 0.54 to 0.03. Notably, ranking-based metrics remain above baseline, revealing a divergence between latent model signal and classification validity. These findings suggest that commonly reported performance metrics may reflect alignment with labeling processes rather than true predictive capability. We therefore reconceptualize evaluation validity as a system-level property shaped by label governance and provide a practical methodology for auditing intelligent systems operating under weak supervision.