Papers

We design adaptive overlapping temporal decomposition (AOTD), a parallel sequential quadratic programming (SQP) method for long-horizon nonlinear optimal control problems (OCPs). As designed, AOTD is the first overlapping temporal decomposition (OTD) algorithm that adapts the overlap size online, eliminating the need to fix a suitable overlap a priori. Specifically, at each iteration, it partitions the long horizon into overlapping nonlinear subproblems, performs one SQP step on each in parallel, and adaptively selects both the overlap size and the accuracy to which the resulting local KKT systems are solved. Two conditions govern the accepted step: an adaptive residual condition bounding the distance from the concatenated inexact direction to the exact Newton direction of the full problem, and a descent condition on the adaptive penalty parameters that certifies descent for a carefully chosen exact augmented Lagrangian. The residual condition is solver independent and accommodates both deterministic iterative solvers and randomized sketching solvers. Under standard regularity assumptions, we prove that the KKT residual converges to zero from any initialization and establish a local linear rate for sequences converging to a strict local minimizer. We validate AOTD on a power-system frequency regulation OCP and a Burgers PDE OCP, attaining a reduction in estimated FLOPs of approximately $2.5\times$ to the best fixed-overlap variant while converging across multiple horizon lengths with a single parameter setting.

High-fidelity simulations of time-dependent partial differential equations (PDEs) are computationally expensive, motivating data-driven reduced-order surrogates for many-query tasks such as uncertainty quantification, design optimization, data assimilation, and optimal control. However, existing surrogate models often exhibit poor temporal stability, which can lead to unstable rollouts and exploding gradients during backpropagation, especially in multistep long-horizon forecasting. To address this, we propose SMORE, a mesh-agnostic framework for model order reduction of time-dependent PDEs. Its latent dynamics are trained with Lyapunov-guided stability regularization, which promotes stable long-horizon rollouts. We provide theoretical guarantees under the stated structural assumptions. Beyond forecasting PDE evolution, the learned latent dynamics, which are interpretable and linear or linear-quadratic, could bring benefits for downstream tasks such as data assimilation and optimal control. Moreover, our framework is capable of predicting continuous PDE solution fields from sparse measurements of the initial condition. We evaluate SMORE on a range of problems, including wave propagation, the Navier-Stokes equations, and the shallow water equations. Our results show that it improves long-horizon rollout generalization and empirical robustness, and achieves competitive accuracy at comparable parameter budgets relative to competitive baselines including DINo, FNO, CNO, and Transolver.

Speech Language Models (SLMs) that understand spoken language questions support only a few high-resource languages, limiting access to millions of people worldwide. This gap stems from the scarcity of multilingual speech instruction-tuning datasets. We present MULTISPEECHQA, a large-scale, synthetically generated and human-verified dataset comprising 9200 hours of 10.8 million spoken question-answer pairs in 23 typologically diverse languages. Using MULTISPEECHQA, we also introduce MULTISPEECH-BENCH, a multi-task benchmark for evaluating SLM performance on 23 languages. We compare the performance of a cascading system to open-weight and closed SLMs on MULTISPEECH-BENCH and find that the cascading system outperforms open-weight SLMs but not all closed SLMs. We use MULTISPEECHQA to finetune Qwen 2.5-Omni, which improves its performance on our benchmark. Our findings show that high-quality synthetic datasets offer a cheap solution to improving the multilingual capabilities of SLMs.

Diffusion Transformers (DiTs) have established themselves as a scalable backbone for high-fidelity image synthesis. However, unlike U-Net based diffusion models that rely on rigid, hand-crafted skip connections, DiTs predominantly use a uniform residual stream that integrates all preceding layers as a monolithic state. In this work, we rethink residual connections in diffusion transformers and propose to transform them from passive summation into an active retrieval mechanism optimized for image denoising. First, we conduct a systematic analysis of DiT's internal representation, revealing a latent preference for early-layer feature reuse and symmetric layer guidance. Motivated by this, we introduce a structured connectivity design that explicitly integrates local residual connections with long-range pathways. Instead of static skip connections or dense all-layer routing, our method enables each transformer block to selectively “attend” to critical earlier representations, dynamically retrieving spatial and semantic cues through direct, differentiable cross-depth paths. Experiments show that our adaptive connectivity leads to faster convergence, with up to $1.73\times$ fewer training iterations, and significant gains in FID and visual quality with less than $0.1\%$ additional parameters, further improving a strong REPA-XL/2 model from $5.9$ to $4.34$ FID without guidance and reaching $1.39$ FID with classifier-free guidance. Our findings suggest that adaptive cross-layer connectivity is a critical yet underexplored factor in diffusion transformers, and that incorporating structured information pathways provides a simple and effective direction for improving scalable generative models.

