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 (math.PR) 2026-06-19

Establishing an $\Omega(\sqrt{d})$ complexity lower bound for PDMP samplers and how to break it: a sub-$\sqrt{d}$ algorithm for Gaussian-tailed targets

arXiv:2606.19909v1 Announce Type: cross Abstract: Despite the theoretical appeal of their non-reversibility, to date, no Piecewise Deterministic Markov Process (PDMP) samplers have been developed that scale better than $\mathcal{O}(\sqrt{d})$ in computational complexity with respect to the target dimension $d$. We prove that this is a fundamental limitation by establishing an $\Omega(\sqrt{d})$ lower bound on the algorithmic complexity of PDMP samplers in a standard setup. By relaxing the assumption that the target density must remain invariant at all continuous times, we then demonstrate how to bypass this barrier. Specifically, we introduce a novel PDMP sampling scheme and show that it achieves an empirical complexity of $\mathcal{O}(d^\alpha)$, where $\alpha \in [0.2, 0.3]$ for Gaussian-tailed targets. In addition, this PDMP scheme is locally adaptive in both trajectory length and distance between velocity updates.

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

Size Doesn't Matter: Cosine-Scored Sparse Autoencoders

arXiv:2606.15054v1 Announce Type: new Abstract: Sparse autoencoders (SAEs) detect features via inner product, so a feature's activation scales with both its directional alignment and the input's norm. Under BatchTopK, high-norm tokens inflate all pre-activations simultaneously, claiming dictionary slots regardless of content alignment. This matters because sublayer normalization has already discarded the magnitude the score measures, so the encoder detects a quantity the model does not read. We replace the score with a learned blend of cosine similarity and input magnitude, letting the optimizer choose how much norm to use; a per-feature extension lets each feature decide independently. In both regimes, training is free to recover inner product but never does, with no feature ever choosing more than half-magnitude dependence. At matched reconstruction, the cosine encoder learns features that align with human-recognizable concepts far more often than standard, filling dictionary slots that inner product wastes on norm detectors. Loss reweighting that equalizes gradients barely closes the gap, confirming forward-pass score geometry as the lever. The advantage is not universal across tasks or depths, but we believe cosine scoring should be the default for dictionary learning on normalized representations.

04.
arXiv (CS.LG) 2026-06-17

Overcoming the Incentive Collapse Paradox

arXiv:2603.27049v2 Announce Type: replace-cross Abstract: AI-assisted task delegation is increasingly common, yet human effort in such systems is costly and typically unobserved. Recent work by Bastani and Cachon (2025); Sambasivan et al. (2021) shows that accuracy-based payment schemes suffer from incentive collapse: as AI accuracy improves, sustaining positive human effort requires unbounded payments. We study this phenomenon in a budget-constrained principal-agent framework with strategic human agents whose output accuracy depends on unobserved effort. Our first contribution is a general impossibility result showing that incentive collapse is not merely a limitation of simple linear payments, but arises for any payment rule based only on observed task accuracy.To overcome this barrier, we propose a sentinel-auditing payment mechanism that enforces a strictly positive and controllable level of human effort at finite cost, independent of AI accuracy. Building on this incentive-robust foundation, we develop an incentive-aware active statistical inference framework that jointly optimizes (i) the auditing rate and (ii) active sampling and budget allocation across tasks of varying difficulty to minimize the final statistical loss under a single budget. Experiments demonstrate improved cost-error tradeoffs relative to standard active learning and auditing-only baselines.

