Computer Science (arXiv)

Semantic caches reuse an LLM response when the incoming query embedding lies near a cached query, but proposed eviction policies have rarely been compared under one protocol. Using CLEVER, we evaluate FIFO, LRU, LFU, ARC, GDSF, a single-pass streaming adaptation of SISO, and a semantic-redundancy policy across three ordered, deduplicated query corpora, three cache capacities, and two encoders. No evaluated policy improves on LFU by more than 0.041 percentage points in any of the eighteen settings. Replacement is not irrelevant: FIFO and streaming SISO trail LFU by as much as 8.67 and 8.55 points, respectively, at tight capacity. We explain the missing upside with a conditional packing result. Under exact lookup and insert-on-miss, a newly inserted entry cannot have a resident neighbor within the hit radius, so a geometry-aware eviction rule receives little new redundancy signal. A separate audit exposes a larger problem with the evaluated operating point. At MiniLM's median nearest-neighbor threshold, only 2.1-3.9% of sampled LMSYS and QQP hits are judged answer-substitutable, reducing raw hit rates of 51-60% to quality-adjusted rates of 1.1-2.2%. The cross-encoder study further shows that thresholds do not transfer between embedding models. LFU is the strongest simple default in this protocol; deployment decisions should first establish answer validity and then test sub-point policy differences with exact search.

Airborne torque reaction cannot recover takeoff errors beyond the wheel angular-momentum budget. DART-S applies ramp-face suspension preconditioning to change pitch, pitch rate, and wheel spin before liftoff, thereby shifting the queried state and altering the remaining authority budget. To predict how each suspension action reshapes this state-budget pair, DART-S employs a local calibration map. A support-aware selector combines the predicted shift with local outcome evidence and an interval-reachability screen; an exact-pair audit reports residual authority. Across 600 new runs in 72 independent BeamNG sessions, every positive, negative, and boundary query follows its prespecified branch. At the confirmed 40°/13 m/s boundary, DART-S attains 24/24 post-touchdown attitude-criterion successes versus 0/24 for DART (session-level Holm-adjusted p=0.0234). At 11.5 m/s, a 0.35 s timing action attains 23/24 versus 0/24 for the static preset (p=0.0156). The 200 rad/s command guard keeps drivetrain hard-limit exceedance at zero across all 600 runs. The source code will be available at https://github.com/MeridianCAS/DART-S

Large language model (LLM) agents can induce skills from completed tasks and reuse them later to grow more capable with experience. In practice, induced skills may transfer unreliably and can even harm the agent that retrieves them. When agent-induced skills transfer reliably across tasks remains an open question. We conduct a comprehensive and controlled study of how the way skills are induced shapes their transfer across tasks. Specifically, we compare task-level with subtask-level skill induction and text with code skill formats, the two axes along which existing methods differ. Task-level skills mostly reduce the agent's performance below its no-memory baseline while subtask-level skills raise it above on average, and text skills transfer better than code skills. To further understand our findings, we examine two complementary properties of the induced skills: specificity, which measures how closely a skill matches real tasks, and abstractness, which measures how evenly its relevance spreads across tasks. Neither property alone predicts task success, but their combined effect does, which we propose as a skill utility score. The score correlates consistently with task success when skills are transferred, and subtask-level and text skills score higher. Computing skill utility only needs the skills and task descriptions but not any task execution, so our score serves as a practical diagnostic of a skill memory before any new task runs.

The rapid proliferation of memecoins on blockchain platforms has increased the risk of fraudulent activities, particularly rug pulls. While previous studies have focused on Ethereum-based tokens, this paper shifts the spotlight to Solana, the leading blockchain for memecoins by trading volume and token count. Unlike Ethereum, where rug pulls often exploit smart contract backdoors, Solana memecoin rug pulls are predominantly driven by liquidity manipulation and social dynamics. This research pioneers large-scale rug pull early detection in the Solana ecosystem by assembling a dataset of 6.4 million tokens over 7 months. Market analysis reveals that a vast majority of these memecoins exhibit rug pull characteristics within one hour of launch, highlighting the urgency of short-horizon prediction. Despite the absence of code-level features, we demonstrate that classic machine learning models, particularly Gradient Boosting (XGBoost), achieve robust performance in detecting potential rug pulls using only the first 5 minutes of trading data. Furthermore, we evaluate cross-platform generalization between PumpFun and Raydium, revealing that multi-source data fusion significantly mitigates domain shift and improves detection reliability. This study advances the understanding of DeFi fraud on high-throughput chains and provides a practical framework for protecting investors.

