Terms of endearment? Bias in first-name eponyms for species named after people
Letter to the Editor
Academic Intelligence · Curated Daily
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。
Letter to the Editor
The heterogeneity of expression levels among genetically identical cells, termed gene expression noise, is a property of the gene expression process whose importance in the biology of organisms and their evolution is increasingly recognized. Measuring gene expression noise requires single-cell expression data, as obtained from single-cell RNA sequencing (scRNASeq). Its estimation, however, is challenging owing to (i) the presence of technical noise in addition to biological noise, and (ii) the heterogeneity of cell types in the sampled population. We propose a maximum-likelihood framework to infer biological noise from scRNASeq data, while accounting for technical noise, dropout probabilities, and distinct cell sequencing depths. We demonstrate the parameter identifiability using simulations and that the resulting noise estimates are uncorrelated from the mean gene expression, and therefore do not need extra correction in downstream analyses, easing intra- and inter- genome comparisons. Using two technical replicates of scRNASeq data from the wild yeast *Saccharomyces paradoxus*, we show that expression noise can be inferred in a reproducible manner.
arXiv:2602.18575v3 Announce Type: replace Abstract: We prove, within the probabilistic framework of Khinchin families, that the generating function $P_k$ of partitions into $k$-th powers is strongly Gaussian in the sense of Báez-Duarte, and even further that it is a Hayman function. Thus the Hardy–Ramanujan asymptotic formula for the number $p_k(n)$ of partitions of $n$ into $k$-th powers which reads \[ p_k(n) \sim \frac{\alpha_k}{n^{(3k+1)/(2k+2)}} \exp\!\Big(\beta_k\, n^{1/(k+1)}\Big), \qquad n\to\infty, \] where $\alpha_k$ and~$\beta_k$ are explicit constants depending only on $k$, follows directly from Hayman's asymptotic formula for strongly Gaussian power series. The proof of strong Gaussianity of $P_k$ combines a Gaussianity criterion for Khinchin families with certain bounds of Tenenbaum, Wu and Li on the generating function; the asymptotic formula is recovered by computing asymptotic approximations of the mean and variance of the associated family. Analogous results are presented for the generating function $Q_k$ of partitions into distinct $k$-th powers.
Background: Cancer survivors face elevated risks of mortality from cardiovascular disease (CVD). The potential importance of physical activity (PA) and other behaviours across the 24-hour day (e.g. sedentary behaviour (SB) and sleep) for CVD-mortality risk is not well understood in this at-risk population. Objectives: To assess the importance of 24-hour movement behaviour, using a compositional approach, for mitigating CVD-mortality amongst cancer survivors. Methods: Participants with a prior cancer diagnosis were drawn from the UK Biobank accelerometry sub-study (n=6,158). Accelerometer-derived movement (moderate-to-vigorous PA (MVPA), vigorous PA (VPA), moderate PA (MPA), light PA (LPA), SB, sleep) was examined in relation to CVD-mortality, identified from health record linkage data (using Fine-Gray Cox proportional-hazards models adjusted for demographic, health, lifestyle covariates). Results: Median follow-up was 8.0 years (Q1-Q3: 7.4-8.5), with n=500 (8.2%) deaths (CVD-deaths: n=118). Greater MVPA, in place of any other behaviour, was inversely associated with CVD-mortality with e.g. 10% lower hazard if MVPA theoretically replaced 7 minutes (mins)/day SB (Hazard ratio (HR): 0.91, (95% Confidence Interval: 0.86-0.95)), 9 mins/day LPA (HR: 0.90, 0.83-0.97), or 11 mins/day sleep (HR: 0.90, 0.83-0.97). The VPA component of MVPA proved critical, requiring only ~1-2 additional mins/day for equivalent hazard reduction. Sleep duration, was also inversely associated with CVD-mortality. A 10% lower hazard required replacing 29 mins/day of SB with sleep (HR: 0.90, 0.84-0.96); no other behavioural replacement amongst SB, sleep or LPA could provide an equivalent risk reduction. Conclusions: Among cancer survivors, the most potent reduction in CVD-mortality followed theoretically reallocating time to higher intensity movement.