05.
arXiv (CS.AI) 2026-06-24

Engineering Reliable Autonomous Systems: Challenges and Solutions

arXiv:2606.23760v1 Announce Type: cross Abstract: Engineering reliable autonomous systems is an important and growing topic in computer science. As autonomous systems become more prevalent, easy-to-use techniques for building them reliably are increasingly important. This workshop report captures and expands on the discussions at the Lorentz Center Workshop "Engineering Reliable Autonomous Systems" (ERAS), held from 10 to 14 June 2024. The workshop was co-organised by the organisers of the Workshop on Formal Methods for Autonomous Systems (FMAS) and the Workshop on Agents and Robots for reliable Engineered Autonomy (AREA). It brought together members of the FMAS and AREA communities, industry practitioners, and representatives from sectors where autonomous systems pose distinctive engineering challenges. The workshop focused on three main research topics: techniques for verification and validation of autonomous systems; engineering real-world autonomous systems; and software architectures for safe autonomous systems. Its main outcome is a catalogue of challenges in these areas and, most importantly, a pathway to solutions. Some challenges can already be tackled by techniques that are well known in academia but have not yet become regularly used in practice. Other challenges remain unresolved and require further research. This roadmap is intended to support future research and industrial collaboration.

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

UniMM: A Unified Mixture Model Framework for Multi-Agent Simulation

arXiv:2501.17015v2 Announce Type: replace Abstract: Simulation plays a crucial role in assessing autonomous driving systems, where the generation of realistic multi-agent behaviors is a key aspect. In multi-agent simulation, the primary challenges include behavioral multimodality and closed-loop distributional shifts. In this study, we formulate a unified mixture model (UniMM) framework for generating multimodal agent behaviors, which can cover the mainstream methods including regression-based mixture models and discrete NTP models. Furthermore, we introduce a closed-loop sample generation approach tailored for mixture models to mitigate distributional shifts. Within the UniMM framework, we recognize critical configurations from both the model and data perspectives. We conduct a systematic examination of various model configurations, and comprehensively characterize their effects. Moreover, our investigation into the data configuration highlights the pivotal role of closed-loop samples in achieving realistic simulations. To extend the benefits of closed-loop samples across a broader range of mixture models, we further introduce a temporal disentanglement-and-alignment mechanism to address the shortcut learning and off-policy learning issues. Leveraging insights from our exploration, the distinct variants proposed within the UniMM framework, including discrete, anchor-free, and anchor-based models, all achieve state-of-the-art performance on the WOSAC benchmark.

07.
arXiv (CS.CV) 2026-06-12

Bounding Boxes as Goals: Language-Conditioned Grasping via Neuro-Symbolic Planning

For robotics to be effectively integrated into household or industrial environments, machines must adapt to natural-language prompts in real time. Although Vision-Language Models (VLMs) have enabled zero-shot generalization in robot task and motion planning (TAMP), current state-of-the-art approaches often remain computationally "heavyweight" or require extensive training on thousands of demonstrations. We present GRASP (Grounded Reasoning and Symbolic Planning), a framework designed as a step toward open-vocabulary tabletop manipulation. Our approach leverages a pretrained VLM to translate natural-language queries into neuro-symbolic goal states, grounded in the physical world via a bounding-box detection pipeline. Unlike methods that rely on fixed color lists or hard-coded coordinates, GRASP enables robots to interpret abstract spatial concepts such as "top shelf" and execute tasks without additional fine-tuning. We achieve 73.3% overall success across 90 real-robot trials at three difficulty levels, requiring no task-specific training.

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

T2S: A Rehearsal-Based Approach for Extraction-Resistant Model Watermarking

arXiv:2606.11698v1 Announce Type: cross Abstract: Model watermarking safeguards AI model intellectual property by embedding distinctive knowledge that induces unique behavioral signatures. The primary technical challenge lies in ensuring watermark robustness against various post-processing attacks on the watermarked model. Model extraction attacks emerge as the most severe threat, where adversaries exploit prediction outputs to train surrogate models that illegally replicate the original model's functionality. In this work, we propose a rehearsal-based watermark embedding framework to enhance the robustness of model watermarks against model extraction attacks. By simulating the extraction process, our method leverages the loss of a simulated stolen model on a trigger set as a training signal to fine-tune the watermark knowledge within the target model. This fine-tuning step encourages the watermark to be embedded in a way that boosts transferability, thereby increasing its chances of persisting and remaining detectable in stolen models. Comprehensive experiments conducted under diverse settings demonstrate that the proposed method significantly improves the robustness of model watermarks against both model extraction and subsequent watermark removal attacks.