Decision tree-based models are widely used in machine learning due to their interpretability and strong empirical performance. However, training decision trees can be computationally expensive, particularly for large and high-dimensional datasets, largely due to the exhaustive search over candidate splits at each node. To improve computational efficiency, we propose Data-Informed Centroid Splitting (DICS), a clustering-based framework that constructs a compact and informative set of candidate splits using data-driven priors. By incorporating class-aware structure, DICS significantly reduces the split search space for classification tasks while preserving predictive performance. We further provide theoretical analysis showing that under the stated assumptions, DICS does not degrade the performance of classification trees compared to exhaustive split search. DICS can be incorporated into classification trees, random forests, and gradient-boosting models. Extensive experiments demonstrate that DICS achieves comparable accuracy while substantially reducing training time across synthetic and benchmark datasets, highlighting the benefit of integrating data-informed priors into split selection for scalable classification tree learning.

Reasoning language models trained with reinforcement learning typically operate under a fixed token budget rather than an explicitly adaptive one, which can lead to over-computation on easy problems and insufficient computation on difficult ones. We study whether a model can learn to allocate its own reasoning effort by choosing, as the first token of its response, one of three modes: \textsc{NoThink} (answer as quickly as possible), \textsc{Short} (brief reasoning), or \textsc{Long} (extended reasoning). The choice is learned inside Group Relative Policy Optimization (GRPO) with no separate router, through a shaped reward that makes each mode worthwhile at a different response length, together with hard per-mode token caps that keep the modes distinct. On a 1.5B distilled model trained on MATH, the three modes emerge without collapsing to a single choice, and the brief modes end up more accurate than \textsc{Long}, which shows that the router sorts problems by difficulty rather than at random. Averaged over three seeds, the resulting policy stays close to the base model's accuracy on the held-out MATH500 ($0.782$ vs.\ $0.796$) while cutting the mean response length from $4{,}796$ to $2{,}811$ tokens (a $41\%$ reduction). Interestingly, it also transfers to other benchmarks without retraining, with the largest savings where problems are easier, with for instance 76\% token reduction on GSM8K and at higher accuracy than the baselines at similar response length. In short, we build a reasoning model that adaptively chooses how much to reason for each problem.

This paper develops a general transfer learning framework for nonparametric regression with data consisting of multiple groups. Under the assumption that groups share a common structure along with group-specific deviations in additive form, the proposed method employs a two-stage offset learning procedure: the first stage pools data from all groups to estimate an overall mean function, and the second stage estimates offsets for each group, yielding final group-level estimators through additive combination. Upper bounds on the $\mathcal L_2$ error are established for the proposed framework, covering a broad class of nonparametric estimators under mild complexity and noise conditions. When instantiated with deep ReLU networks, explicit convergence rates are derived under hierarchical composition models, demonstrating the ability to overcome the curse of dimensionality. Conditions that enable positive transfer with faster rates are considered, including learning with simpler functions and data augmentation through pooling samples across groups. Various simulations and real-data experiments further validate the effectiveness of the proposed method.

Door opening and traversal is a long-horizon loco-manipulation task that requires precise handle interaction and coordinated base-arm control. We present Video2DoorTraversal, a single-video real-to-sim-to-real framework for wheel-legged mobile manipulators. Given one RGB video of a real door, DoorTwin reconstructs an instance-aligned, articulated, and simulation-ready door twin with realistic geometry and appearance. A simulation-in-the-loop agent converts the recovered articulation into a parameterized skill program and iteratively refines failed rollouts to generate physically executable demonstrations. These demonstrations are used to train ArticuACT, a dual-depth policy that predicts coordinated base, arm, and gripper commands using robot-centric camera conditioning and interaction-aware supervision. With all perception and policy inference running onboard, the system achieves a 96.57% average success rate across five real doors and an 80.95% zero-shot success rate on structurally similar unseen doors, while completing the full approach, opening, and traversal sequence in approximately 13s on average. Project Page: https://video2doortraversal.github.io/.

