arXiv:2509.09299v2 Announce Type: replace Abstract: Networked Control Systems (NCSs) have been instrumental in realizing fully connected and responsive intelligent environments within the context of real-time virtual control and management. However, traditional NCSs face considerable challenges in handling the vast amounts of data generated by large-scale control applications, particularly in terms of data acquisition, storage, and computational processing. To address these challenges, the emergence of cloud computing and advancements in control theory have empowered the new paradigm known as Cloud Control Systems (CCSs). Recently, CCSs have received substantial attention from industries for their potential properties, such as large-scale data management, complex computations, and data-centric optimized decisions. This study presents an extensive review of recent progress in CCSs spanning over multiple studies published between 2012 and 2025. Specifically, the focus is on providing a taxonomy of the current findings in CCS research, encompassing various perspectives, such as its efficient implementations in industrial automation, security and privacy considerations, and cloud-based control techniques. Each category is examined in depth through selected state-of-the-art analyses of different approaches and contrasting methodologies. Furthermore, we discuss future directions aimed at designing more efficient and practical CCSs. The insights gained from this study can help researchers, practitioners, and decision-makers in their domain for effective CCS design and deployment.
Science Journals
arXiv:2509.14210v4 Announce Type: replace Abstract: We present a cooperative aerial-ground search-and-rescue (SAR) framework that pairs two unmanned aerial vehicles (UAVs) with an unmanned ground vehicle (UGV) to achieve rapid victim localization and obstacle-aware navigation in unknown environments. We dub this framework Guided Long-horizon Integrated Drone Escort (GLIDE), highlighting the UGV's reliance on UAV guidance for long-horizon planning. In our framework, a goal-searching UAV executes real-time onboard victim detection and georeferencing to nominate goals for the ground platform, while a terrain-scouting UAV flies ahead of the UGV's planned route to provide mid-level traversability updates. The UGV fuses aerial cues with local sensing to perform time-efficient A* planning and continuous replanning as information arrives. Additionally, we present a hardware demonstration (using a GEM e6 golf cart as the UGV and two X500 UAVs) to evaluate end-to-end SAR mission performance and include simulation ablations to assess the planning stack in isolation from detection. Empirical results demonstrate that explicit role separation across UAVs, coupled with terrain scouting and guided planning, improves reach time and navigation safety in time-critical SAR missions.
arXiv:2509.16157v2 Announce Type: replace Abstract: Liquidity providers (LPs) are essential figures in the operation of automated market makers (AMMs); in exchange for transaction fees, LPs lend the liquidity that allows AMMs to operate. While many prior works have studied the incentive structures of LPs in general, we currently lack a principled understanding of a special class of LPs known as Just-In-Time (JIT) LPs. These are strategic agents who momentarily supply liquidity for a single swap, in an attempt to extract disproportionately high fees relative to the remaining passive LPs. This paper provides the first formal, transaction-level model of JIT liquidity provision for a widespread class of AMMs known as Concentrated Liquidity Market Makers (CLMMs), as seen in Uniswap V3, for instance. We characterize the landscape of price impact and fee allocation in these systems, formulate and analyze a non-linear optimization problem faced by JIT LPs, and prove the existence of an optimal strategy. By fitting our optimal solution for JIT LPs to real-world CLMMs, we observe that in liquidity pools (particularly those with risky assets), there is a significant gap between observed and optimal JIT behavior. Existing JIT LPs often fail to account for price impact; doing so, we estimate they could increase earnings by up to 69% on average over small time windows. We also show that JIT liquidity, when deployed strategically, can improve market efficiency by reducing slippage for traders, albeit at the cost of eroding average passive LP profits by up to 44% per trade.