arXiv:2606.24504v1 Announce Type: new Abstract: We discuss reasons why the scaling exponents of current Large Language Models (LLMs) applications are indicating an unsustainable regime in terms of energy resources. We further show that attributing the smallness of such exponents to a numerical bias due to the neglect of a non-zero value of the loss function in the limit of infinite data (``pedestal effect") does not remove the unsustainability issue. Finally, the effects of the smoothness (roughness) of the data on the scaling exponents is commented upon based on an analogy with phenomenological models of fluid turbulence.
arXiv:2606.15767v1 Announce Type: cross Abstract: Understanding when and why deep neural networks are uncertain is crucial for deploying reliable machine learning systems in safety-critical domains. While existing uncertainty quantification methods provide scalar measures of model confidence, they offer limited insight into which spatial regions of an input contribute to different types of uncertainty. We propose a novel visualization framework, Uncertainty Activation Map (UAM), that combines Evidential Deep Learning (EDL) with Full-Gradient Class Activation Mapping (FullGrad) to generate interpretable spatial uncertainty activation maps. Our approach distinguishes between two fundamental types of uncertainty: vacuity, representing lack of evidence, and dissonance, capturing conflicting evidence between competing hypotheses. By leveraging the complete gradient decomposition property of FullGrad and the principled uncertainty quantification of Subjective Logic, our method produces theoretically grounded visualizations that highlight specific image regions responsible for model uncertainty. With this framework, vacuity and dissonance activation maps are generated by computing belief-weighted attributions, enabling identification of where models lack knowledge versus where they encounter ambiguous evidence. Extensive evaluations across multiple benchmark datasets demonstrate that the proposed framework effectively addresses the critical gap between uncertainty quantification and explainability, providing intuitive visual feedback to assess model reliability in complex visual recognition tasks.
Long-context language modeling requires not only extending context windows but maintaining coherent understanding of entity states and relationships across thousands of tokens – a challenge that semantic similarity alone cannot address. KGERMAR addresses this by constructing dynamic, context-specific knowledge graphs from input text during inference, enabling domain-adaptive retrieval that leverages both semantic similarity and explicit entity relationships. The framework performs real-time entity and relation extraction to build contextual knowledge graphs, then integrates graph-structural embeddings with textual semantics through a multi-component memory architecture. Three memory banks – contextual, semantic, and structural – are maintained with retrieval signals fused via learned weights to capture both surface-level semantics and deeper relational patterns. Evaluated on SlimPajama (84.7K training examples), WikiText-103 (4,358 examples), PG-19 (100 examples), and Proof-pile (46.3K examples), KGERMAR achieves up to 8.5\% lower perplexity and 2–2.5x better memory efficiency than memory-augmented baselines across context lengths from 1K to 32K tokens, with superior in-context learning performance across five NLU tasks. The dynamic knowledge graph construction approach advances memory-augmented language modeling by enabling domain-specific knowledge representation that adapts to input contexts rather than relying on fixed knowledge bases.
Recent advances in Large Language Models (LLMs) have substantially transformed Automated Essay Scoring (AES), yet the internal mechanisms underlying LLM-based scoring remain poorly understood. In this work, we systematically analyze the hidden representations of eight LLMs across two English essay datasets (ASAP++, CSEE) and one Portuguese dataset (ENEM). Using linear probing, cross-prompt generalization, dimensionality reduction, and neuron-level analyses, we find consistent evidence that essay quality information is encoded in a linearly accessible form within LLM representations. These representations emerge progressively across layers, remain robust across prompting strategies, and partially transfer across essay prompts despite differences in scoring rubrics. In addition, nonlinear probes provide only marginal and inconsistent improvements over linear probes, suggesting that most essay quality information is already linearly decodable. We further identify individual ``essay scoring neurons'' whose activations strongly correlate with essay scores and whose behavior is sensitive to targeted intervention. Moreover, the layer-wise distribution of these neurons systematically shifts with essay length, with longer essays relying more heavily on deeper layers. Overall, our findings provide evidence that LLMs encode structured representations related to essay quality and offer new insights into the interpretability of LLM-based AES systems.