Flow-matching models generate high-quality images and videos, but repeated neural network evaluations make sampling expensive. Skipping evaluations reduces this cost by extending an available velocity estimate over a longer span. However, local velocity agreement alone does not determine a suitable span, and checking each candidate endpoint adds costly model calls. We introduce ReAL, a training-free sampler that selects how far to advance using one shared lookahead. Our key insight is that the discrepancy between uncorrected and lookahead-corrected endpoint proposals can be computed directly from the observed velocity mismatch and the candidate span beyond the lookahead. This relation provides a span-dependent selection criterion without additional endpoint evaluations. The same lookahead selects the longest passing candidate span, corrects the accepted update, and supplies its velocity as the next starting estimate. After initialization, each regular iteration requires only one fresh evaluation. ReAL uses the pretrained velocity output and original noise schedule, with no additional training or access to internal features. Experiments cover four image-generation backbones, video generation, and image editing. ReAL achieves 4.91x measured speedup on FLUX.1-dev while retaining 97.0% of dense mean ImageReward. On HunyuanVideo, it achieves a 5.49x speedup while maintaining a VBench score close to that of dense sampling.

In this paper we fully categorize all hybrid-variable (HV) discretizations of linear advection equations according to their stability property and establish the stability barrier of these methods. In the HV discretization framework, we find numerical approximations to both cell-averaged solutions and nodal solutions and evolve them in time simultaneously. We prove that the HV methods have a similar stability barrier as the classical finite difference schemes, that the semi-discretized HV scheme is stable if and only if it uses a stencil such that the number of unknowns in the upwind direction is precisely that in the downwind direction plus one or two, with only one exception when the two numbers are three and zero, respectively. It is also proved that all central HV schemes are neutrally stable, in the sense that the L2-norm of the solution remain uniformly bounded at all times. The theoretical predictions are verify by extensive numerical tests.

On-policy self-distillation trains reasoning models on their own trajectories using dense token distribution guidance from a privileged teacher with access to a verified solution. This supervision transfers what the teacher predicts without directly transferring where it attends within the preceding context. We introduce On-Policy Attention Self-Distillation (OPASD), which complements token-level supervision with solution-conditioned attention distillation. Because the privileged teacher can attend to verified solution tokens unavailable to the student, OPASD projects teacher attention onto student-visible positions and renormalizes the resulting distribution before alignment. Across three model sizes and four competition-level mathematics benchmarks, OPASD consistently outperforms token-only OPSD, improving average accuracy by 4.98 to 8.40 percentage points. OPASD also avoids the response-length inflation and performance degradation observed with token-only distillation, reducing generated rollout tokens by 73.9% and estimated model compute by 72.6% while training 1.53x faster. These results show that solution-conditioned attention provides a complementary supervision signal that makes on-policy self-distillation more accurate, stable, and compute-efficient.

With the growing number of binary black hole (BBH) candidates in the Gravitational-Wave Transient Catalog 5.0 (GWTC-5.0), an increasing number of BBH systems with distinct properties have been reported. The individual spins are relatively less constrained in the parameter estimations for most events. While the LIGO-Virgo-KAGRA Collaboration (LVK) adopts a uninformative spin prior by default, some binary evolution scenarios have strong predictions on component spins. In this work, we perform Bayesian parameter inference for a subset of BBH events spanning a broad range of properties in the chirp mass, mass ratio, and effective inspiral spin ($\mathcal{M}_{\rm c}$-$q$-$\chi_{\rm eff}$) parameter space. We consider three different spin priors: the LVK spin prior and two astrophysically motivated priors, assuming a non-spinning primary BH ($\chi_1 = 0$) or a non-spinning secondary BH ($\chi_2 = 0$). We find that the non-spinning primary BH prior is disfavored by the data for several systems with high effective inspiral spins, including GW190517_055101, GW190412, GW241113_163507, and GW231028_153006. We suggest that standard common-envelope evolution represents a promising formation channel that can be tested with gravitational-wave observations, and propose a novel method for estimating the merger rate of systems formed through alternative evolutionary channels. We further suggest that the binary evolution involving mass-ratio reversal may provide a possible formation pathway for these systems.