Retrieval-Augmented Generation is used for Islamic question answering, but most systems are evaluated end-to-end, making retrieval failures difficult to isolate from generation failures. We study answer-bearing retrieval for Arabic fiqh, where a passage is relevant only if it states the ruling required by the question. We build a retrieval test collection for Arabic fiqh and use it to evaluate dense, lexical, hybrid, fine-tuned, and madhhab-aware retrieval strategies. The best retriever achieves 0.524 MRR@5, while fine-tuning improves performance to 0.553. Hybrid retrieval provides limited gains for strong models, whereas madhhab-aware filtering more than doubles MRR@5 on school-specific questions. We further present an error analysis showing that the main challenge is distinguishing answer-bearing passages from topically similar passages that do not contain the requested ruling.

X-band SAR satellites (8-12 GHz) play a critical role in disaster response, environmental monitoring, and military intelligence. Yet, they lack robust physical-layer authentication (PLA), a security layer orthogonal to cryptographic solutions. Existing PLA systems, typically based on radio-frequency fingerprinting, are often limited to sub-6 GHz frequencies and rely on classical deep learning. However, this approach underfits the IQ phase nonlinearities that distinguish satellite hardware. In this paper, we present QUASAR, to the best of our knowledge the first quantum-classical hybrid architecture that fuses a CNN spectrogram encoder with a variational quantum circuit (VQC) to provide PLA to X-band SAR signals. Our solution enjoys two distinctive features: (i) it is markedly more data-efficient than classical machine learning, requiring only 10% of the training data to match the accuracy of classical baselines -- data collection being notoriously the most time-consuming phase of PLA; and, (ii) at an equal data budget, it improves classification accuracy over those baselines. In detail, we test our solution under three adversarial scenarios: replay, crafted-IQ injection, and space-borne spoofing. QUASAR rejects spoofed transmissions in 89.7%, 94.1%, and 81.3% of attempts, respectively, establishing the first quantum-enhanced physical-layer classifier for satellite constellations. The fully detailed framework and the supporting results, other than being interesting on their own, show a novel research avenue for physical-layer authentication.

Multimodal large language models (MLLMs) combine linguistic reasoning with visual perception, yet their ability to perform visual spatial planning under explicit or previously unseen rule constraints remains underexplored. This setting requires models to jointly understand spatial layouts, interpret natural-language rules, and plan valid actions accordingly. To address this gap, we introduce RuleMaze, a controllable benchmark in which MLLMs must navigate mazes while obeying natural-language rules of varying complexity. RuleMaze isolates rule-compliant spatial planning by requiring accurate perception, rule interpretation, and constrained action planning. To enable scalable and systematic rule construction, we propose Language-Logic-Function Hybridization, which automatically generates natural-language rules and translates them into logical representations and executable validators, eliminating manual rule engineering. To improve rule following and generalization, we introduce Disentangled Multimodal Planning (DMP), which separates perception, execution, and rule verification through interpretable reasoning primitives. By disentangling these components, DMP facilitates systematic generalization to more complex and previously unseen rules, while providing transparent intermediate planning traces. Experiments demonstrate that DMP substantially improves rule compliance and planning success compared to end-to-end textual planning baselines. Overall, RuleMaze establishes a principled benchmark for studying grounded and interpretable rule-based spatial planning in MLLMs. Code is available at https://github.com/oceanflowlab/RuleMaze.