arXiv:2606.17235v1 Announce Type: cross Abstract: Machine Learning (ML) models for grain growth prediction are typically trained on idealized synthetic data, yet practical applications require generalization to conditions outside the training distribution. This study evaluated the Out-Of-Distribution (OOD) generalization capability of the trained model from our previous study across three test cases, including experimental microstructures, microstructures characterized by a bimodal grain size distribution, and abnormal grain growth. To further probe whether physics-informed architectural design could improve robustness under these different conditions, a boundary-masked attention mechanism was proposed specifically for grain growth, constraining attention to grain boundary pixels. Both the baseline and the proposed physics-informed attention model were evaluated without retraining or fine-tuning on the OOD data. Both models successfully generalized to all three test cases, yet the boundary-masked attention mechanism provided substantial improvements, with the most notable gains for microstructures characterized by a bimodal grain size distribution, where Structural Similarity Index Measure (SSIM) improved from \num{0.6221} to \num{0.7609} and mean grain size ($\overline{R}$) error decreased from \operatorname{SI}{8.75}{\percent} to \operatorname{SI}{3.57}{\percent}. The attention heatmap analysis revealed that the boundary-masked attention model learned to concentrate attention on large grain boundaries in a manner consistent with curvature-driven grain growth physics, emerging from training without being explicitly encoded into the architecture. These results indicate that models trained on synthetic data can generalize to diverse OOD conditions without retraining, and that physics-informed attention may improve accuracy when the boundary morphology matches the training domain.
We present a survey on 4D generation and reconstruction, a fast-evolving subfield of computer graphics whose developments have been propelled by recent advances in neural fields, geometric and motion deep learning, as well as 3D generative artificial intelligence (GenAI). While our survey is not the first of its kind, we build our coverage of the domain from a unique and distinctive perspective of 4D representations, to model 3D geometry evolving over time while exhibiting motion and interaction. Specifically, instead of offering an exhaustive enumeration of many works, we take a more selective approach by focusing on representative works to highlight both the desirable properties and ensuing challenges of each representation under different computation, application, and data scenarios. The main take-away message we aim to convey to the readers is on how to select and then customize the appropriate 4D representations for their tasks. Organizationally, we separate the 4D representations based on three key pillars: geometry, motion, and interaction. Our discourse will not only encompass the most popular representations of today, such as neural radiance fields (NeRFs) and 3D Gaussian Splatting (3DGS), but also bring attention to relatively under-explored representations in the 4D context, such as structured models and long-range motions. Throughout our survey, we will reprise the role of large language models (LLMs) and video foundational models (VFMs) in a variety of 4D applications, while steering our discussion towards their current limitations and how they can be addressed. We also provide a dedicated coverage on what 4D datasets are currently available, as well as what is lacking, in driving the subfield forward. Project page:https://mingrui-zhao.github.io/4DRep-GMI/
arXiv:2606.19172v1 Announce Type: new Abstract: Personal memory in a language model is two problems: content and reasoning skill. The brain keeps the two apart (a sparse, local engram in the hippocampus for each episode, a slow neocortex for the shared skills that interpret it), so a new fact need not overwrite everything else. Most personalization today keeps a user's facts outside the weights, in a natural-language memory file or a retrieval index. When facts are written into the model instead, the standard recipe is the per-user LoRA adapter, which does the opposite of the brain, folding content and skill into one global weight delta. Writing a user's facts as a LoRA contaminates text unrelated to them; writing the same facts as local Engram rows leaves it mathematically untouched, resulting in a roughly 33,000x smaller memory footprint. We therefore propose User as Engram: store a user's content as surgical edits to the hash-keyed memory table of an Engram model, and carry the reasoning skill in one shared adapter. This layered design matches per-user LoRA's direct recall while delivering 5.6x higher indirect-reasoning accuracy on average, and never makes a single user worse at reasoning than the untouched base. The edit is a glass box: writing a fact switches on its lookup at exactly the trigger, adds the value the answer needs, leaves every other position unchanged to the last bit, and fails if written into the wrong layer. Because different users' facts land in disjoint hash slots, their edits compose: many users live in one shared table at once, stacking additively and losslessly, where a per-user LoRA, a single global weight delta, admits only one. Upon retrieval, a per-user Engram table does not grow with the population the retriever must search, so past ~100 facts it overtakes a retrieval pipeline on a 2.5x larger model.