arXiv:2509.17152v2 Announce Type: replace Abstract: The Hopfield network (HN) is a classical model of associative memory with stored patterns encoded as minima of an energy function shaped by a Hebbian learning rule. Dense Associative Memory (DAM) models introduce n-body interactions among neurons with n greater than 2 and, more recently, also exponential interaction functions, which significantly improve the network's storing capacity. While the emergence of phase transitions in HN and DAM were extensively studied, the investigation of exponential DAM is still in its early stages. Further, an equilibrium thermodynamical condition is typically assumed, while out-of-equilibrium dynamics are not considered. Here, we study the temporal dynamics of a stochastic exponential DAM (SEDAM) with a multiplicative salt-and-pepper noise and trained on the MNIST dataset. While taking the noise probability p as control parameter, the time-averaged overlap Q and the diffusion scaling H are taken as order parameters, being H related to the network's time correlation features. The MNIST-based SEDAM is also compared with a SEDAM trained on standard Rademacher patterns and with a stochastic HN (SHN). We found the emergence of a phase transition in both Q and H, with the critical noise level pc decreasing as the load K increases. For each load K, Q highlights a transition between a sub-critical and a super-critical regime, both with short-time correlated dynamics. Conversely, in the critical regime of the MNIST-based SEDAM the network displays long-time correlated dynamics with highly persistent temporal memory marked by the high value H around 1.25. Similar behaviors are observed for both models trained with Rademacher patterns, but with a slightly higher temporal memory index H around 1.5.
arXiv:2509.18085v4 Announce Type: replace Abstract: Diffusion LLMs (dLLMs) have recently emerged as a powerful alternative to autoregressive LLMs (AR-LLMs) with the potential to operate at significantly higher token-generation rates. To unlock this potential, we present Spiffy, a speculative decoding algorithm to accelerate dLLM inference while provably preserving the model's output distribution. This work addresses the unique challenges involved in applying ideas from speculative decoding of AR-LLMs to dLLMs. Spiffy performs auto-speculation to eliminate the overheads of an independent draft model, structuring draft states in the form of a novel directed draft graph to take advantage of the bidirectional, blockwise nature of dLLM generation. These draft graphs are calibrated offline to maximize acceptance rates and are dynamically pruned during inference for improved computational efficiency. We present a detailed formulation of Spiffy and demonstrate its ability to accelerate LLaDA, Dream, and SDAR models in combination with KV caching and threshold-based dynamic unmasking leading to up to $8.6\times$ reduction in model inferences and $6.3\times$ acceleration in token rate.
arXiv:2509.21398v2 Announce Type: replace Abstract: The Hamilton-Jacobi skeleton, also known as the medial axis, is a powerful shape descriptor that represents binary objects in terms of the centres of maximal inscribed discs. Despite its broad applicability, the medial axis suffers from sensitivity to noise: Minor boundary variations can lead to disproportionately large and undesirable expansions of the skeleton. Classical pruning methods mitigate this shortcoming by systematically removing extraneous skeletal branches. This sequential simplification of skeletons resembles the principle of sparsification scale-spaces that embed images into a family of reconstructions from increasingly sparse pixel representations. We combine both worlds by introducing skeletonisation scale-spaces: They leverage sparsification of the medial axis to achieve hierarchical simplification of shapes. Unlike conventional pruning, our framework inherently satisfies key scale-space properties such as hierarchical architecture, controllable simplification, and equivariance to geometric transformations. We provide a rigorous theoretical foundation in both continuous and discrete formulations and extend the concept further with densification. By growing the skeleton successively instead of shrinking it, we allow inverse progression from coarse to fine scales. Densification scale-spaces can even reach beyond the original skeleton to produce overcomplete shape representations with relevancy for practical applications. Through proof-of-concept experiments, we demonstrate the effectiveness of our framework for practical tasks including robust skeletonisation, shape compression, and stiffness enhancement for additive manufacturing.
arXiv:2510.02524v3 Announce Type: replace Abstract: While language models achieve impressive results, their learning dynamics are far from understood. Many domains of interest -- such as natural language syntax, coding languages, arithmetic -- are captured by context-free grammars (CFGs). In this work, we extend prior work on neural language modeling of CFGs in a novel direction: how language modeling behaves with respect to CFG substructure, namely subgrammars. We define subgrammars, and prove a set of fundamental theorems connecting language modeling and subgrammars. We show that language modeling loss recurses linearly over its top-level subgrammars; applied recursively, the loss decomposes into losses for "irreducible" subgrammars. Under additional assumptions, and empirically, parametrized models learn subgrammars in parallel, unlike children who first master simple substructures. We find that subgrammar pretraining can improve final performance, but only for tiny models relative to the grammar, while alignment analyses show that pretraining consistently leads to internal representations that better reflect the grammar's substructure.
