Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达,利用大语言模型自动生成交叉领域文献分析简报。

01.
arXiv (quant-ph) 2026-06-16

Inflationary branch decoherence and the cosmological arrow of time

作者:

arXiv:2602.21263v3 Announce Type: cross Abstract: We analyze branch decoherence in inflationary quantum cosmology by computing reduced density matrices and branch-overlap factors for long-wavelength perturbations. The Hartle-Hawking no-boundary state is real in the semiclassical regime and contains both expanding and contracting WKB components, whereas the tunneling state is selected as an outgoing complex WKB branch; expanding-contracting decoherence is therefore central for the former and mainly diagnostic for the latter. Using the influence-functional formalism, we derive the noise kernel for a light spectator environment and evaluate decoherence under horizon-based and EFT-motivated coarse grainings. We then compute the single-mode branch overlap directly from the Bunch-Davies mode functions, obtaining $|\mathcal{D}_k(z)|=[z^2/(z^2+1)]^{1/4}$ in the massless limit and $|\mathcal{D}_k(z)|\sim z^\nu$ on superhorizon scales for massive fields, where $z=-k\eta$ is the dimensionless wavenumber with $\eta$ the conformal time. In the massless case, the accumulated geometric branch functional is evaluated in closed form, with a leading cutoff-sensitive phase-space term and a universal subleading contribution. The calculation provides an explicit quantitative bridge between quantum-cosmological boundary conditions, inflationary squeezing, and the emergence of effectively classical cosmological histories.

02.
arXiv (CS.LG) 2026-06-11

Data-Driven Dynamic Assortment in Online Platforms: Learning about Two Sides

arXiv:2606.11118v2 Announce Type: replace Abstract: We study a dynamic assortment problem on a two-sided service platform with incomplete information and heterogeneous customers in a discrete-time setting. In each period, a customer arrives seeking service, and the platform chooses an assortment of sellers to display. The customer then proposes a transaction to at most one seller in the assortment according to a multinomial logit choice model. After a fixed number of periods, sellers review the proposals they have received and each chooses at most one customer according to another multinomial logit choice model, after which the cycle repeats. A key challenge is that the platform does not know the choice-model parameters of either customers or sellers in advance. To our knowledge, this is the first study of a dynamic assortment problem in which both sides' choice parameters are unknown. We develop a data-driven algorithm that learns these parameters while optimizing the platform's objective over time. We evaluate performance using regret, which measures revenue loss relative to a clairvoyant benchmark that knows all parameters and customer arrivals in advance. We show that the algorithm's worst-case regret grows polylogarithmically over time, and we derive a matching lower bound, establishing its rate optimality.

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

David vs. Goliath in Next Activity Prediction: Argmax vs. LSTM, Transformer, and LLM

arXiv:2606.15868v1 Announce Type: new Abstract: Next activity prediction (NAP) is a cornerstone of predictive process monitoring (PPM), enabling organizations to move from retrospective analysis to proactive process steering. The PPM field has progressed from classical machine learning through deep learning architectures such as LSTMs and Transformers to large language models (LLMs). Despite growing model complexity, no benchmark jointly compares LLMs, Transformers, LSTMs, and simple baselines in a direct sequence modeling setting for NAP. In this paper, we fill this gap with a systematic benchmark. We compare vocabulary-adapted LLMs, Transformers trained from scratch, LLM-distilled Transformers, and LSTMs against a simple counting-based argmax baseline across seven real-life event logs. Our results tell a David vs. Goliath story: pretraining confers no consistent improvement over training from scratch, model size shows little effect on performance, and on most datasets the argmax baseline matches or approaches the performance of billion-parameter LLMs.

04.
arXiv (CS.LG) 2026-06-19

Insulin4RL: Real-Time Insulin Management in the Intensive Care Unit for Offline Reinforcement Learning

arXiv:2606.19481v1 Announce Type: new Abstract: Offline reinforcement learning (ORL) offers the potential to improve the quality of clinical decision-making using historical electronic health record (EHR) data. Current training and evaluative practices in this field rely heavily on EHR datasets that have been temporally discretised into fixed, regular time intervals. Discretisation creates fictional representations of complex clinical scenarios and compromises the generalisability of retrospective model evaluations. In this paper, we introduce Insulin4RL, a healthcare ORL dataset featuring naturally irregular inputs and actions from real clinical trajectories. Derived from MIMIC-IV, Insulin4RL comprises over 375,000 labelled decisions across 12,209 patients requiring insulin infusion titration in the Intensive Care Unit. The dataset can thus be used for research into ORL model performance under realistic clinical sampling assumptions. We provide a description of the dataset's structure and characteristics, baseline performance metrics using model-free offline reinforcement learning, and a standardised evaluation protocol using fitted Q-evaluation. We conclude with suggested areas for future research that could be addressed using this resource.