arXiv:2606.15458v1 Announce Type: cross Abstract: Variational inference (VI) is a core engine of modern AI, enabling scalable approximate Bayesian learning and uncertainty-aware training of large probabilistic and generative models. In this paper, we propose Structured Nonparametric Variational Inference (SN-VI), a novel framework for modeling complex dependencies among latent variables in posterior approximation, leveraging multivariate spline techniques. Unlike traditional methods that rely on the mean-field assumption, SN-VI preserves intricate latent variable dependencies, providing a flexible and accurate approximation of posteriors with arbitrary shapes. We establish rigorous theoretical guarantees, including the derivation of the lower bound for the variational objective and proof of asymptotic consistency in posterior estimation. To facilitate practical implementation, we develop an algorithm that automatically identifies dependent latent variables and their underlying dependence structure, without requiring manual specification. Simulation studies validate the effectiveness of SN-VI in approximating posterior distributions with bounded support and complex dependencies. The proposed method has been successfully applied to high-dimensional structured data, including computer vision datasets and spatial transcriptomics. In these applications, SN-VI demonstrates improved generative model performance and effectively uncovers coupled biological signals through the learned dependency structure.
arXiv:2510.11228v2 Announce Type: replace Abstract: The present paper is devoted to the study of mean-field backward stochastic differential equations (MFBSDEs) with double mean reflections whose generators are not Lipschitz continuous. With the help of the Skorokhod problem and some a priori estimates for MFBSDEs, we establish the existence and uniqueness results for doubly mean reflected MFBSDEs.
Out-of-distribution (OOD) detection in dynamic open-world environments requires a model to continually adapt to evolving data distributions while generalizing to covariate-shifted inputs and rejecting semantic-shifted OOD examples. Most existing OOD detection methods optimize only the current-step objective and do not explicitly account for how post-deployment environment changes affect future OOD behavior. In this paper, we establish a theoretical grounding for dynamic OOD detection using a reinforcement learning (RL)-guided optimizer that explicitly favors updates that reduce the semantic OOD false positive rate over time. We develop a novel augmented optimizer that uses an RL-guided correction term on top of standard gradient descent (GD) and show its improvement over both future-domain generalization and semantic-OOD rejection. We analyze temporal error decomposition in terms of model-change and environment-change generalization errors and develop a new theoretical framework for comparing the generalization errors under both GD and RL-guided optimizers.
Long-term memory promises LLM agents that grow more capable across sessions, maintaining an accurate, evolving understanding of the user that interaction forms. In practice, however, this memory is evaluated mostly through downstream behavior, such as later answers, personalization quality, or task success, which tests that understanding only indirectly and leaves the memory artifact itself largely unaudited. We argue that long-term memory should instead be evaluated as an auditable post-interaction artifact: after ordinary assistance, what structured user state can be reconstructed from the memory the agent leaves behind? We instantiate this view in MEMPROBE, a benchmark in which a memory-equipped agent assists simulated users, each carrying a hidden, taxonomy-anchored user-state bank, across a trajectory of leak-controlled tasks, after which that bank is reconstructed from the agent's resulting memory under both full-store and top-k access. Built on synthetic ground truth for efficient, scalable measurement, MEMPROBE spans 50 simulated users with 31 hidden dimensions each (1,550 recovery targets) and tests 5 representative memory systems. Testing state-of-the-art memory agents, we find that successful assistance and recoverable memory behave as distinct capabilities. Task completion nearly saturates, even for a memoryless baseline, while category-balanced recovery stays moderate (about 0.6) and drops further under top-k retrieval. MEMPROBE is the first benchmark to study memory recovery directly, reconstructing the user state a system retains and scoring it against ground truth. We see recovery as a concrete objective for future memory agents to optimize, and MEMPROBE as a step toward an environment where agents are trained to remember their users, growing more faithful the longer they know them.