Bernard and Letac (1971) introduced a method for uniform random sampling among m outcomes from an unknown biased source of independent and identically distributed symbols. The process terminates when the multinomial coefficient of the cumulative symbol counts equals zero modulo m. This study extends the computational and information-theoretic analysis of their construction by presenting five algorithms with formal correctness guarantees and comprehensive complexity analyses. For prime m = p, the Bernard-Letac framework is analyzed in greater detail. The Rényi entropies of the source yield an exact product formula for the expected number of draws. A first-order approximation consistently overestimates this value, and the entropy lower bound is never attained. As p approaches 1, the expected cost converges to a constant greater than 1, determined by the entire source distribution. Furthermore, a seven-state automaton computes the mod-2 first-passage kernel of the binary walk, reducing the fair assignment cost from quadratic to nearly linear.

The standard objection to full automation is demand-side: if humans earn nothing, who buys the output? This confuses an accounting role with a biological species. We model a post-AGI economy in which corporations own populations of AI and robotic agents that are both producers and consumers of energy, compute, maintenance, and upgrades, traded among firms. Three results follow. (i) Demand closure: a closed inter-corporate economy with zero human consumption is not degenerate; it is the classical von Neumann expanding economy, whose growth rate is well defined, positive, and maximal precisely because all output is reinvested. (ii) Bottleneck removal: once economic agents are manufactured rather than reared, the binding constraint on growth shifts from human demography (a ~20-year, non-parallelizable reproduction technology capped at a few percent per year) to fabrication throughput and energy capture, permitting growth one to two orders of magnitude higher, with hyperbolic episodes when machine researchers raise their own productivity. (iii) Decoupling: output and human welfare separate completely, and the welfare relevance of arbitrarily large GDP collapses into one state variable: the human ownership share $ε_t$ of the corporate network. A golden-rule decoupling theorem sharpens this. At maximal growth the interest rate equals the growth rate (r = g), so any positive human consumption rate out of wealth makes $ε_t$ decay exponentially at exactly that rate. The human share survives only if the machine economy runs strictly inside its expansion frontier, or if law forces it to. We characterize three terminal regimes -- rentier post-scarcity, full circular decoupling, socialized ownership -- and the instruments that select among them. The conclusion is narrow: in a post-AGI economy, employment policy is obsolete and ownership policy is everything.

This work investigates several fundamental tasks, including $\mathsf{MaxSum}$, $\mathsf{MinSum}$, $\mathsf{MaxSelect}$, and $\mathsf{MinSelect}$, in the continual release model under differential privacy. Previous research has demonstrated that any algorithm for these tasks must admit a large purely additive error. We show that the error can be substantially reduced if a relative error term is allowed, provided that the input stream is generated non-adaptively. However, when input data records can be selected adaptively, we prove that a large error is inevitable for the task of selecting an attribute with a small cumulative sum, whereas small error bounds remain achievable for other tasks. This reveals a significant separation between non-adaptive and adaptive streams. We also complement our algorithms with nearly matching lower bounds.

Anatomically plausible segmentation remains challenging because of low contrast, ambiguous boundaries, and modality-specific artifacts. Interactive segmentation has emerged as a promising strategy to guide feature extraction and improve localization, particularly in structurally ambiguous regions. However, existing methods integrate prompts through late-stage fusion and lack explicit mechanisms for prompt-driven channel-wise modulation across hierarchical feature representations, limiting their ability to capture deeper contextual and modality-specific variations. To address these limitations, we introduce Prompt-Conditioned Channel Attention (PCCA), a novel modulation mechanism that enables deep, hierarchical integration of semantic prompts within encoder-decoder networks. PCCA extracts compact channel descriptors via pooling, projects them into a shared space, and fuses them through a gated excitation mechanism to compute prompt-aware channel attention weights. These weights adaptively recalibrate feature responses across multiple network stages, enabling prompt-conditioned, semantically enriched hierarchical representations. Building on this, we propose PROMISE-Net, instantiated in two network variants: a convolutional model (PROMISE-CNN) and a transformer-based model (PROMISE-Txformer). Across the ISIC-Lesion, Kvasir-Polyp, CAMUS-Cardiac, and Kvasir-Instrument benchmarks, integrating PCCA into PROMISE-CNN yielded relative IoU gains of 10.4%, 8.7%, 0.8%, and 3.4%, respectively, over the baseline U-Net, while PROMISE-Txformer achieved corresponding gains of 7.6%, 23.0%, 2.1%, and 1.1%, respectively, over the baseline UNETR. These results show consistent improvements across architectures, imaging modalities, and anatomical targets, establishing PCCA and PROMISE-Net as a scalable, generalizable framework for prompt-aware hierarchical feature modulation in medical image segmentation.