09.
arXiv (CS.AI) 2026-06-12

Parthenon Law: A Self-Evolving Legal-Agent Framework

arXiv:2606.04602v3 Announce Type: replace Abstract: As agents grow more capable, legal-domain LLM agents promise to turn document-heavy matters into reviewable work products – yet reliable deployment faces three obstacles: no large-scale evidence on how today's strongest model-and-harness combinations behave on end-to-end legal matters; no agent architecture adapted to the legal vertical, only general-purpose harnesses; and, in a setting that keeps shifting with new facts, authorities, and deadlines, no mechanism for systems to learn from their own outcomes. We address each. A large-scale empirical study on Harvey LAB – $12{,}510$ agent trajectories – shows that even frontier agents remain far from completing matters in a single pass: per-criterion accuracy climbs with stronger models while strict matter completion stalls. We then introduce \textsc{Parthenon}, a self-evolving legal-agent framework that factors Model, Harness, Agent roles, legal Knowledge, deterministic Tools, and procedural Skills into auditable surfaces for source traceability, date and number grounding, deliverable compliance, and issue closure. Finally, an anti-leakage learning loop converts scored failures into task-agnostic edits to skills, tools, and knowledge, letting the system improve with experience – as a firm refines its checklists and playbooks after each matter – without touching model weights. Across our large-scale empirical analysis, \textsc{Parthenon} substantially improves the performance of state-of-the-art models and harnesses on legal-matter tasks.

10.
arXiv (CS.LG) 2026-06-19

Neural network surrogates with uncertainty quantification for inverse problems in partial differential equations

arXiv:2606.20417v1 Announce Type: new Abstract: Inverse problems for differential equations arise throughout science and engineering, where one seeks to infer unknown model parameters from noisy or incomplete observations. Traditional numerical methods for these problems are often computationally expensive, particularly in Bayesian settings where evaluating the likelihood becomes costly for complex forward models and high-dimensional parameter spaces. To address this challenge, we introduce DeepGaLA, a neural-network surrogate for differential equation solvers that provides uncertainty-aware predictions, reducing overconfident inference when training data are limited. To evaluate the fidelity of the surrogate-induced posterior approximations in practice, we show that a short run of delayed-acceptance Markov chain Monte Carlo can serve as an effective diagnostic. Across a range of numerical experiments, DeepGaLA delivers forward-model approximations with accuracy comparable to established Gaussian-process surrogates, while better maintaining efficiency as parameter dimension grows. Moreover, it can incorporate differential-equation constraints, including in nonlinear settings. Overall, these results indicate that uncertainty-quantified neural surrogates can enable scalable and reliable Bayesian inference for inverse problems in complex systems.