arXiv:2606.08594v2 Announce Type: replace Abstract: Deep learning EEG denoising architectures have scaled from tens of thousands to tens of millions of parameters, yet no prior study has isolated model capacity as the experimental variable or tested whether reconstruction metrics predict downstream neural-signal utility. We address both gaps by fixing architecture, loss, data split, and training recipe while sweeping only channel width from 1.05K to 40.26K parameters in a minimal depthwise-separable convolutional U-Net. Models were evaluated on the EEGDenoiseNet benchmark, cross-dataset BCI transfer tests, controlled baseline retraining, and downstream motor-imagery classification with five decoder families across all nine BCI Competition IV-2a subjects. Reconstruction performance saturated by 3-6.5K parameters, with post-elbow gains of at most 0.015 correlation coefficient per log10-parameter unit. An 8.46M-parameter baseline retrained under the same pipeline matched the 40.26K compact variant on EOG–a 200x parameter gap yielding no advantage–while a Patch-Transformer control reproduced the same diminishing-return shape. Downstream evaluation exposed a classifier-dependent metric-utility gap: reconstruction-optimized denoising significantly degraded CSP+LDA classification across all nine subjects and three artifact types (best denoised accuracy 0.547 vs. 0.612 noisy baseline; Bonferroni p=0.0488), persisting on naturally recorded trials (Delta=-0.047; BH-FDR q=0.0049). End-to-end neural decoders showed variable or neutral effects. Standard EEG denoising benchmarks are saturated far below current model capacity, and reconstruction metrics do not predict BCI utility. Ultra-compact models at 33-46 KB and 1.27-2.61M FLOPs/segment are practical for edge deployment. These findings argue for capacity-controlled evaluation, harder task-aware benchmarks, and mandatory downstream validation.
arXiv:2606.18091v1 Announce Type: new Abstract: Collective behaviour such as flocking (the collective motion of a spontaneously formed group along a common direction) or aster formation (the binding of opposing flocks, inhibiting each others motion) are intriguing emergent phenomena in active systems with local alignment rules. Until recently, their occurrence was mainly studied for classical systems, a prime example being the active Ising model (AIM), which translates the main ingredients of flocking and aster formation (i.e., alignment and self-propulsion) to a lattice framework. Here we introduce and study a one-dimensional (1D) quantum lattice variant of the AIM, based on ideal bosons with a spin degree of freedom. We find that both the collective behaviours of the 1D classical model, flocking and aster formation, are markedly enhanced by the bosonic quantum statistics. This contrasts with a recent quantum generalization of the AIM based onto hard-core bosons [Khasseh et al., Phys. Rev. Lett. 135, 248302 (2025)], where flocking, but neither its quantum-statistical stabilization nor aster states were observed as a consequence of interactions. Moreover, we investigate the competition of this quantum statistical stabilization of collective phases with their suppression by the quantum fluctuations induced by a transverse external magnetic field.