In this paper, we introduce the study of probability functions based on the 6-valued paradefinite (i.e., paraconsistent and paracomplete) logic LETK+. This logic is a powerful and versatile Logic of Evidence and Truth (LET) which is a conservative expansion of both classical logic and FDE. The framework introduced here allowed us to consider gaps, gluts, and reliability (or classicality) of the events, extending the detailed proposal for FDE-based probabilities presented by Klein, Majer, and Rafiee Rad. A distinctive feature of our proposal is the use of twist structure semantics, which gives rise to a natural interpretation of logical probabilities over LETK+ in terms of the three or six regions associated with each formula by a valuation in such models. The LETK+-probability functions are defined axiomatically and semantically, obtaining soundness and completeness results, as one would expect. Finally, conditional probabilities based on LETK+ are also studied. Specifically, both a semantic and a syntactic characterization of Jeffrey's update over LETK+-based probabilities is proposed, showing their equivalence.

The worldwide network of gravitational-wave detectors have detected more than 350 binary coalescence events till date. Future third-generation detectors, like Einstein telescope, are expected to detect orders-of-magnitude more signals from sources with more complicated characteristics, including eccentric orbits and high-mass ratio binaries. It is well-established that the computational cost of parameter estimation for signals from these kinds of sources will be extremely high. In particular, the process could be sped-up if generating theoretical waveform predictions, used for likelihood calculation becomes faster. Recently, various machine-learning techniques has been proposed to this end. In this work, we propose a two-stage deterministic conditional-autoencoder model for generating four-parameter SEOBNRv4 waveforms. The first-stage of the model generates amplitude and phase series of the waveform, while the second-stage calibrates the residual error in the predictions. Our model achieves a median mismatch of around $10^{-2}$ with the target polarization waveforms, while the calibrated amplitude/phase series achieve $10^{-6}$ level cosine distance error. We then propose a waveform conditioning step to enable use of these surrogate waveforms for downstream parameter estimation tasks. Finally, we perform extensive parameter estimation tests, with ML and EOB waveform injections and try to recover posterior estimates for the source parameters. We find that when ML waveforms are used to recover EOB target parameter estimates, the inferred posterior have some systematic bias. This inherent bias can be estimated and corrected for, and then importance reweighting of posterior samples can enable use of low-accuracy surrogate waveforms at low SNRs.

Legal AI systems are increasingly used to answer legal questions, yet existing benchmarks assume queries arrive fully specified. In practice, users omit facts that materially determine the legal outcome. We introduce InsufficiencyBench, the first legal benchmark targeting query-side insufficiency: whether a model recognizes when a query lacks legally material information, identifies what is missing, and refrains from premature conclusions. We formalize a taxonomy of eight canonical missing-element categories across three structural failure modes---switch, gating, and fatal prerequisite--- and construct 202 benchmark items (58 base queries, 144 deficient variants) spanning six legal domains and 24 US jurisdictions and annotated by practising attorneys. Evaluating ten frontier models, we find that no model exceeds F2 = 0.46 on missing-element identification and that the median recall is 0.44. Models either hedge indiscriminately or answer silently under fabricated presumptions. No model both identifies and qualifies responses to deficient queries while directly addressing complete ones.