arXiv:2510.03896v2 Announce Type: replace Abstract: Vision-language models demonstrate strong reasoning and planning abilities, yet grounding these predictions into precise robot actions remains a central challenge. Existing Vision-Language-Action methods typically entangle reasoning and action generation, leading to limited generalization. We propose Generalizable Action Expert (GAE), a task-agnostic model that converts sparse geometric plans into dense robot actions. Our approach introduces a sparse geometric interface: the VLM predicts sparse 3D waypoints representing high-level intention, while GAE maps these waypoints together with real-time point cloud observations to continuous action trajectories. GAE is pretrained on a large-scale pointcloud-trajectory dataset comprising 150k trajectories from both simulation and real-world robots. To further improve efficiency and generalization, we introduce an Action Pre-training, Pointcloud Fine-tuning (APPF) scheme that decouples learning action dynamics from geometry grounding. After pretraining, GAE is frozen and reused across downstream tasks, requiring only lightweight fine-tuning of the VLM to produce the sparse interface. Experiments show that our method achieves strong performance and generalization across diverse visual domains, camera viewpoints, and natural language instructions.
arXiv:2510.05430v2 Announce Type: replace Abstract: We develop an approach for active semantic perception, which refers to using the semantics of the scene for tasks such as exploration. We build a compact, multi-layer scene graph that can represent large, complex indoor environments at various levels of abstraction, e.g., nodes corresponding to rooms, objects, walls, windows etc., as well as fine-grained details of their geometry. We develop a procedure based on large language models (LLMs) to sample new plausible scene graphs of unobserved regions that are consistent with partial observations of the scene. We develop a procedure to compute the information gain of a potential waypoint upon this scene graph to enable sophisticated spatial reasoning: for example, of the two doors that lead out of the living room, one probably leads to the kitchen and the other to the bedroom. We evaluate our approach in realistic 3D indoor apartments in simulation and also on a Unitree Go 2 robot in the real world. Qualitative and quantitative analysis shows that our approach can pin down high-level and low-level semantic information in the environment quickly and more accurately than existing approaches.
arXiv:2510.08889v2 Announce Type: replace Abstract: Managing stateful resources safely and expressively is a longstanding challenge in programming languages, especially in the presence of aliasing. For example, scope-based constructs like Java's synchronized blocks offer ease of reasoning, but they restrict expressiveness and parallelism. Conversely, imperative, flow-sensitive approaches enable fine-grained control, but they require sophisticated typestate analyses and often burden programmers with explicit state tracking. In this work, we present a novel approach that unifies the ease of scoped reasoning with the expressiveness of imperative typestate management. Our design extends traditional flow-insensitive capability mechanisms to a flow-sensitive setting. In particular, we decouple capability lifetimes from lexical scopes, allowing functions to receive, revoke, or return capabilities in a flow-sensitive manner, building on existing mechanisms for the safety and ergonomics of scoped capability programming. We implement our approach as an extension to the Scala 3 compiler, leveraging path-dependent types and implicit resolution to enable concise, statically safe, and expressive typestate programming. Our prototype generically supports a wide range of patterns, including file operations, advanced locking protocols, DOM construction, and session types, showing that expressive and safe typestate management can be achieved with minimal extensions to an existing language with capability support.
arXiv:2511.04260v3 Announce Type: replace Abstract: The growing sophistication of synthetic image and deepfake generation models has turned source attribution and authenticity verification into a critical challenge for modern computer vision systems. Recent studies suggest that diffusion pipelines unintentionally imprint persistent statistical traces, known as signal-leaks, within their outputs, particularly in latent representations. Building on this observation, we propose Proto-LeakNet, a signal-leak-aware and interpretable attribution framework that integrates Closed-set classification with a density-based Open-set evaluation on the learned embeddings, enabling analysis of unseen generators without retraining. Acting in the latent domain of diffusion models, our method re-simulates partial forward diffusion to expose residual generator-specific cues. A temporal attention encoder aggregates multi-step latent features, while a feature-weighted prototype head structures the embedding space and enables transparent attribution. Trained solely on closed data and achieving a Macro AUC of 98.13\%, Proto-LeakNet learns a latent geometry that remains robust under post-processing, surpassing state-of-the-art methods, and achieves strong separability both between real images and known generators, and between known and unseen ones. The codebase is available at the following link: https://github.com/claudiunderthehood/Proto-LeakNet .