Stereo-based 3D object detection still faces two critical safety challenges: real-time performance and open-set generalization. Existing stereo 3D methods typically achieve twice the accuracy of monocular methods but suffer from significantly lower inference speeds, making them unsuitable for real-time applications. Meanwhile, recent advances in open-world detection have introduced open-set and open-vocabulary algorithms in monocular 2D and 3D settings, yet stereo-based open-set detection remains largely unexplored. To bridge this gap, we propose DDStereo, a novel Dual-Decoder Stereo Transformer for real-time open-set 3D object detection. DDStereo features two lightweight decoder branches: one for open-set foreground 2D detection and the other for 3D attribute regression. These decoders share object-level queries to achieve unified target-level alignment. To enhance inference efficiency, we designed a compact disparity feature extractor and a streamlined decoder architecture. Experiments on public stereo 3D benchmarks demonstrate that DDStereo achieves state-of-the-art accuracy under both closed-set and open-set protocols. Notably, our method surpasses existing stereo 3D detectors in inference speed and, for the first time, achieves real-time performance comparable to monocular approaches.
arXiv:2602.21062v2 Announce Type: replace Abstract: We introduce a broad class of families of branching random walks on a countable set $X$, which we refer to as germ-monotone branching random walks (GMBRWs). The processes in each family are parametrized by a positive parameter $\lambda>0$, which controls the overall reproductive speed, and they are monotonically increasing in $\lambda$ with respect to the germ order, a notion that extends classical stochastic domination. This framework encompasses a wide range of models, including classical continuous-time branching random walks, as well as discrete-time counterparts of certain non-Markovian processes such as ageing branching random walks. We define a general notion of critical parameter $\lambda(A)$ associated with each subset $A \subseteq X$, which serves as a threshold separating almost sure extinction in $A$ from positive probability of survival in $A$. This unifies and extends the classical global and local critical parameters $\lambda_w$ and $\lambda_s$, which can be recovered as special cases. We then investigate how modifications of the reproduction laws, either on a finite set or on a more general subset of $X$, affect these critical parameters. Our results extend earlier contributions in the literature.
arXiv:2606.13747v1 Announce Type: cross Abstract: Accurate power estimation is important for understanding and optimizing CPU power behavior, yet practical workflows often rely on simulation-derived information or post-silicon analysis. In this work, we present BigPower, a hierarchical source-level surrogate model for fine-grained module-level power estimation during CPU design. BigPower leverages large language model-based representations together with architectural hierarchy, module connectivity, configuration parameters, and workload context to estimate module-level power consumption directly from source-level design information, without requiring additional simulation during inference. Experimental results in the open-source XiangShan processor family demonstrate practical fine-grained power estimation across diverse configurations and workloads, offering an efficient alternative to conventional simulation-based workflows.
Computational antibody design methods can generate large libraries of candidate binders for a target epitope, but prioritizing which candidates to test experimentally remains a major bottleneck. Existing scoring approaches, including physics-based affinity estimators, structure-prediction-derived confidence measures, and inverse-folding likelihood models, provide useful proxy signals but are not explicitly optimized for early enrichment of binders among many structurally similar candidates. Here we investigate epitope-conditioned antibody binder ranking as a dedicated learning problem and introduce AGZArank, a geometric deep learning framework trained with structure-derived synthetic supervision based on normalized pseudo-energy targets. On a benchmark of 45 experimentally validated antibody-antigen interfaces, AGZArank recovered the true binder within the top ten candidates in 44.4% of cases and showed stronger generalization on post-2021 structures than ProteinMPNN, ESM-IF, and PRODIGY. Ablation experiments indicate that ranking performance depends primarily on training scale and alignment between the optimization objective and retrieval-based evaluation, rather than architectural complexity alone. These results support candidate prioritization as a distinct and tractable problem in computational antibody design.