11.
bioRxiv (Bioinfo) 2026-06-11

TMO: ASYMMETRIC CROSS-MODAL ATTENTION FOR LEARNINGCELL-STATE-DEPENDENT REGULATORY LAGS FROM SINGLE-CELL MULTIOMIC DATA

Abstract Background: Single-cell multi-omics technologies simultaneously measure chromatin accessibility (ATAC) and gene expression (RNA), providing a unique window into the temporal ordering of regulatory events during differentiation. However, most computational models treat the two modalities symmetrically, ignoring the directional relationship between chromatin and transcription, and existing lag-aware methods estimate a single global lag per gene, failing to capture cell-state-dependent dynamics. Methods and Results: We introduce Temporal Multi-Omics (TMO), a deep learning framework that learns signed, cell-state-conditional regulatory lags ({Delta}{tau}) using asymmetric cross-modal attention. TMO projects RNA and ATAC into 50 latent components each, tokenises each cell as a sequence of 100 tokens, and uses a two-pass transformer in which a data-driven lag prior - derived from a sliding-window cross-correlation function - directly biases attention asymmetrically. On four independent 10x Multiome datasets (mouse brain, human brain, mouse kidney, human PBMC), the asymmetric model achieves Lag Concordance Scores (LCS) of 0.988-0.999, compared to 0.048-0.108 for an architecturally identical symmetric baseline. A stratified 80/20 held-out experiment confirms that the learned component-lag ordering generalises to unseen cells (held-out LCS 0.85-0.99). Clustered {Delta}{tau} heatmaps show positive {Delta}{tau} (ATAC-led priming) in early pseudotime and negative {Delta}{tau} (RNA-led, activity-dependent regulation) in late pseudotime; the ATAC-RNA correlation heatmap exhibits a U-shaped pattern indicative of developmental decoupling. Components with the most positive {Delta}{tau} are enriched for chromatin organization and stem cell differentiation (FDR < 0.05), while those with the most negative {Delta}{tau} are enriched for synaptic signalling and immune activation. Ablating the cell-state information from the lag predictor reduces the LCS and collapses per-component temporal dynamics (KS p [&le;] 0.039 in all four tissues), proving that TMOs dynamic lag patterns depend on cell-state conditioning. Independent ChIP-seq validation for four transcription factors (PAX5, Pax6, ASCL1, Hnf4) confirms highly significant separation between target genes and expression-matched background (p < 10-4 in all cases). Two Multiome Perturb-seq screens provide causal validation: SMARCB1 knockout shows a directional trend (1.5-fold target shift, p = 0.056, n = 147 perturbed cells), and SMARCE1 knockout reaches statistical significance (p = 0.0089, n = 3,394 perturbed cells). Gene-level cross-correlation independently validates that the regulatory lag signal is present in the raw data, and TMO further identifies rare, statistically significant biphasic gene programs where the regulatory direction reverses across pseudotime. Conclusions: TMO is the first method to make regulatory lag a learnable, cell-state-conditional, and architecturally encoded parameter. It is scalable, interpretable, and open-source, providing a powerful tool for studying regulatory timing in development, disease, and perturbation screens.

12.
arXiv (CS.CL) 2026-06-11

Debiasing Without Protected Attributes: Latent Concept Erasure from Textual Profiles

Most fairness research in NLP assumes direct access to protected attributes such as gender, race, or nationality. In practice, however, such information is often unavailable due to privacy constraints, missing metadata, or legal restrictions, even though models may infer it from indirect textual cues. This raises a key question: can debiasing succeed without direct access to sensitive attributes? We propose H-SAL, which performs post-hoc concept and attribute erasure using self-description text as an implicit debiasing signal. To support this setting, we introduce a multi-domain Stack Exchange-based fairness benchmark for helpfulness prediction that includes both explicit and implicit signals, enabling comparison between standard debiasing with protected labels and debiasing without access to sensitive information. Across encoder and decoder-only language models, we find that implicit self-description often matches or outperforms explicit-label-based debiasing. Our results broaden representation-level fairness research and provide a new benchmark for studying debiasing under realistic data constraints.

13.
arXiv (CS.LG) 2026-06-15

Running the Gauntlet: Re-evaluating the Capabilities of Agents Beyond Familiar Environments

arXiv:2606.14397v1 Announce Type: new Abstract: As agentic systems continue to evolve and are widely deployed in real-world scenarios, there is a growing demand to faithfully evaluate their capabilities. However, current benchmarks are typically built on popular applications with relatively simple tasks and focus on a narrow set of capabilities while overlooking broader dimensions, resulting in saturated performance on modern agents and failing to probe their limitations. To this end, we introduce GauntletBench, a web-based benchmark for evaluating agent generalisation in challenging scenarios, focusing on three underexplored capabilities (temporal perception, graphical understanding, and 3D reasoning), across five less-covered professional applications (Video Editor, Workflow Builder, 3D Modeller, Flight Analyser, and Circuit Designer), each with 20 vision-intensive tasks (100 in total). Our benchmark provides a modular pipeline that comprises an environment compatible with both open- and closed-source agent frameworks, a controlled web-based application, a well-structured task suite, and an automated evaluation engine with diverse metrics. Contrary to widespread expectations, our empirical results reveal that frontier agentic systems remain far from achieving human-level performance. Even the state-of-the-art agent achieves only a 19.1% success rate on our GauntletBench, highlighting the limitations in these overlooked capabilities and generalisation. By comparison, non-expert human annotators achieve over 80% success on our challenging yet feasible tasks, revealing the substantial gap between current agent capabilities and those required for complex real-world scenarios.