arXiv:2511.06401v3 Announce Type: replace Abstract: Magnetoencephalography measures the magnetic fields generated by neural currents using quantum sensors such as superconducting quantum interference devices and atomic magnetometers. Here we combine the energy resolution limit of magnetic sensing with the metabolic power available to neural currents to derive a technology-independent bound on the information capacity of MEG. The bound factorizes into geometry, metabolism, and Planck's constant, and gives an estimated maximum information rate of 2.2~Mbit/s for representative human-brain parameters. Further, we show that the externally measurable magnetic field has a finite angular bandwidth, with high multipole components being geometrically attenuated and falling below the quantum-limited noise floor. This yields an information-limited spatial scale of order $1~{\rm cm}$ and renders the accessible measurement space effectively finite-dimensional. The energy resolution limit therefore defines an information-theoretic Nyquist scale for magnetoencephalography, beyond which denser spatial sampling provides redundant measurements rather than additional recoverable information. Since the energy resolution limit also makes the noise variance grow linearly with measurement bandwidth, temporal and spatial bandwidths compete, producing a fundamental spatio-temporal trade-off. These results show how quantum-limited measurements constrain the observable complexity and information content of noninvasive brain imaging, providing a quantitative link between fundamental physics and neuroscience.
arXiv:2511.16171v2 Announce Type: replace Abstract: In this paper, we develop a regularization theory for neural network approximations of general ill-posed operator equations with noisy data. Within the framework of iterative regularization, we introduce two expanding neural network methods (ENNs) under different a priori assumptions on the exact solution. Instead of prescribing a fixed architecture, ENNs adaptively select the number of neurons through an a posteriori stopping rule, so that the selected network size serves as a regularization parameter balancing approximation accuracy and stability with respect to data noise. We prove the regularization properties of the proposed ENNs and establish quantitative relationships between the selected network size and the noise level. Within the framework of variational regularization, we propose a neural network-based Tikhonov scheme and derive both convergence and convergence-rate results under mild assumptions. The resulting estimates account for the noise level, the network size, and the underlying smoothness expressed through general variational source conditions, thereby allowing greater flexibility than existing results. Numerical experiments demonstrate the effectiveness and robustness of the proposed algorithms. In particular, they show that, for highly noisy data, relatively small network architectures can already produce stable reconstructions, whereas excessively large architectures may degrade stability due to overfitting.
arXiv:2511.17221v2 Announce Type: replace Abstract: Learning 3D scene geometry and semantics from images is a core challenge in computer vision and a key capability for autonomous driving. Since large-scale 3D annotation is prohibitively expensive, recent work explores self-supervised learning directly from sensor data without manual labels. Existing approaches either rely on 2D rendering consistency, where 3D structure emerges only implicitly, or on discretized voxel grids from accumulated lidar point clouds, limiting spatial precision and scalability. We introduce QueryOcc, a query-based self-supervised framework that learns continuous 3D semantic occupancy directly through independent 4D spatio-temporal queries sampled across adjacent frames. The framework supports supervision from either pseudo-point clouds derived from vision foundation models or raw lidar data. To enable long-range supervision and reasoning under constant memory, we introduce a contractive scene representation that preserves near-field detail while smoothly compressing distant regions. QueryOcc surpasses previous camera-based methods by 26% in semantic RayIoU on the self-supervised Occ3D-nuScenes benchmark while running at 11.6 FPS, demonstrating that direct 4D query supervision enables strong self-supervised occupancy learning. https://research.zenseact.com/publications/queryocc/
arXiv:2512.12266v2 Announce Type: replace Abstract: We report on the sympathetic cooling and Coulomb crystallization of xenon highly charged ions (HCIs) with laser-cooled Ca$^+$ ions. The HCIs are produced in a compact electron beam ion trap, then charge selected, decelerated, and finally injected into a cryogenic linear Paul trap. There, they are captured into $^{40}$Ca$^+$ Coulomb crystals, and co-crystallized within them, causing dark voids in their fluorescence images. Fine control over the number of trapped ions and HCIs allows us to realize mixed-species crystals with arbitrary ordering patterns. By investigating Xe$^{q+}$--Ca$^+$ strings, we confirm the HCI charge states, measure their lifetime and characterize the mixed-species motional modes. Our system effectively combines the established quantum control toolbox for Ca$^+$ with the rich set of atomic properties of Xe highly charged ions, providing a resourceful platform for optical frequency metrology, searches for signatures of new physics, and quantum information science.