05.
medRxiv (Medicine) 2026-06-22

Evidence-guided AI regularization for suicidal ideation prediction in pediatric bipolar disorder

Background: Suicide prediction models in psychiatry often rely on purely data-driven feature selection, which can produce unstable and clinically opaque predictor sets in modest-sized samples. We developed Evidence-Based AI LASSO (EBAL), an evidence-guided regularization framework that incorporates curated clinical evidence into feature-specific penalty factors for interpretable prediction. Methods: Baseline data from 136 youth with confirmed bipolar spectrum disorder in the Greater Houston Area Bipolar Registry were analyzed using 20 candidate clinical predictors. Forty higher-level evidence documents on suicidality and related predictor domains were curated through a structured evidence synthesis workflow and indexed as an auditable evidence corpus. An open-weight large language model assigned feature-specific penalty factors using a prespecified scoring rubric, and these penalties were used to fit a weighted LASSO model. EBAL was compared with a standard evidence-agnostic LASSO using nested leave-one-out cross-validation. Results: For suicidal ideation, EBAL achieved an AUROC of 0.768, balanced accuracy of 0.757, sensitivity of 0.758, and specificity of 0.757. The standard LASSO achieved an AUROC of 0.760 and balanced accuracy of 0.715. EBAL improved balanced accuracy (+0.042, p=0.010) and Matthews correlation coefficient (+0.079, p=0.010), while retaining fewer stable predictors than standard LASSO (11/20 vs 18/20). The strongest positive predictors were current depressed mood, duration of mood disorder illness, and comorbid generalized anxiety disorder. For suicidal behavior, both models performed near chance and retained all candidate predictors. Limitations: The study was cross-sectional, single-site, and modest in sample size, with no external validation cohort. Conclusions: EBAL produced a sparser and more clinically coherent model for suicidal ideation in pediatric bipolar disorder, but did not improve prediction of suicidal behavior. These findings support evidence-guided regularization as a transparent strategy for aligning psychiatric prediction models with prior clinical knowledge while preserving interpretability.

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

Beyond English: Uncovering the Multilingual Gap in Vision-Language-Action Models

Vision-Language-Action models have recently demonstrated promising capabilities in learning generalist robot policies from large-scale multimodal data. However, most existing VLA systems are trained and evaluated primarily with English instructions, leaving their ability to understand and execute instructions in other languages largely unexplored. While the underlying large language models often possess multilingual capabilities, it remains unclear whether these multilingual capabilities transfer to VLAs during training. In this work, we present the first systematic study of multilingual instruction following in VLA models. We first construct multilingual instructions by extending existing benchmarks with translations of their instructions. Using these instructions, we evaluate several representative VLA models across a range of tasks in simulation settings. Our experiments reveal a significant multilingual gap: models trained primarily on English instructions exhibit substantial performance degradation when evaluated on other languages, even when the underlying language backbone is multilingual. We provide several findings and analyses to understand the multilingual gap. Cross-lingual transfer behavior analysis shows that performance drops correlate with both instruction understanding and action execution. Representation analyses suggest that multilingual instruction-caused representation shifts may contribute to the multilingual gap. Motivated by these findings, we further explore strategies to improve multilingual performance in VLAs. We propose a simple yet effective multilingual fine-tuning approach, Multilingual Principal Component Alignment, which leverages Principal Component Analysis to get the principal component subspace and align projected multilingual representations, effectively reducing the multilingual performance gap.