Screened scalar fields provide a possible connection between dark energy and laboratory-scale new physics while remaining compatible with existing fifth-force constraints. Solar production offers a direct probe of these models, and ionization-only measurements extend the sensitivity of liquid-xenon detectors into the sub-keV regime. We present the first search for solar chameleons using low-energy S2-only data, based on the 7.83 tonne-year XENONnT ionization-only data set. The predicted electronic-recoil spectrum is folded through the run-dependent detector response and tested directly in corrected-S2 (cS2) space. A joint binned likelihood combines the seven cS2 bins of science runs SR0, SR1, and SR2. We sample the five-dimensional parameter space $\{\beta_e,\beta_\gamma,M_e,\Lambda,n\}$ and derive posterior constraints on the conformal and disformal electron couplings. The data extend the recoil-energy reach to approximately $0.04~{\rm keV}_{ee}$, where the few-electron detector response and instrumental backgrounds must be modeled explicitly. In the disformal-dominated regime, the result is expressed in terms of $\beta_{\rm eff}\equiv\beta_\gamma({\rm eV}/M_e)^4$. We obtain the upper limit $\log_{10}\beta_{\rm eff}<-3.64$. Although the present analysis reaches a lower energy threshold than our previous XENONnT electronic-recoil analysis, the resulting constraint is weaker, primarily because the disformal absorption rate decreases rapidly toward low recoil energies, where the background rate also becomes larger. This analysis establishes low-energy ionization data as a new experimental channel for testing solar chameleons and screened dark-energy models.

Vision-Language-Action (VLA) models provide a unified framework for grounding high-level semantic information into low-level robot actions, enabling scalable robotic manipulation across diverse tasks. However, existing VLA models lack explicit mechanisms to disentangle the states and intents of the two arms, leading to unintended cross-arm interference that degrades task execution success. To address this issue, we propose a symmetric Dual-Arm Expert (DAE) architecture built upon a shared Vision-Language Model (VLM) backbone with decoupled, arm-specific expert towers. Expert selection is carried out through a two-stage dual-arm intent routing scheme, in which experts are routed either by explicit language instructions in the first stage or by implicit visual semantics in the second stage. Moreover, we introduce a lightweight task progress prediction module that leverages cross-attention between the pre-chunk temporal features and semantic representations of proprioceptive and visual observations to accurately estimate frame-wise task completion progress. This module facilitates task progress synchronization to support coordinated scheduling for collaborative multi-robot tasks. Experimental results demonstrate the effectiveness of our model in dual-arm intent routing and the disentanglement of cross-arm interference, and further provide preliminary evidence of emergent skill generalization from single- to dual-arm tasks (as well as the reverse), together with cross-arm motion-domain skill transfer.

Evaluating large language models (LLMs) relies heavily on benchmark scores, yet aggregate metrics can obscure whether individual items serve their intended measurement purpose. BSDProbe provides an item-level framework for benchmark structural diagnosis that jointly characterizes correctness transitions and their supporting response evidence. Repeated responses along an empirically ordered model axis yield complementary correctness and normalized entropy trajectories, from which four-dimensional profiles of boundary position, boundary width, boundary-signal validity, and order consistency are constructed for benchmark-level aggregation. Experiments on six benchmarks identify varying proportions of potentially anomalous items. With Qwen3 as both the selection and reference axis, BSDProbe achieves a 17.9% mean relative improvement in normalized discrimination ratio over the best baseline per benchmark and matches or exceeds all baselines in ranking agreement with the full benchmark.