arXiv:2605.00327v3 Announce Type: replace-cross Abstract: In large language model (LLM)-based recommendation systems, direct preference optimization (DPO) effectively aligns recommendations with user preferences, requiring multi-negative objective functions to leverage abundant implicit-feedback negatives and sharpen preference boundaries. However, our empirical analyses reveal a counterintuitive phenomenon, preference optimization collapse, where increasing the number of negative samples can lead to performance degradation despite a continuously decreasing training loss. We further theoretically demonstrate that this collapse arises from gradient suppression, caused by the dominance of easily discriminable negatives over boundary-critical negatives that truly define user preference boundaries. As a result, boundary-relevant signals are under-optimized, weakening the model's decision boundary. Motivated by these observations, we propose DynamicPO (Dynamic Preference Optimization), a lightweight and plug-and-play framework comprising two adaptive mechanisms: Dynamic Boundary Negative Selection, which identifies and prioritizes informative negatives near the model's decision boundary, and Dual-Margin Dynamic beta Adjustment, which calibrates optimization strength per sample according to boundary ambiguity. Extensive experiments on three public datasets show that DynamicPO effectively prevents optimization collapse and improves recommendation accuracy on multi-negative preference optimization methods, with negligible computational overhead. Our code and datasets are available at https://github.com/xingyuHuxingyu/DynamicPO.
arXiv:2606.20544v1 Announce Type: new Abstract: Calibration aligns a model's predictive uncertainty with the frequencies of its empirical outcomes and is important for understanding and trusting reported probabilities. Recent work shows that enforcing calibration at the level of individual predictors can improve ensemble accuracy and calibration, with mixture-of-experts (MoE) models showing strong empirical improvements in particular; however, the conditions under which calibration helps MoE are not well understood. In this work, we study how MoE models behave under distribution shift, focusing on how routing mechanisms interact with expert-level calibration. We show that expert calibration is sufficient to ensure calibration of the overall model under a broad class of distribution shifts in hard-routed models, but is insufficient for calibrating soft-routed models. To address this, we propose an adversarial reweighting that penalizes calibration errors of the routed aggregate under distribution shift, and we demonstrate that it improves the accuracy-calibration tradeoff both on average and on difficult subsets of the data, across model classes, prediction tasks, and distribution shifts.
Distilling the tool-use capabilities of large language models (LLMs) into small language models (SLMs) is essential for their practical application. The predominant approach, supervised fine-tuning (SFT), suffers from poor out-of-domain (OOD) generalization due to its rigid alignment with static teacher trajectories. While reinforcement learning (RL) offers an alternative, the capacity limitations of SLMs pose a severe dilemma: sparse outcome rewards provide insufficient guidance, whereas strict trajectory matching imposes overly restrictive constraints. To bridge this capacity-driven gap, we propose MENTOR, which introduces a flexible yet process-aware reward structure. Instead of enforcing rigid replication, MENTOR uses the teacher's reference to guide tool-use behavior, balancing behavioral alignment with downstream performance. Extensive experiments on controlled executable-tool benchmarks demonstrate that MENTOR improves OOD tool-use performance compared to SFT and strict RL baselines. Our findings suggest that within verifiable tool-use environments, flexible tool-use alignment offers a more effective approach than strict trajectory replication for developing adaptable small models.
Autonomous vehicles must operate safely in the real world, where errors can have severe consequences. Although modern end-to-end driving policies excel in routine scenarios, their reliability is limited by the scarcity of safety-critical ``long-tail'' events in real driving datasets. These rare interactions define the practical safety boundary of the learned policy, yet they are difficult to collect at scale in the real world. Here we show that this fundamental limitation can be addressed by post-training pre-trained driving models on synthesized high-stakes interactions. We introduce World Engine, a generative framework that reconstructs high-fidelity interactive environments from real-world logs and systematically extrapolates them into realistic safety-critical variations. This paradigm enables reinforcement-based post-training to align policies with safety constraints, circumventing the physical risks inherent in real-world exploration. On a public benchmark built on nuPlan, World Engine substantially reduces failures in rare safety-critical scenarios and yields significantly larger gains than scaling pre-training data alone. Furthermore, when deployed on a production-scale autonomous driving system, the resulting policy reduces simulated collisions and demonstrates measurable improvements in on-road testing, showing that post-training on synthesized, safety-critical interactions offers a scalable and effective pathway to safer autonomous driving. The full codebase suite, including training, is released to the public.