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

ROVE: Unlocking Human Interventions for Humanoid Manipulation via Reinforcement Learning

arXiv:2606.17011v1 Announce Type: cross Abstract: Human interventions provide crucial corrective signals for post-training Vision-Language-Action (VLA) models. However, enabling seamless humanoid interventions is a formidable systems challenge due to complex whole-body kinematics and dexterous-hand control. Consequently, the collected intervention trajectories are often suboptimal, and methods that rely on human interventions as expert supervision can absorb hesitant, inefficient, or even erroneous behaviors. To address both the system and algorithmic challenges, we propose ROVE, a reinforcement learning framework for humanoid VLA post-training with imperfect human interventions. First, ROVE introduces a human-in-the-loop pipeline capable of collecting deployment and intervention data for humanoid manipulation. Second, it utilizes Optimistic Value Estimation (OVE) to prioritize high-value behaviors from mixed-quality trajectories. To further robustify value estimation, we incorporate cross-embodiment human experience videos to provide rich supervision for long-tailed failure and recovery modes. The resulting critic yields informative advantage signals, steering the VLA actor to focus on high-value behaviors rather than indiscriminately imitating all actions. On challenging real-world contact-rich and fine-grained humanoid manipulation tasks, ROVE outperforms experience-learning baselines and consistently improves across multiple rollout-intervention iterations.

08.
arXiv (CS.LG) 2026-06-12

NetCause: Counterfactual Learning for Root Cause Analysis in Large-Scale Networks

arXiv:2606.13543v1 Announce Type: cross Abstract: Can a learned model capture how faults propagate through a large-scale network and use this knowledge to causally attribute customer impact to its underlying root cause? Existing root cause analysis techniques often rely on static rules, correlation heuristics, or topology-local reasoning, which struggle to generalize in dynamic environments where faults propagate across complex physical and logical dependencies. We present NetCause, a self-supervised learning-based framework that models network incidents as graph-temporal processes and uses counterfactual simulation to rank candidate root causes. This approach produces an interpretable ranking of root cause hypotheses and integrates naturally with operator-defined mitigation and remediation actions. We train the model on over 1,500 incidents collected over six months from a leading cloud provider's production network and evaluate it on 31 expert-labeled incidents. NetCause consistently improves root cause ranking quality in the regime most relevant to operational decision-making, achieving a 16.1% accuracy improvement over a rule-based heuristic baseline. While training is computationally intensive, inference is lightweight, requiring only seconds of GPU runtime per incident (well below typical telemetry collection latencies).

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

Where Do Backdoors Live? A Component-Level Analysis of Backdoor Propagation in Speech Language Models

Speech language models (SLMs) are systems of systems: independent components that unite to achieve a common goal. Despite their heterogeneous nature, SLMs are often studied end-to-end; how information flows through the pipeline remains obscure. We investigate this question through the lens of backdoor attacks. We first establish that backdoors can propagate through the SLM, leaving all tasks highly vulnerable. From this, we design a component analysis to discover the role each component takes in backdoor learning. We find that backdoor persistence or erasure is highly dependent on the targeted component. Beyond propagation, we examine how backdoors are encoded in shared multitask embeddings, showing that poisoned samples are not directly separable from benign ones, challenging a common separability assumption used in filtering defenses. Our findings emphasize the need to treat multimodal pipelines as intricate systems with unique vulnerabilities, not solely extensions of unimodal ones.

10.
arXiv (CS.CV) 2026-06-17

BusterX++: Towards Unified Cross-Modal AI-Generated Content Detection and Explanation with MLLM

The rapid advancement of generative AI has substantially improved image and video synthesis, amplifying the risk of multimodal visual misinformation. Recent MLLMs have shown promise for transparent AI-generated content detection through reasoning and explanation, yet existing approaches largely treat image and video forensics as isolated tasks, leaving cross-modal synergies underexplored. To address this, we present BusterX++, a unified MLLM for joint image and video detection with interpretable reasoning. We also introduce GenBuster-Bench++, a meticulously curated, difficulty-aligned benchmark containing balanced image and video samples spanning recent generation models and diverse real-world scenarios. Using this controlled setting, we revisit the widely adopted $SFT \rightarrow RL$ post-training paradigm. Notably, our findings demonstrate that a single-stage, pure RL strategy driven strictly by sparse outcome rewards consistently matches or surpasses a strong SFT+RL baseline across both unified and single-modality settings. Our key insight reveals that SFT imposes lower policy entropy, which restricts the policy search space and dampens exploratory freedom. In contrast, single-stage pure RL maintains higher policy entropy throughout training, effectively unlocking the spontaneous emergence of cross-modal capability transfer between image and video forensics. Extensive experiments demonstrate that BusterX++ achieves state-of-the-art performance, highlighting the powerful potential of RL for unified cross-modal visual reasoning.