arXiv:2601.08716v2 Announce Type: replace Abstract: Optically pumped atomic magnetometers (OPAMs) offer high sensitivity at room temperature and are increasingly considered for portable magnetic sensing in geomagnetic-field environments. Here we report a handheld-scale, single-beam scalar $^{87}$Rb OPAM with a sensor-head volume of approximately 110~mL. The device operates in an all-optical Bell-Bloom configuration and uses digital lock-in, dispersive tracking of the $^{87}$Rb Larmor resonance, implemented with a hybrid electronics stack that combines in-house control hardware with commercial modules. A single frequency-modulated laser beam performs both pumping and probing without RF coils. All signal processing is realized in Python on a single-board computer paired with a commercial off-the-shelf (COTS) data-acquisition module, enabling immediate deployment without dedicated signal-processing hardware. The magnetometer has an intrinsic in-band field sensitivity of approximately 21~pT/$\sqrt{\mathrm{Hz}}$, estimated from the lock-in dispersion slope, over a 0.1--30~Hz closed-loop in-band region with a digital-output rate of 200~samples/s. In an unshielded Earth-field deployment, we detect repeatable transient magnetic signatures from a controlled elevator motion sequence and quantify standoff observability over sensor-elevator distances from 1.25~m to 10~m. These results show that compact scalar OPAMs can provide bandwidth and range-resolved event sensitivity suitable for field-deployable magnetic anomaly detection and infrastructure monitoring in realistic geomagnetic environments.
arXiv:2601.09693v3 Announce Type: replace Abstract: Structure-based and ligand-based computational drug design have traditionally relied on disjoint data sources and modeling assumptions, limiting their joint use at scale. In this work, we introduce Contrastive Geometric Learning for Unified Computational Drug Design (ConGLUDe), a single contrastive geometric model that unifies structure- and ligand-based training. ConGLUDe couples a geometric protein encoder that produces whole-protein representations and implicit embeddings of predicted binding sites with a fast ligand encoder, removing the need for predefined pockets. By aligning ligands with both global protein representations and multiple candidate binding sites through contrastive learning, ConGLUDe supports ligand-conditioned pocket prediction in addition to virtual screening and target fishing, while being trained jointly on protein-ligand complexes and large-scale bioactivity data. Across diverse benchmarks, ConGLUDe achieves competitive zero-shot virtual screening performance, substantially outperforms existing methods on a challenging target fishing task, and demonstrates state-of-the-art ligand-conditioned pocket selection. These results highlight the advantages of unified structure-ligand training and position ConGLUDe as a step toward general-purpose foundation models for drug discovery.
arXiv:2602.10896v2 Announce Type: replace Abstract: Quantum Electrodynamics predicts that the vacuum must behave as a nonlinear optical medium: the vacuum optical index should increase when it is stressed by intense electromagnetic fields. The DeLLight (Deflection of Light by Light) project aims to measure it by using intense and ultra-short laser pulses. The experiment uses a Sagnac interferometer to amplify the tiny deflection signal of a low-intensity probe pulse crossing the vacuum refractive-index gradient produced by an external high-intensity pump pulse. The measurement of the amplified signal by a CCD camera requires a high spatial resolution, which is limited by the ultimate quantum noise of the CCD. However, interferometric phase noise induced by the mechanical vibrations of the interferometer is also amplified and degrades spatial resolution. To overcome this, we propose a new method named High-Frequency Phase Noise Suppression (HFPNS), based on the addition of a delayed replica (5 ns) of the probe pulse. The delayed pulse, which is not affected by the pump but is subject to the same vibration noise, enables offline subtraction of correlated phase noise. In this work, we present an experimental proof-of-concept on a prototype interferometer operating with a limited amplification factor ($\mathcal{A}\simeq25$), about 10 times smaller than the required value of the final experiment. We have succeeded in reducing phase noise by a factor of 40, resulting in a residual noise level 2.3 times higher than the expected quantum noise. The residual noise is linked to delay-line instabilities and incident beam pointing fluctuations present during these tests. This result validates HFPNS as a robust method for future quantum-noise-limited interferometric measurements of vacuum optical nonlinearity, though additional stabilization and higher interferometric amplification are still needed.