Vision-language models (VLMs) support multimodal fake news detection (FND) by producing explicit analyses. Recent methods further improve interpretability by organizing verification knowledge into explicit concepts. However, two questions remain: how to improve the reliability and applicability of verification concepts, and how to effectively apply reusable concepts to verify unseen news. We propose ReVR, a dual-path reasoning framework that constructs and applies reusable verification concepts for multimodal fake news detection. An agentic workflow grounds and consolidates candidate concepts, while statistical profiles characterize their historical behavior. During inference, a coverage-oriented path aggregates evidence from the complete concept library using a trainable encoder, while a query-focused path prompts a frozen VLM to reason over selected concepts and their observations. A learned conflict resolver selects between the two predictions when they disagree. Experiments on fake news benchmarks demonstrate the effectiveness of the method regarding detection performance and generalizability.

Matrix-valued optimizers such as Muon exploit the spectral structure of neural network updates through Newton–Schulz orthogonalization, but their near-flattening of the singular spectrum discards relative magnitude information across gradient modes. We introduce Soren (Spectral Orthogonal Reshaping), a matrix-valued optimizer that preserves the singular subspaces of the gradient while applying a bounded, monotone sigmoid transformation to its singular values. This smoothly compresses dominant modes without fully flattening the spectrum. We interpret Soren as a positive-definite preconditioned gradient method and establish convergence guarantees under relative smoothness and metric Polyak–Łojasiewicz geometry. To avoid explicit singular value decomposition, we further develop a finite-depth Soft Newton–Schulz (SNS) polynomial realization of the sigmoid spectral map and characterize how its spectral approximation affects the induced convergence geometry. Experiments across LLM pre-training, supervised fine-tuning, and direct preference optimization demonstrate the effectiveness and robustness of Soren against established optimizers.

How many directions in weight space does training need? The intrinsic dimension answers this with the smallest number of random directions in which training still reaches a target accuracy, and small values have motivated parameter-efficient methods such as LoRA. We measure it for variational Monte Carlo (VMC), which trains a neural network to represent the ground state of a quantum many-body system. VMC is a demanding test, because the network generates its own training samples and every gradient is noisy, and a revealing one, because the exact answer is known and every run can be scored. We train only a small latent vector that a frozen random map turns into the network's weights, with no change to the standard natural-gradient optimizer. We find that a small dimension can be misleading, while the stability it brings is real. On a magnet with a hard sign pattern, a network that cannot represent signs reaches its best energy in 8 of 28,642 directions, but only because no such network can go lower; once signs are learnable, neither the signs nor the magnitudes are cheap. The dimension rises across a quantum phase transition, so it tracks how difficult a state is at far less compute than fitting a scaling law, yet it never falls below a floor set by the random subspace itself, even where the ground state is nearly trivial. Training in the subspace, in contrast, never diverged in our experiments, whereas full-parameter training with the same settings did, and a control with matched solvers attributes the difference to the reduced dimension.

Scientific code generation can produce executable programs that fail to compute the intended scientific object. We study this problem in language-model synthesis of Clifford circuits, which prepare the stabilizer states used in quantum error correction and admit exact classical verification. In our target-conditioned framework, each target is given as compact signed stabilizer generators, and an exact verifier checks the generated OpenQASM circuits. We supervise models with Aaronson-Gottesman chain-of-thought (AG-CoT) traces checked by the verifier, and continue training on model generations that the verifier accepts. Across two independently trained model families (3B and 7B), AG-CoT supervision multiplies greedy-decode state-equivalence accuracy by four to six times over circuit-only baselines, and verifier-filtered continuation training adds a further consistent gain atop both. A complementary 32B study shows that supervised models achieve near-perfect syntax and Clifford validity while the strongest direct model reaches 6.14% state equivalence per target, rising to over 10% under verifier-guided selection with multiple candidates. These results show that algorithmic trace supervision gives a large, statistically significant gain in both model families and that verifier-filtered continuation adds a further repeated gain. The persistent gap between Clifford validity and state equivalence confirms that exact verification is necessary: a circuit can be syntactically and physically valid yet prepare the wrong quantum state.