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

ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models

Multimodal large language models (MLLMs) may memorize sensitive cross-modal information during pretraining, making machine unlearning (MU) crucial. Existing methods typically evaluate unlearning effectiveness based on output deviations, while overlooking the generation quality after unlearning. This can easily lead to hallucinated or rigid responses, thereby affecting the usability and safety of the unlearned model. To address this issue, we propose ASRU, a controllable multimodal unlearning framework that incorporates generation quality as a core evaluation objective. ASRU first induces initial refusal behavior through activation redirection, and then optimizes fine-grained refusal boundaries using a customized reward function, thereby achieving a better trade-off between target knowledge unlearning and model utility. Experiments on Qwen3-VL show that ASRU significantly improves unlearning effectiveness (+24.6%) on average and generation quality (5.8X) on average while effectively preserving model utility, using only a small amount of retained supervision data.

12.
arXiv (quant-ph) 2026-06-19

Propagating Collective Spin-valley Modes in Twisted WSe2

arXiv:2507.18770v2 Announce Type: replace-cross Abstract: The emergence of neutral collective modes is a hallmark of correlated quantum phases but is often challenging to probe experimentally. In two-dimensional flatband systems, charge responses have been intensively investigated yet neutral excitations remain largely unexplored. In particular, intervalley coherent state (IVC) features a neutral Goldstone mode due to spontaneously broken valley U(1) symmetry. While IVC state has been proposed as a unifying theme across graphene and semiconductor based systems, its defining feature, the neutral Goldstone mode, remains elusive in experiment. Here we investigate space and time resolved transport of neutral modes in twisted WSe2 moire superlattices through a novel ultrafast imaging technique. We uncover two new propagating collective modes with very different velocities, which emerge near the van Hove singularity (VHS) in both intermediate (3.5 to 4 degree) and large (around 5 degree) angle twisted WSe2. The fast-propagating mode has a large speed of about 3 km/s and is consistent with a Goldstone mode for an IVC state, while the slow-moving mode is likely a gapped amplitude mode. They can be understood as the spin-valley analogues of collective modes of a superfluid, whose propagation is imaged for the first time in a condensed matter system. Our study demonstrates a powerful new approach for probing charge-neutral modes in quantum materials and offers key insights into the interplay between charge and spin-valley physics in moire superlattices.

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

KeepLoRA++: Continual Learning with Layer-Scaled Residual Gradient Adaptation

Continual learning for pre-trained vision-language models requires balancing three competing objectives: retaining pre-trained knowledge, preserving knowledge from a sequence of learned tasks, and maintaining the plasticity to acquire new knowledge. This paper presents KeepLoRA++, balancing these objectives through a unified dual-dimensional knowledge retention mechanism. We analyze knowledge distribution of Transformer architecture from both inter-layer and intra-layer perspectives. The inter-layer perspective examines how retention is distributed across layers, while the intra-layer perspective focuses on the parameter space within each layer. Our analysis reveals a structural property: general transferable knowledge is mainly encoded in the shallow layers and the principal subspace of the parameters, while task-specific adaptations are localized in the deep layers and the residual subspace. Motivated by this insight, KeepLoRA++ introduces a layer-scaled residual gradient adaptation method. New tasks are learned by restricting LoRA parameter updates to the residual subspace, combined with a shallow-to-deep layer scaling, to prevent interference with previously acquired capabilities. Specifically, the gradient of a new task is projected onto a subspace orthogonal to both the principal subspace of the pre-trained model and the dominant directions of previous task features, while simultaneously assigning smaller update magnitudes to shallow layers and larger ones to deeper layers. Our theoretical analysis and empirical evaluations confirm that KeepLoRA++ successfully balances these three competing objectives, consistently outperforming representative baselines across image classification, visual question answering, and video understanding tasks.

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

Image Quality Assessment of Identity Cards Using Measures from Open Face Image Quality

This paper addresses the challenge of assessing image quality in ID cards in remote verification systems by applying capture-related quality measures from the Open Face Image Quality (OFIQ) standard to ID card images. Our preprocessing pipeline includes corner detection, perspective normalization, and comprehensive foreground masking to ensure accurate and unbiased quality measure computation. We evaluate the effectiveness of these measures by analyzing their correlation with the performance of three presentation attack detection (PAD) algorithms across four diverse ID card datasets, where two datasets contain bona fide, i.e. pristine, images and two contain printed mock ID cards. Our results suggest that quality assessment based on some OFIQ measures can significantly improve PAD performance.