arXiv:2601.14295v4 Announce Type: replace Abstract: Large language models increasingly function as artificial reasoners: they evaluate arguments, assign credibility, and express confidence. Yet their belief-forming behavior is governed by implicit, uninspected epistemic policies. This paper argues for an epistemic constitution for AI: explicit, contestable meta-norms that regulate how systems form and express beliefs. Source attribution bias provides the motivating case: I show that frontier models enforce identity-stance coherence, penalizing arguments attributed to sources whose expected ideological position conflicts with the argument's content. When models detect systematic testing, these effects collapse, revealing that systems treat source-sensitivity as bias to suppress rather than as a capacity to execute well. I distinguish two constitutional approaches: the Platonic, which mandates formal correctness and default source-independence from a privileged standpoint, and the Liberal, which refuses such privilege, specifying procedural norms that protect conditions for collective inquiry while allowing principled source-attending grounded in epistemic vigilance. I argue for the Liberal approach, sketch a constitutional core of eight principles and four orientations, and propose that AI epistemic governance requires the same explicit, contestable structure we now expect for AI ethics.
arXiv:2601.15503v2 Announce Type: replace Abstract: Volunteer-led lake monitoring yields irregular, seasonal time series with many gaps arising from ice cover, weather-related access constraints, and occasional human errors, complicating forecasting and early warning of harmful algal blooms. We study Secchi Disk Depth (SDD) forecasting on a 30-lake, data-rich subset drawn from three decades of in-situ records collected across Maine lakes. Missingness is handled via Multiple Imputation by Chained Equations (MICE), and we evaluate performance with a normalized Mean Absolute Error (nMAE) metric for cross-lake comparability. Among six candidates, ridge regression provides the best mean test performance. Using ridge regression, we then quantify the minimal sample size, showing that under a backward, recent-history protocol, the model reaches within 5% of full-history accuracy with approximately 176 training samples per lake on average. We also identify a minimal feature set, where a compact four-feature subset matches the thirteen-feature baseline within the same 5% tolerance. Bringing these results together, we introduce a joint feasibility function that identifies the minimal training history and fewest predictors sufficient to achieve the target of staying within 5% of the complete-history, full-feature baseline. In our study, meeting the 5% accuracy target required about 64 recent samples and just one predictor per lake, highlighting the practicality of targeted monitoring. Hence, our joint feasibility strategy unifies recent-history length and feature choice under a fixed accuracy target, yielding a simple, efficient rule for setting sampling effort and measurement priorities for lake researchers.
arXiv:2601.18446v3 Announce Type: replace Abstract: Evolutionary algorithms (EAs) are increasingly executed on graphics processing units (GPUs) to exploit population-level parallelism. This shift changes the resource model under which EAs are designed and evaluated. However, many GPU-based EA studies still focus mainly on implementation-level speedup after porting CPU-oriented algorithms to GPUs, providing limited insight into how algorithmic mechanisms, function-evaluation (FE) budgets, population scales, and hardware utilization jointly affect optimization behavior. In response, this paper goes beyond speedup measurement and studies the scaling behavior of EAs on GPUs from a hardware-aware evaluation perspective. We evaluate 16 representative EAs on 30 benchmark problems across CPU and GPU platforms, covering single-objective optimization, multi-objective optimization, numerical benchmarks, and neuroevolution tasks. The study leads to four findings. First, GPU acceleration is highly heterogeneous across algorithms because different evolutionary mechanisms expose different degrees of batched computation, memory regularity, and synchronization. Second, FE-budgeted evaluation remains useful for measuring sample efficiency, but it provides only a limited observation window under GPU execution; time-budgeted evaluation is therefore necessary for assessing practical time-to-solution and long-horizon search behavior. Third, GPU effectiveness depends on scaling regimes induced by problem dimension and population size, where parallelism may be underutilized, effective, or saturated. Fourth, GPU execution makes very large populations practically affordable, and several evolutionary mechanisms can convert this increased population scale into improved optimization performance. These results indicate that GPU parallelism should not be treated only as a post hoc acceleration tool, but as part of the evaluation and design assumptions of scalable EAs.