We prove that the five-column recursive-line Zarankiewicz number satisfies $z_{RL}(m,5)=3m+5$ for every integer $m\ge 12$. The proof is based on an explicit $12\times5$ seed configuration combined with a recursive one-row extension scheme. The seed attains the five-column cell bound and contains no unoccupied cells. Two selected pairs in the seed are opened and replaced by parallel paths, ensuring that each inserted row increases the total number of augmented edges by three while preserving the required simple configuration. The main challenge lies in verifying the strengthened recursive-line condition ${\rm (RW3+)}$ uniformly across all extension lengths. To this end, we develop a distance-reducing rectangle lemma that transfers certified inner-product relations along extension paths, reducing relations between distant labels to those at smaller path distances. This confines the verification for arbitrarily long extensions to a finite collection of seed and interface certificates. The resulting construction successfully satisfies pair identification, preserves distinct selected-edge classes, and certifies the orthogonality of distinct edge representatives. Furthermore, combining this constructed lower bound with the parameter hierarchy yields $z_2(m,5)=z_{RL}(m,5)=3m+5$ throughout this range.

Driving decisions depend on both where to focus and how to act on what is seen. Effective driving reasoning must establish which objects matter, where they are, and how they inform the intended action. Text-based rationales can describe a driving response while leaving its correspondence to specific visual evidence implicit. Visual focus provides a concrete starting point for this connection by identifying what matters in the scene and where it is. We propose FocusDrive, a structured multimodal reasoning framework that organizes end-to-end planning around explicit visual focus. It pairs descriptions of decision-relevant objects with image-patch references, bringing explicit visual focus into the reasoning that generates driving plans and trajectories. We first assess this focus representation through driver gaze prediction, then investigate its role in planning reasoning using driving-focus annotations within existing NAVSIM training scenes. Experiments on W3DA and NAVSIM demonstrate competitive gaze prediction and end-to-end planning performance, with FocusDrive improving over text-based chain-of-thought. These results support visual focus as an effective link between scene understanding and driving action.

When the human reference scores zero on a metric, the released GTRS-Dense label generator for NAVSIM marks every candidate trajectory in the scene as passing it. NAVSIM's authors introduced this human-reference forgiveness to avoid penalizing contextually justified maneuvers when scoring one trajectory, and warned that it could overlook important failures. In label generation it sets a whole column of 16,384 candidate targets to passing. To measure the consequences for supervision, we re-run the generator with the overwrite disabled and compare the pre-overwrite targets with the released labels on all 103,288 navtrain scenes. The rule erases a candidate distinction that the training loss reads on 11,237 of them (10.8793%). Firing usually changes most of a column: lane keeping carries 9,982 of the 13,042 forgiven loss columns, and its median forgiven column had 14,391 of 16,384 candidates failing before the overwrite. On held-out navtest scenes forgiven on lane keeping, the released lane-keeping head's median AUC against the pre-overwrite outcome is 0.7095; on unforgiven scenes matched on failing-candidate count it is 0.9807. For the Hydra-MDP checkpoint released with GTRS, whose configuration takes the same label file, the two values are 0.6627 and 0.9761. Continuing the released GTRS-Dense checkpoint for 300 optimizer steps with three paired seeds, we observe the forgiven-scene AUC 0.1086-0.1251 higher with pre-overwrite than with published targets, and a narrower gap between matched groups, still above zero. Scoring with forgiveness disabled, we observe lane keeping higher by 2.478-3.524 points on navtest scenes forgiven on any of five loss metrics, with lower adjacent-frame plan consistency. Both changes are larger there than on the rest. EPDMS, scored the same way, does not separate the two target sets.

Entanglement harvesting transfers quantum field correlations to localized probes, yet whether these correlations can alter stable strategic behavior has remained unclear. We formulate a quantum-input game in which two Unruh-DeWitt detectors harvest vacuum entanglement from a free massless scalar field in $(3+1)$-dimensional Minkowski spacetime. A referee converts this state into payoffs using independent question qubits and phase-calibrated local Bell measurements. To leading perturbative order, the mean reward is proportional to the signed entanglement margin $\sqrt{\mathcal{H}^2+\mathcal{Q}^2}-\mathcal{P}$, where $\mathcal{H}$ is the Hadamard correlation contribution, $\mathcal{Q}$ is the Pauli-Jordan causal contribution, and $\mathcal{P}$ is the local excitation noise. Away from payoff degeneracies, allowing each player to exit or to choose an early or late interaction yields $1+2n_+$ trembling-hand perfect equilibria, where $n_+$ counts the positive-reward timing equilibria. For one fixed protocol, five analytically certified settings realize the sequence $7\to3\to1\to7\to1$. They represent the spacelike region, the first commutator lobe, central carrier cancellation, the second lobe, and the post-wavefront Huygens regime. Thus, within this fixed quantum-input verification protocol, distinct field-correlation regimes generate different stable strategic structures.