16.
Nature (Science) 2026-06-10

The Amazon can be saved — with concerted action inside and outside Brazil

作者: 未知作者

As deforestation in the Amazon falls, fresh evidence shows that the rainforest can withstand global warming, but only if there is a worldwide effort to stop cutting it down. As deforestation in the Amazon falls, fresh evidence shows that the rainforest can withstand global warming, but only if there is a worldwide effort to stop cutting it down.

17.
arXiv (CS.CV) 2026-06-19

How Fragile Are Training-Free AI-Generated Image Detectors? A Controlled Audit of Score Direction, Preprocessing, and Compression

Training-free detectors of AI-generated images promise generator-agnostic deployment without classifier training, yet their reported numbers are rarely compared under a single controlled protocol. We audit two representative training-free scores – an autoencoder-reconstruction score (AEROBLADE-style) and a noise-perturbation feature-similarity score (RIGID-style) – plus a naive feature-kNN control, on a common 1,500-image GenImage-derived benchmark spanning seven generators and JPEG compression at quality 70 and 50. The audit yields three cautionary findings. (i) Implementation details masquerade as method differences: replacing the LPIPS backbone (AlexNet -> VGG-16) changes overall AUROC by +0.085, and switching between resize-to-512 and native-resolution preprocessing flips per-generator conclusions by up to 0.38 AUROC. (ii) Score direction is not a property of the method but of its hyperparameters: the RIGID-style score is inverted (AUROC < 0.5) on SD1.5 and Wukong at noise level sigma=0.05, recovers to >0.5 for every generator at sigma=0.01, and collapses to 0.15 at sigma=0.3. (iii) Dataset format bias inflates robustness claims: without unified re-encoding, AUROC under JPEG-50 exceeds the clean condition for the AlexNet-backbone reconstruction score; after bias correction the residual anomaly localizes to a single generator (BigGAN). The audited scores have complementary per-generator failure sets, but naive z-score fusion does not beat the best single score, indicating that exploiting complementarity requires direction-aware combination.

18.
arXiv (quant-ph) 2026-06-19

Approximating optimal decoding of quantum LDPC codes with narrow frontiers

arXiv:2606.20513v1 Announce Type: new Abstract: We introduce the Frontier decoder, a pruned dynamic-programming decoder for sparse quantum decoding problems. Frontier processes error variables in a chosen order, merges prefixes with the same residual syndrome and logical label, and approximates logical-coset posterior masses by retaining only a narrow scored frontier. Without pruning, the recursion is exact ordered inference with exponential complexity. In the code-capacity setting, the decoder reaches thresholds close to optimal for the surface code and the color code. In the circuit-level noise model, it achieves state-of-the-art performance with a very small average retained list size: less than 100 for the gross code $[[144,12,12]]$ at a physical error rate of $0.001$. When the list size is constant, the decoder has linear complexity, suggesting the possibility of low-latency implementations.

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

MortarBench: Evaluating Mortgage Loan Origination Agents

arXiv:2606.19416v1 Announce Type: new Abstract: Loan origination is the process by which a lender creates a new loan, from application and underwriting through approval and funding. This process serves a critical role in evaluating the eligibility and level of risk posed by an applicant. Recently, firms have begun using mortgage loan agents to augment human loan officers, despite a lack of any public benchmark. To fill this gap, we present MortarBench, a loan origination agent benchmark. MortarBench uses a financial data synthesis and mutation pipeline to generate examples with broad edge case coverage that match real-world distributions and questions. We find that state-of-the-art large language models (LLMs) perform poorly, with closed-source models achieving at most 77.1\% exact match accuracy. We also discover systematic biases in LLM perception of foreignness related to non-English names. Noting these weaknesses, we introduce CRIT, a confidence calibration framework. Our method increases accuracy to 80.5\% while improving risk management steering and reducing bias.

20.
arXiv (CS.CL) 2026-06-18

Redact or Keep? A Fully Local AI Cascade for Educational Dialogue De-Identification

Educational dialogue is a valuable but sensitive resource for research: the same transcripts that capture authentic learning often capture personally identifiable information (PII) entangled with curricular content, where "Riemann" may refer to a real student or to a mathematical concept. Existing approaches force a tradeoff between governance and accuracy. Commercial Large Language Models (LLMs) can handle this ambiguity but require sending student data to third parties, while local named entity recognition (NER) systems preserve governance but over-redact curricular terms. We propose a fully local cascade framework that reframes de-identification from open-ended entity recognition to constrained privacy triage. A recall-first union proposer combines two lightweight encoders with deterministic rules to over-generate candidate spans; a context-aware reviewer then makes a binary Redact/Keep decision for each candidate using surrounding dialogue and speaker role. We evaluate three reviewer configurations against same-family LLM-only baselines and a commercial API on math tutoring transcripts from two large platforms. The strongest local configuration reaches 0.958 macro F1, compared with 0.767 for a same-family LLM-only baseline and 0.706 for the commercial API, while running entirely on a single laptop. On a targeted challenge set of curricular-personal name ambiguity, the same configuration degrades by only 0.03 F1 versus 0.19 to 0.25 for smaller reviewers. These results suggest that for educational de-identification, problem formulation matters more than model scale.