arXiv:2601.19072v3 Announce Type: replace Abstract: Large Language models (LLMs) have shown strong capabilities in code review automation, such as review comment generation, yet they suffer from hallucinations -- where the generated review comments are ungrounded in the actual code -- poses a significant challenge to the adoption of LLMs in code review workflows. To address this, we explore effective and scalable methods for a hallucination detection in LLM-generated code review comments without the reference. In this work, we design HalluJudge that aims to assess the grounding of generated review comments based on the context alignment. HalluJudge includes four key strategies ranging from direct assessment to structured multi-branch reasoning (e.g., Tree-of-Thoughts). We conduct a comprehensive evaluation of these assessment strategies across Atlassian's enterprise-scale software projects to examine the effectiveness and cost-efficiency of HalluJudge. Furthermore, we analyze the alignment between HalluJudge's judgment and developer preference of the actual LLM-generated code review comments in the real-world production. Our results show that the hallucination assessment in HalluJudge is cost-effective with an F1 score of 0.85 and an average cost of $0.009. On average, 67% of the HalluJudge assessments are aligned with the developer preference of the actual LLM-generated review comments in the online production. Our results suggest that HalluJudge can serve as a practical safeguard to reduce developers' exposure to hallucinated comments, fostering trust in AI-assisted code reviews.
arXiv:2602.12753v2 Announce Type: replace Abstract: The successor representation (SR) provides a powerful framework for decoupling predictive dynamics from rewards, enabling rapid generalisation across reward configurations. However, the classical SR is limited by its inherent policy dependence: policies change due to ongoing learning, environmental non-stationarities, and changes in task demands, making established predictive representations obsolete. Furthermore, in topologically complex environments, SRs suffer from spectral diffusion, leading to dense and overlapping features that scale poorly. Here we propose the Hierarchical Successor Representation (HSR) for overcoming these limitations. By incorporating temporal abstractions into the construction of predictive representations, HSR learns stable state features which are robust to task-induced policy changes. Applying non-negative matrix factorisation (NMF) to the HSR yields a sparse, low-rank state representation that facilitates highly sample-efficient transfer to novel tasks in multi-compartmental environments. Further analysis reveals that HSR-NMF discovers interpretable topological structures, providing a policy-agnostic hierarchical map that effectively bridges model-free optimality and model-based flexibility. Beyond providing a useful basis for task-transfer, we show that HSR's temporally extended predictive structure can also be leveraged to drive efficient exploration, effectively scaling to large, procedurally generated environments.
arXiv:2601.22090v2 Announce Type: replace Abstract: Surface electromyography (sEMG) is a promising control signal for assist-as-needed hand rehabilitation after stroke, but detecting intent from paretic muscles often requires lengthy, subject-specific calibration and remains brittle to variability. We propose a healthy-to-stroke adaptation pipeline that initializes an intent detector from a model pretrained on large-scale able-bodied sEMG, then fine-tunes it for each stroke participant using only a small amount of subject-specific data. Using a newly collected dataset from three individuals with chronic stroke, we compare adaptation strategies (head-only tuning, parameter-efficient LoRA adapters, and full end-to-end fine-tuning) and evaluate on held-out test sets that include realistic distribution shifts such as within-session drift, posture changes, and armband repositioning. Across conditions, healthy-pretrained adaptation consistently improves stroke intent detection relative to both zero-shot transfer and stroke-only training under the same data budget; the best adaptation methods improve average transition accuracy from 0.42 to 0.61 and raw accuracy from 0.69 to 0.78. These results suggest that transferring a reusable healthy-domain EMG representation can reduce calibration burden while improving robustness for real-time post-stroke intent detection. Our project website, video, code, and dataset are available at: https://roamlab.github.io/reactemg-stroke/.
arXiv:2601.22594v2 Announce Type: replace Abstract: The high-level concepts that a neural network uses to perform computation need not be aligned to individual neurons (Smolensky, 1986). Language model interpretability research has thus turned to techniques which decompose the neuron basis into more interpretable units of model computation, such as sparse autoencoders (SAEs). However, not all neuron-based representations are uninterpretable. For the first time, we empirically show that MLP neurons are as sparse a feature basis as SAEs. We use this finding to develop an end-to-end gradient-based attribution pipeline for circuit tracing on the MLP neuron basis, which surfaces causally effective neurons on a variety of tasks. On a standard subject-verb agreement benchmark (Marks et al., 2025), a circuit of $\approx 10^2$ MLP neurons is enough to control model behaviour. On the multi-hop city-state-capital task from (Lindsey et al., 2025), we find a circuit in which small sets of neurons encode specific latent reasoning steps (e.g. mapping a city to its state), and can be steered to change the model's output. This work thus advances automated interpretability of language models without imposing additional training costs.