For an integer $r\geq 3$ and real numbers $a_1,\ldots,a_r\in(0,1)$, let $a=a_1\cdots a_r$. We show that \[ \prod_{i=1}^r(a_i+a_i^2+\cdots+a_i^r)\geq \prod_{i=1}^r(a_i+a_i^2+\cdots+a_i^{r-1}+a). \] This inequality enables us to bound the measure of $r$-cross $t$-intersecting families.

Evaluating a target policy using data generated by a different behavior policy remains a fundamental challenge in reinforcement learning. While most existing work relies on the standard one-step Bellman residual, we consider a convex combination of one-step and two-step residuals with a fixed mixing weight. In ideal settings, this mixed Bellman formulation can provide a natural bias–variance trade-off between approximation error under a restricted value-function class and the increased variance arising from multi-step importance weighting. To solve this mixed residual optimization, we adopt a minimax formulation involving a critic function. Unlike standard approaches that rely on a fixed functional class, we construct a data-dependent critic representation using predicted future feature directions which effectively induces a kernel adapted to the underlying transition dynamics. This allows the critic to focus on directions that are most relevant for the estimated Bellman error. To control overfitting, we use sample splitting to construct the critic and estimate the value function on separate data subsets. Simulation studies and MetaWorld tasks illustrate the effect of the mixing parameter and show that intermediate residual combinations can improve value estimation in challenging settings.

Let $\cX$ be a curve of genus $g\ge2$ and $p$-rank zero over an algebraically closed field of odd characteristic $p$, and let $G\le\Aut(\cX)$ satisfy $|G|>24g(g-1)$. We prove that $G$ fixes a point unless $(\cX,G)$ belongs to an explicit list. For $g\ge4$, the curve is a cyclic quotient of a Hermitian curve, of a Giulietti–Korchmáros curve, or of Skabelund's cover of a Ree curve. We determine the equations and all the groups $G$ that occur, including their central kernels. The exceptional $A_7$ action in characteristic $5$ is on the Hermitian curve of degree $6$; genera $2$ and $3$ are treated separately. The equations are recovered from the group action by two rigidity lemmas for equivariant one-point $p$-group covers of the affine line. If $G$ has no common fixed point and its Sylow $p$-subgroups are noncyclic, then, unless $G$ induces ${}^2G_2(3)$ on the orbit of Sylow fixed points, the bound $|G|>\max\{84(g-1),2g^2\}$ gives a rational quotient with exactly one wild and one tame short orbit, and a rational Sylow quotient.

Speculative decoding reduces sequential Target model calls by verifying multiple tokens from the Draft model in parallel. Yet KV Cache growth limits long-context serving under constrained GPU memory. Offloading KV to CPU memory relieves this pressure. However, existing offloading schemes restore the full KV history before attention and fail to fully exploit the benefits of KV sharing across queries within a verification round. Existing parallel speculative decoding methods also overlook idle GPU compute capacity while memory-bandwidth-bound Target verification waits for historical KV. We present SpecStream, a speculative decoding system that begins verification without waiting for the full KV history to be restored and exploits compute bubbles during KV transfers for concurrent drafting on the same GPUs. It offloads only Target-committed history, keeping candidate rollback local to the GPUs. Each streamed KV chunk serves all queries in the round, with online softmax preserving full attention. Target-priority scheduling controls Draft concurrent execution under resource limits to constrain interference with Target verification. Experiments show that SpecStream maintains task quality close to SGLang speculative decoding while supporting more concurrent requests under limited GPU memory. Across different datasets, it achieves average throughput speedups of 1.41$\times$ and 1.32$\times$ over the offloading baseline for Qwen3 and InternLM2.5, respectively. Compared with parallel speculative decoding on separate Target and Draft GPUs, SpecStream improves output throughput per GPU by an average of 55.4%.