14.
arXiv (math.PR) 2026-06-18

Formation of clusters and coarsening in weakly interacting diffusions

arXiv:2510.17629v3 Announce Type: replace-cross Abstract: This paper studies the clustering behavior of weakly interacting diffusions under the influence of sufficiently localized attractive interaction potentials on the one-dimensional torus. We describe how this clustering behavior is closely related to the presence of discontinuous phase transitions in the mean-field PDE. For local attractive interactions, we employ a new variant of the strict Riesz rearrangement inequality to prove that all global minimizers of the free energy are either uniform or single-cluster states, in the sense that they are symmetrically decreasing. We analyze different timescales for the particle system and the mean-field (McKean-Vlasov) PDE, arguing that while the particle system can exhibit coarsening by both coalescence and diffusive mass exchange between clusters, the clusters in the mean-field PDE are unable to move and coarsening occurs via the mass exchange of clusters. By introducing a new model for this mass exchange, we argue that the PDE exhibits dynamical metastability. We conclude by presenting careful numerical experiments that demonstrate the validity of our model.

15.
arXiv (CS.LG) 2026-06-16

Anomaly Detection via Mean Shift Density Enhancement

arXiv:2602.03293v2 Announce Type: replace Abstract: Unsupervised anomaly detection stands as an important problem in machine learning. Existing unsupervised anomaly detection algorithms rarely perform well across different anomaly types, often excelling only under specific structural assumptions. This lack of robustness also becomes particularly evident under noisy settings. We propose Mean Shift Density Enhancement (MSDE), a fully unsupervised framework that detects anomalies through their geometric response to density-driven manifold evolution. MSDE is designed as a general purpose anomaly detection framework, based on the principle that normal samples, being well supported by local density, remain stable under iterative density enhancement, whereas anomalous samples undergo large cumulative displacements as they are attracted toward nearby density modes. To operationalize this idea, MSDE employs a weighted mean-shift procedure with adaptive, sample-specific density weights derived from a manifold learning-based fuzzy neighborhood graph. We evaluate MSDE on an anomaly detection benchmark comprising 46 real-world tabular datasets, four realistic anomaly generation mechanisms, and six noise levels. Compared to 13 established unsupervised baselines, MSDE achieves consistently strong, balanced and robust performance for several standard classification metrics, at several noise levels and on average over several types of anomalies. These results demonstrate that displacement-based scoring provides a robust alternative to the existing state-of-the-art for unsupervised anomaly detection.

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

ScholaWrite: A Dataset of End-to-End Scholarly Writing Process

Writing is a cognitively demanding activity that requires constant decision-making, heavy reliance on working memory, and frequent shifts between tasks of different goals. To build writing assistants that truly align with writers' cognition, we must capture and decode the complete thought process behind how writers transform ideas into final texts. We present ScholaWrite, the first dataset of end-to-end scholarly writing, tracing the multi-month journey from initial drafts to final manuscripts. We contribute three key advances: (1) a Chrome extension that unobtrusively records keystrokes on Overleaf, enabling the collection of realistic, in-situ writing data; (2) a novel corpus of full scholarly manuscripts, enriched with fine-grained annotations of cognitive writing intentions. The dataset includes \LaTeX-based edits from five computer science preprints, capturing nearly 62K text changes over four months; and (3) analyses and insights into the micro-dynamics of scholarly writing, highlighting gaps between human writing processes and the current capabilities of large language models (LLMs) in providing meaningful assistance. ScholaWrite underscores the value of capturing end-to-end writing data to develop future writing assistants that support, not replace, the cognitive work of scientists.