22.
arXiv (quant-ph) 2026-06-19

Benchmark of quantum algorithms for ground state preparation in the presence of noise

arXiv:2606.20551v1 Announce Type: new Abstract: We compare the performance of representative cooling, adiabatic, and optimization algorithms for ground-state preparation in the presence of noise. Using an exactly solvable family of quadratic fermionic Hamiltonians subject to depolarizing noise, we derive the scaling of the achievable relative energy as a function of the noise rate and support these results with numerical simulations. The Hamiltonian exhibits two phases, separated by a quantum phase transition. As expected, the performance of the different algorithms depends on the phase: adiabatic evolution is favorable in the trivial phase, while a multi-frequency cooling algorithm, as proposed in [1], becomes competitive or superior in the topological phase, where gap-closing limits adiabatic protocols. We further present numerical results for the quantum approximate optimization algorithm [2], showing that it performs competitively with cooling in the trivial phase but is typically outperformed in the topological regime. Finally, we show that for this model the cooling protocol exhibits enhanced robustness to parameter imperfections, highlighting its potential advantage for realistic implementations of noisy quantum state preparation. The analytical approach developed here, in conjunction with numerical validation, establishes an extendable approach to benchmarking ground-state preparation algorithms.

23.
arXiv (math.PR) 2026-06-16

Malliavin Calculus for the stochastic Cahn-Hilliard equation driven by fractional noise

arXiv:2601.10490v2 Announce Type: replace Abstract: The stochastic partial differential equation analyzed in this work is the Cahn-Hilliard equation perturbed by an additive fractional white noise (fractional in time and white in space). We work in the case of one spatial dimension and apply Malliavin calculus to investigate the existence of a density for the stochastic solution $u$. In particular, we show that $u$ admits continuous paths almost surely and construct a localizing sequence through which we prove that its Malliavin derivative exists locally, and that its law is absolutely continuous with respect to the Lebesgue measure on $\bf R$, establishing thus that a density exists. A key contribution of this work is the analysis of the stochastic integral appearing in the mild formulation: we derive sharp estimates for the expectation of the $p$-th power ($p \geq 2$) of the $L^{\infty}(D)$-norm of this stochastic integral as well as for the integral involving the $L^{\infty}(D)$-norm of the operator associated with the kernel appearing in the integral representation of the fractional noise, all of which are essential for this study.

24.
arXiv (CS.LG) 2026-06-12

EPM-JEPA: Operator-Side Experience Modulation in JEPA-Family World Models

arXiv:2606.12979v1 Announce Type: new Abstract: JEPA-family world models use a static predictor whose weights do not adapt when test-time dynamics diverge from training. We compare two mechanisms for incorporating accumulated experience into a JEPA predictor under distribution shift: operand-side injection, where a compressed experience representation is added as a residual to the predictor's hidden state (EI-JEPA), and operator-side modulation, where the same representation generates low-rank weight deltas via LoRA applied to the predictor's weights (EPM-JEPA). On a pre-registered comparison (Moving MNIST, gravity shift), EPM-JEPA (D_shift^{n=50} = 0.7848 +/- 0.0078, three seeds) differs from EI-JEPA (0.8238) by delta = 4.74% - Outcome C: a null result - by our stated criterion, a valid outcome. As a secondary, non-pre-registered observation, EPM-JEPA improves 1.90% over a no-memory baseline (0.8000), consistently across seeds, while EI-JEPA underperforms the baseline, indicating the benefit is specific to weight-level modulation. Our primary contribution is a mechanism analysis: the D_shift^{n=50} trajectory reflects three independent dynamical processes - buffer cycling, EMA target drift, and an intrinsic LoRA settling transient of +0.021 - rather than convergence to equilibrium. These findings motivate PEM-JEPA, a physics-grounded successor addressing this dynamical-peak limitation.

25.
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.