Large Language Models (LLMs) suffer from temporal misalignment, often due to the contradictory nature of their training corpora. While current mitigation strategies rely on computationally expensive fine-tuning or context-heavy retrieval augmented generation (RAG), the internal mechanisms governing time-sensitive recall remain under-explored. Unlike prior studies that identify temporal components such as attention heads and MLP layers, we provide the first feature-level map of temporal recall by isolating individual MLP features via transcoder circuit tracing. We identify three node categories (common temporal, common to the year, and chrono-semantic) which interact to generate a temporal filter during factual recall. By analysing Gemma 2 2B, LLaMA 3.2 1B, and Qwen3-4B, we show that these features do not follow a simple linear pipeline but represent time through a parallel and mixed syntactic-semantic interplay across layers. We additionally discover a class of higher-layer temporal components invisible to existing EAP-IG methods, establishing transcoders as a more complete lens for temporal interpretability in time-sensitive factual recall. These findings present MLP components for potential targeted interventions in time-sensitive factual recall

Large language models (LLMs) are increasingly expected to adapt to individual users, yet effective personalization remains challenging when only limited user-specific samples are available. In this work, we take an alternative perspective: pretrained LLMs may already possess latent capacity for personalization, and a few user samples may therefore suffice to guide the model toward user-aligned behavior with minimal user-specific adaptation. From this perspective, we propose LatentPersonal, a framework that formulates personalization as navigation in a shared latent adaptation space. LatentPersonal infers a compact latent representation from a few user samples to guide user-specific model adaptation, regularized with a variational information bottleneck to encourage compact preference representations. We instantiate LatentPersonal with LoRA, leveraging its low-rank parameterization as a natural low-dimensional adaptation space for personalization. By simply inserting a user-specific guidance vector between the shared low-rank factors, the model can navigate toward personalized adaptations through lightweight inference of this compact representation, without updating the shared LoRA parameters. Experiments across multiple personalization datasets demonstrate that LatentPersonal substantially reduces user-specific adaptation overhead while achieving effective personalization from only a few user-specific interactions, with particularly strong performance in the one-shot regime.

Recent advances in online policy self-distillation (OPSD) have demonstrated that large language models (LLMs) can improve their capabilities by leveraging external privileged information (PI), such as manual annotations or feedback from external environments. However, obtaining accurate annotations and constructing sophisticated environments often require substantial human effort and computation, limiting the scalability of OPSD. While a few recent studies have explored self-improvement without external PI, the resulting gains remain limited. In this work, we explore whether LLMs can achieve comparable self-improvement without external PI. Our key observation is that a single LLM can support multiple reasoning modes, such as deep-thinking and non-thinking modes, with deep thinking generating additional information during reasoning. Based on this observation, we propose Self-Evolving Online Policy Distillation (SeOPD), which enables LLMs to distill and internalize information generated by their own chain of thought (CoT). Specifically, it (1) generates CoT with the deep-thinking mode, (2) produces responses with the non-thinking mode, and (3) uses the generated CoT as PI to provide token-level supervision for the non-thinking response, allowing new information inferred during reasoning to guide the non-thinking mode and be internalized into the shared model parameters, thereby improving both non-thinking and deep-thinking capabilities. Extensive experiments across LLMs and tasks demonstrate the effectiveness of SeOPD.

As recursive self-improvement (RSI) rapidly advances, reliable evaluation becomes critical for guiding adaptive search. RSI typically relies on finite evaluation resources, such as fixed benchmarks, to determine which modifications are retained and what is proposed next. However, when these finite resources are repeatedly reused, new candidates are proposed based on feedback from the same evaluation set, so the search trajectory can adaptively overfit and empirical improvement may not reflect genuine population improvement on the underlying task distribution. Some existing methods account for multiple comparisons but assume that candidates are chosen independently of the evaluation set, and therefore do not control this adaptive dependence. To address this, we propose REUSE (Risk-controlled Evaluation Under Sequential Evolution), a certified evaluation and promotion framework that allows a fixed evaluation set to support repeated adaptive decisions while providing statistical guarantees. For a user-specified error level $\alpha$, with probability at least $1-\alpha$, every promoted modification is a genuine population improvement on the underlying task distribution. REUSE achieves this by strictly limiting the evaluation feedback returned to the search process and accounting for possible promotion histories within the error budget. We develop detailed statistical theory for RSI evaluation in this setting, including simultaneous error control, valid lower bounds on cumulative improvement, and a characterization of the fundamental limits of adaptive evaluation reuse. In live self-improvement experiments, REUSE commits substantially fewer false promotions than evaluation frameworks from current RSI systems and error-controlled baselines, reducing the proportion of false promotions from up to 20.7% to 0%, while achieving final true population performance comparable to the best baselines.