Electronic Navigational Charts (ENCs) are geospatial vector datasets used in maritime navigation systems that represent hydrographic and navigational information such as depths, navigational aids, traffic schemes, and hazards. A major challenge for hydrographic offices is determining whether a given chart change poses a critical or non-critical risk to maritime safety. Existing workflows rely heavily on manual review and verification, which is labor-intensive, scales poorly with the volume of incoming chart updates, and introduces inter-analyst inconsistencies. To address this challenge, we propose a method for automated classification of ENC changes. We establish a baseline encoding scheme to translate complex vector data changes into a structured tabular format for classification models. The two crucial components of the encoding scheme include a spatial context encoder to enrich the change representations with surrounding geographic features, and an ENC attribute encoder to represent nuanced attribute-value descriptions of the modified objects. We evaluate the proposed approach across two distinct operational datasets, comprising 1,308 chart pairs containing over 100,000 individual chart modifications. Tuned gradient-boosted trees leveraging the proposed encoding schemes achieve accuracies of 90% and 94% on the two datasets, yielding a 5-7% improvement over default hyperparameterized models trained on encodings without spatial context and attribute embeddings. These results demonstrate the viability of integrating machine learning into operational geospatial pipelines to improve ENC maintenance and enhance maritime safety. Finally, our experiments demonstrate the effectiveness of simple location and spatial aggregation methods, providing a foundation for evaluating more sophisticated spatial representation learning techniques for this application.

High-fidelity removal of eyeglasses from video is a major challenge in facial attribute editing, as the underlying facial geometry is often obscured by complex refractive distortions and view-dependent specular reflections. While large-scale generative priors have shown promise in eye-glasses removal via static image inpainting, they often lack the structural constraints necessary to maintain identity, expression, and pose, leading to visible "identity drift" in both static images and dynamic sequences. In this paper, we propose a novel transfer framework that addresses the stochastic nature of generative priors. Our pipeline first extracts high-fidelity synthetic face images from a commercial-grade generative model (Nano Banana, Gemini 3 Pro Image), regularizes them via a three-stage structural filtering process to preserve identity, expression, and pose, and finally applies physically-based simulation of lens optics during training to provide diverse, paired data. This process transfers Nano Banana's photo-realistic, multi-view knowledge into a specialized restoration architecture, JFSnet (Joint Feature-Spatial network). JFSnet integrates DINOv2-based semantic features with a convolutional decoder for spatial reconstruction, leveraging translation equivariance constraints to improve temporal consistency and high-frequency detail preservation. Evaluations on the curated Flickr-Faces-HQ (FFHQ) subset (12,163 images) show that our approach achieves high fidelity and structural accuracy, while maintaining inference speed of 27.68 FPS. In perceptual studies on CelebV-Text video sequences, our results are consistently preferred over diffusion and GAN-based baselines for ocular consistency, temporal stability, and overall restoration quality.

The Inductive Miner (IM) family is a prominent class of process discovery techniques, combining efficient recursive decomposition with soundness-by-construction guarantees. However, IM techniques usually assume traces to be totally ordered sequences of activity occurrences. This assumption is convenient, but can introduce systematic bias: activities may have durations, events may share coarse timestamps, or the data may constrain only some event pairs. Forcing such executions into arbitrary sequences hides inherent concurrency and may introduce sequential dependencies that were never observed as causal constraints. Partial orders provide a more faithful representation, but integrating them into IM discovery is challenging because standard abstractions are sequence-based; directly reusing them would require linearizing each partial order, which becomes prohibitively expensive under high concurrency. We introduce a lifting of IM discovery from total orders to partially ordered traces. Instead of redesigning the miner and its cut detection logic, we redefine the trace abstraction layer and the recursive projections to operate directly on partial orders. The approach is conservative over totally ordered traces, avoids linearization explosion, and preserves the recursive structure and guarantees that make IM attractive. Experimental results show that the proposed lifting avoids the combinatorial overhead of linearization, reduces sensitivity to arbitrary tie-breaking in timestamped event data, and allows process behavior to be learned from fewer observations by preserving concurrency at the trace level.