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

When Multiple Scripts Matter: Evaluating ASR in Clinical Settings

Automatic speech recognition (ASR) in non-English clinical settings is challenged by multiscript variability, where the same term may appear in multiple valid orthographic forms. Conventional string-matching evaluation metrics often underestimate ASR performance by treating orthographic variants as errors. To address this issue, we introduce MultiClin, a clinical ASR benchmark designed to evaluate robustness to multiscript variability. Experiments across diverse ASR models show that multiscript-aware evaluation provides a fairer assessment of recognition quality than conventional single-reference evaluation. We further investigate the impact of script consistency during training and find that inconsistent script mappings increase orthographic uncertainty and hinder model convergence, with a balanced 50% mapping ratio producing the highest entropy. In contrast, script unification consistently yields the best ASR performance. Our dataset and code are publicly available at: https://github.com/aitrics-ronaldo/Interspeech_MultiClin.

18.
bioRxiv (Bioinfo) 2026-06-20

RNAStabFormer: Region-Aware Multi-Task Hybrid Learning for RNA Stability Prediction from Pulse-Chase Transcriptomics

Authors:

RNA stability is a central layer of post-transcriptional gene regulation, yet large-scale stability labels derived from pulse-chase transcriptomics depend strongly on quantification region, time-window definition, and replicate quality control. We present RNAStabFormer, a controlled learning framework for predicting human RNA stability proxies from transcript sequence. Its core model, RAMHT, combines region-specific nucleotide Transformer encoders for CDS, and sequence, a CDS codon stream, engineered sequence-grammar features, gated fusion, and four task-specific regression heads. We construct four strict consensus labels from ENCODE BrU-seq/BruChase-seq data by crossing gene-sense and exon-sense quantification with late-chase 6 h/2 h and total-chase 6 h/0 h retention ratios, and evaluate all models on fixed repeated-random and chromosome-holdout splits. Across chromosome holdouts, XGBoost remains the strongest standalone model, with median Pearson correlations of 0.504, 0.544, 0.546, and 0.778 on the four labels. RAMHT is competitive with raw-sequence deep models but does not universally exceed engineered-feature baselines. A strict nested RAMHT–XGBoost blend nevertheless improves gene total-chase prediction by 0.017 mean Pearson and exon late-chase prediction by 0.004 mean Pearson over XGBoost. Region and mechanism analyses show that CDS, local k-mer composition, and codon-sensitive signals dominate predictive information. RNAStabFormer therefore provides both a multi-task neural model and a leakage-controlled evaluation protocol for RNA stability prediction from pulse-chase data.

19.
Nature (Science) 2026-06-10

Mitochondria directly interact with the nuclear pore complex

Mitochondria regulate cellular processes through direct and indirect interactions with other organelles. A well-studied example has been contact with the endoplasmic reticulum at mitochondrial-associated endoplasmic reticulum membranes1, which control pathways including redox and calcium homeostasis2,3. Recent studies have also reported direct mitochondria–nuclear membrane contacts in cancer cells and yeast that promote pro-survival signalling4,5. Here we identify direct interactions between mitochondria and nuclear pores. Using two unbiased proteomic screens, GST pulldown and BioID, we found that VDAC1 was the top mitochondrial candidate that interacts with the filamentous nuclear pore protein RANBP2. In vitro RANBP2 CRISPR knockout,&nbsp;RANBP2 truncation&nbsp;or site-directed mutagenesis of RANBP2–VDAC1 interacting amino acids resulted in reduced mitochondria–nucleus proximity and decreased nuclear ATP and phosphocreatine levels. This was accompanied by a decline in the levels of the nuclear phosphoproteome and downregulation of pathways involved in histone modification, cellular differentiation and transcriptional regulation in vitro. Moreover, deletion of the RANBP2 C-terminal domain in vivo in mice resulted in embryonic lethality due to cardiac and neural crest differentiation defects. Collectively, these results describe a mechanism by which mitochondria directly interact with the nuclear pore complex, a phenomenon critical for regulation of nuclear energetics and cellular differentiation. Undoubtedly, additional roles of this interaction remain to be revealed. Mitochondria interact directly with the nuclear pore complex via VDAC1–RANBP2&nbsp;binding to sustain nuclear ATP levels.