Constrained-option scoring reads a model's probabilities for a fixed set of permitted answers, so it returns a score even when the model is about to write something else. We study a prompt that quotes a multiple-choice item, ending in the item's own answer instruction, and asks for a forecast of whether a reader model given only a short note will answer it correctly. The model can begin answering the quoted item: on three Qwen3 checkpoints an answer letter is the most probable token on nearly every passage, and the forecast ranks the reader's correctness indistinguishably from chance. Beyond prior work's argmax check and prefill repair, we contribute a label-free diagnostic of the declared options' support, their first-token mass in context; a deletion test tracing the Qwen3 answer-letter start to the quoted instruction; and rendering and tokenisation checks for dropped separators and coinciding first tokens. Prefilling an answer stem lifts the median option mass from below to above one half on all nine checkpoints scored with this prompt and, with options and renormalisation unchanged, raises the Qwen3-averaged AUC from 0.4929 to 0.6152. lm-polygraph's default P(True) estimator shows a related failure: it reads True where Qwen3 without thinking favours an answer letter (MMLU) or the prompt's (A)/(B) label (TriviaQA), and its Qwen3-averaged AUROC is indistinguishable from chance on MMLU and below chance on TriviaQA. Scoring and renormalising the (A)/(B) labels after an answer stem raises it to 0.632 and 0.868, renormalising True against False in place reaches 0.660 and 0.860, and on TriviaQA both exceed all fourteen default single-answer library estimators. The original forecast is renormalised too, yet stays indistinguishable from chance: option mass flags low support in both settings without labels, and only checking the intended target shows which score still ranks.

Molecularly guided pathology foundation models (PFMs) exploit transcriptomic or proteomic information to enrich whole-slide image (WSI) representations, yet effectively leveraging large standalone molecular corpora remains challenging. First, existing molecular foundation models encode protein sequences or single-cell states, not the patient-level bulk expression profiles paired with WSIs. Second, because cross-modal supervision is restricted to paired WSI-omics samples, knowledge from standalone molecular corpora reaches the pathology encoder only indirectly, creating a paired-support bottleneck. To address these challenges, we propose a three-stage framework that decouples proteomic knowledge acquisition from cross-modal transfer, yielding ProSlide, a slide-level hierarchical pathology foundation model. First, to close the modality gap, we pretrain a Proteomic Foundation Encoder (PFE) on 12,695 sample-level bulk protein profiles using virtual profile generation and expression-space multi-view pretraining. Second, we pretrain ProSlide, a patch-region-slide encoder, to predict protein expression from paired WSI-protein samples. Third, to relax the paired-support bottleneck, we introduce Prot2Path, a cross-modal relational distillation objective. For each paired sample, it aligns the similarity distributions of the WSI and its protein profile over a shared, frozen bank of PFE-encoded paired and standalone profiles. We evaluate ProSlide on 12 downstream tasks across breast, lung, and renal cancers. Despite being pretrained with only 2,229 WSIs and 12,695 sample-level protein profiles, ProSlide achieves the highest mean accuracy and AUC within each cancer group.

World-Action Models (WAMs) offer visual foresight for robotic manipulation, but pixel-space models repeatedly reconstruct entire future scenes, incurring high computational cost and spatio-temporal redundancy. In physical manipulation, consecutive frames often share most of their visual context; the changes between them are what an action policy needs to anticipate. We introduce DeltaWAM, a change-centric WAM that makes a compact delta token the unit of future prediction. Each token is a single vector encoding changes between consecutive dense DINO feature maps. DeltaWAM builds on DeltaWorld, a latent world model pretrained on large-scale videos, to autoregressively predict one delta token per future frame. A flow-matching action expert then conditions on the predicted transitions and current DINO features, which serve as spatial anchors, to generate action chunks. Trained for 256 GPU hours on two H100 GPUs, DeltaWAM has 0.725B parameters and achieves 92.8% average success on LIBERO. It also shows robust generalization under procedural perturbations on LIBERO-Pro. Inference takes 142.1 ms per action chunk with 3.86 GB peak memory.