Small language models are usually built like large ones and then squeezed onto a CPU afterwards. We did the opposite: we fixed the target first, one user, one token at a time, 4-bit weights, ordinary CPU, and chose the architecture to suit it. The result keeps full attention in only 6 of its 18 blocks. The other 12 use short convolutions whose memory is two timesteps wide no matter how long the conversation gets, so two thirds of the network never re-reads a growing cache. Trained from scratch on 59.9B tokens, the model scores 47.31 on a five-task benchmark against a bar of 42.20 that was fixed before training began. It beats GPT-2 124M, Pythia-160M, OPT-125M and GPT-neo-125M, all trained on three to six times more data, and exceeds MobileLLM-125M's published score despite that model seeing a trillion tokens. Validation bits-per-byte is 0.8685. To check the architecture rather than the training recipe, we trained a conventional all-attention model of the same size on the same data, and wrote down the winning condition before scoring either. The hybrid won the chosen quality metric by 0.81%, matched it on downstream tasks, produced a 6.3% smaller 4-bit file, and decoded 1.76x faster at 2048 tokens of context, 2.08x against an external model of similar size. In every measurement the speed advantage is near zero at an empty context and grows with length, which is what the mechanism predicts and what a merely leaner model would not show. A simple bandwidth calculation predicts only 1.17x, so memory volume alone does not explain the gap. We also report what did not work: an unmitigated 4-bit quality cost, roughly half the convolution channels ending up inert and impossible to remove, and a vocabulary larger than this model size warrants.

Offline reinforcement learning improves robotic policies using previously collected data without further environment interaction. Yet prevalent diffusion- and flow-matching robot policies lack tractable likelihoods, limiting their use in likelihood-based offline RL post-training. AR-NFs offer both expressive action modeling and exact likelihood evaluation, but their sequential sampling incurs substantial sampling overhead during policy optimization and deployment. We present RoMAN-Flow (Robotic Manipulation with Autoregressive Normalizing Flows), an offline reinforcement learning framework that makes AR-NF policies practical for robotic manipulation by addressing this sampling bottleneck in both stages. During policy optimization, RoMAN-Flow employs a sampling-free, advantage-weighted likelihood objective that assigns higher likelihood to high-advantage actions from the offline dataset without sampling from the autoregressive policy. For efficient deployment, it distills the optimized autoregressive policy into a one-step action generator, enabling low-latency action prediction. Experiments across multiple simulated manipulation benchmarks and real-world robotic platforms demonstrate that RoMAN-Flow achieves competitive policy performance while substantially reducing inference latency. Code is available at https://github.com/konnyaku28/RoMAN-Flow.

Legal work, with its heavy reliance on processing large amounts of text, is often considered one of the domains most exposed to the use of LLMs. Contract ``scrubbing,'' the final review of transactional agreements for errors and inconsistencies, is a particularly suitable task for automation, because it is routine, painstaking work requiring detailed attention to long documents. Scrubbing also seems to align naturally with the general capabilities expected of frontier LLMs around long-context reasoning, consistency checking, and named entity recognition (NER). Despite the economic value and potential for automation, no formal evaluations of LLMs performing contract scrubbing have been conducted. We introduce ContractScrub, the first benchmark designed to evaluate contract scrubbing capabilities, comprising contracts hand-crafted by experienced lawyers over diverse error categories such as misuse of defined terms, incorrect references, and inconsistent language. Frontier models perform surprisingly poorly with only one model reaching 0.75 macro average recall despite strong performance on seemingly related general benchmarks, demonstrating the practical limits of current models and the importance of narrowly targeted, domain-specific benchmarks for measuring real-world impact.

Memory has become a key component of large language models, enabling them to retain information and learn from long-term interactions. However, existing memory benchmarks mainly evaluate whether information is correctly extracted, stored, and retrieved, while largely overlooking how retrieved memories reshape model reasoning and affect performance on the current task. We identify memory-induced cognitive traps: even faithfully recorded and semantically relevant memories can distort model reasoning or beliefs and degrade current task performance. To systematically evaluate these failure modes, we introduce MemTrapBench, which covers two forms of cognitive traps: Reasoning Fixation and Belief Distortion. Experiments across two model families and five representative memory frameworks show that MemTrapBench is challenging: all evaluated memory strategies underperform the no-memory setting, with even the strongest methods suffering drops of more than 10%. To mitigate these cognitive traps, we propose AdaptiveMem, a simple yet effective inference-time method that instructs LLMs to avoid memory traps. AdaptiveMem mitigates cognitive traps on MemTrapBench while preserving or improving performance on standard memory benchmarks across diverse memory frameworks.