20.
arXiv (CS.CL) 2026-06-15

LLMs Contain Multitudes: How Deployment Context Reshapes Model-Level Preferences and Values

Large language models (LLMs) are increasingly characterised in recent evaluation work as having stable, model-level preference and value systems. However, accompanying robustness checks are limited to incidental prompt perturbations such as syntax variation and option reordering. This leaves open whether the measured properties survive when the surrounding task context changes, as it does in most real deployments. We test this directly across two established pairwise paradigms: ranking country preferences and eliciting utility judgements. In both, we make the deployment context – the high-level task the model is performing while making concrete value-dependent choices – our controlled variable, varied across framings such as writing a Reddit post or a news article. Across five LLMs and over 1.2M pairwise decisions, deployment context produces variation far larger than prompt paraphrasing and temperature controls. In country preference rankings over 15 countries, context induces widespread, statistically significant rank shifts; the aggregate Global North favouritism reported in prior work is itself context-dependent, with each model's bias shifting systematically across contexts. In utility elicitation over 50 outcomes, broad cross-category ordering is preserved, but fine-grained rankings within domains vary substantially, and cardinal exchange rates between outcomes (e.g. how many lives in one region equal one in another) shift by a factor of 2.47 at the median. Reported model-level preferences and utilities are therefore better understood as context-conditioned measurements than fixed model-level properties: safety guarantees obtained under one framing provide limited assurance in another.

21.
arXiv (CS.AI) 2026-06-11

Towards Responsibly Non-Compliant Machines

arXiv:2606.12147v1 Announce Type: new Abstract: We consider the problem of engineering autonomous intelligent agents that are capable to responsibly not comply with user requests. We argue that machine non-compliance comes in many different forms, and sketch the issues we should pursue on the road of accomplishing responsibly non-compliant intelligent machines. We anchor responsible non-compliance in justifications for task refusal, pathways to override the non-compliance, as well as careful tracking of security risks and liability transfers.

22.
arXiv (CS.CL) 2026-06-18

Output Vector Editing for Memorization Mitigation in Large Language Models

Large language models memorize and reproduce sequences from their training data, creating privacy, copyright, and security risks. Existing neuron-level mitigation methods equate editing with zeroing out neuron activations, but the activation only controls whether a neuron engages; the output vector is what writes to the residual stream and, through superposition, encodes multiple features. We propose output vector editing, a constrained-optimization weight edit that locates a small set of MLP neurons responsible for a memorized continuation and minimally modifies their output vectors to introduce a distractor in vocabulary space, redirecting their residual-stream contributions while leaving activations unchanged. Evaluating on four models from 360M to 7B parameters (SmolLM-360M, OLMo-1B, OLMo-7B, Llama2-7B), we center on OLMo-7B (whose open weights and pretraining corpus enable systematic mining) and mine 6831 memorized sequences, achieving up to 87.9% suppression. The 2.7$\times$ gap over zero ablation on the same located neurons shows the suppression comes from the output-vector edit, not localization alone. Four edit modes span a spectrum from aggressive suppression to minimal redirection; in ensemble they cover 96.5% of memorized sequences, while our recommended single-mode configuration reaches 81.5% with no catastrophic locality failures. We further identify a mechanistic boundary at ${\sim}14%$ of sequences unreachable by MLP-only editing; while these failures are not attention-driven overall, ablating the top contributing attention heads recovers 60–64% of them, with stronger recovery on continuations that copy tokens from the prefix, positioning attention as a complementary fallback rather than a primary mechanism. Edit mode ordering and the success-locality trade-off transfer across all four models, with success rates scaling with model size rather than family.

23.
arXiv (CS.LG) 2026-06-12

Dense Supervision, Sparse Updates: On the Sparsity and Geometry of On-Policy Distillation

arXiv:2606.13657v1 Announce Type: new Abstract: On-policy distillation (\textsc{OPD}) has recently become a prominent post-training recipe as it combines two desirable ingredients: on-policy student trajectories and dense teacher supervision, yet how this hybrid changes a model's parameters remains unclear. Across several language and vision-language model pairs and use cases, our analysis yields two main findings. On sparsity, \textsc{OPD}-style updates are small and coordinate-sparse. They are distributed across layers and are usually FFN-heavy. This sparse structure is operationally useful: training only the discovered subnetwork recovers nearly the same performance as full \textsc{OPD}. However, the sparsity-inducing SGD optimizer underperforms AdamW in our optimizer ablation, likely because dense teacher supervision preserves heterogeneous coordinate-wise gradient scales where AdamW's adaptive scaling remains useful. On geometry, the updates are numerically full-rank but spectrally concentrated; they lie mostly away from the principal singular subspaces of the source weights and fall disproportionately on coordinates where the source weights are close to zero. These findings suggest that dense teacher supervision does not turn \textsc{OPD} into ordinary dense parameter rewriting; instead, \textsc{OPD} retains important geometric signatures of on-policy post-training.

24.
arXiv (CS.AI) 2026-06-17

MemTrace: Probing What Final Accuracy Misses in Long-Term Memory

arXiv:2606.17328v1 Announce Type: new Abstract: LLM agents increasingly maintain long-term memory of user facts across sessions. Yet such memory is usually evaluated by aggregating accuracy over question rows or episodes. Because this approach scores question rows independently, even when several questions probe the same fact, it cannot show how that fact behaves as conditions change. We introduce MemTrace, a benchmark whose unit of measurement is the knowledge point: a single typed fact about the user, rather than an individual question. MemTrace probes each fact along three controlled dimensions: memory age, defined by how many sessions ago the fact appeared in the history; question type, covering current state, earlier state, and trajectory of change; and evidence condition, covering present, missing, and contradicted-by-false-premise settings. Evaluating 13 memory-system configurations across four paradigms, we find that similar pooled accuracy hides different failures: recovering a fact's current and earlier states does not imply tracking how it changed, and safe abstention does not imply correcting a false premise. The dominant bottleneck is evidence use, not retrieval: when systems fail, the evidence was retrievable 10 times more often than it was missing. These results suggest that improving long-term memory requires better use of reachable evidence, not simply more storage or retrieval.

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

Using Seismic Statistical Features and VQ-VAE to Improve Spatiotemporal Seismicity Predictability

arXiv:2606.10069v2 Announce Type: replace Abstract: In this paper we build upon a previous study in which we demonstrated, using XGBoost and earthquake catalogue data from Japan and Chile, that a set of 60 seismic statistical features (SSFs) had much greater predictive value than a set of 428 generic time series features from the tsfresh package. We here extend this previous work in two key ways, focusing on data from Japan as a large dataset is necessary in order to allow for the training of a deep learning (autoencoder) model. First, we move from whole-region prediction (considering, for each candidate event, the likelihood of an event M $\geq$ 5.0 anywhere in the region in the next 15 days) to localised predictions in which both the region of feature computation and the region of prediction are restricted to a circle of radius 24 km around the candidate event, and we show that performance remains excellent, similar to our previous whole-region study for the same area. Second, we here couple this proven set of SSFs, based on one-dimensional (catalogue) data, with a novel feature based on two-dimensional seismic maps, obtained by training a VQ-VAE model to reproduce such maps as output and identifying a measure of its error in doing so with a localised build-up of crustal stress. We show that while localised prediction based on SSFs can be effective alone, with test AUC values as high as those obtained in the case of Japan in our previous whole-region study, the inclusion of the new natively-spatial VQ-VAE-derived feature, top-ranked by SHAP analysis, can enhance performance and additionally appears to near-wholly replace the traditionally-computed $b$-value in terms of feature usage.