Protein homology search underlies function annotation, structure prediction, and evolutionary analysis, but remains challenging in the "twilight zone," where global sequence similarity is weak and classical alignment methods lose sensitivity. Protein language models provide context-aware representations that could improve alignment sensitivity in this regime. However, prior protein embedding-based retrieval pipelines often pool these representations into a single vector, potentially obscuring local motifs, domains, or conserved residues that reveal remote homology. We introduce ProtoCol, a model which represents proteins as sets of residue embeddings and uses ColBERT-style late interaction to test whether residue-level comparison improves homolog retrieval. ProtoCol encodes proteins independently, keeps candidate representations pre-computable, and scores candidates with MaxSim over residue embeddings. On SCOPe superfamily and Pfam clan benchmarks, ProtoCol outperforms sequence-composition, alignment-based, pooled PLM, and trained single-vector baselines, supporting late interaction as an effective retrieval layer for remote homology search.

Large Language Models (LLMs) have become the central paradigm in artificial intelligence, yet the core computational primitive of attention has remained structurally unchanged. Local Linear Attention (LLA) is an attention mechanism derived from nonparametric statistics in the test-time regression framework. In contrast to prior research on efficient attention variants, LLA upgrades the local constant estimate in softmax attention to a local linear estimate, yielding provably superior bias-variance tradeoffs for associative memory. However, LLA has not been scaled in LLM pretraining due to computational and numerical stability concerns. We introduce Parallax, a parameterized Local Linear Attention that is scalable for LLMs. Parallax eliminates the numerical solver in LLA and learns an extra query-like projector that probes the KV covariance. We place Parallax within a family of attention mechanisms connected by the bandwidth, the probe construction and the affine structure. We propose a hardware-aware algorithm that increases the arithmetic intensity over FlashAttention, shifting attention into a more compute bound regime. Our prototype decode kernel matches or outperforms FlashAttention 2/3 across diverse batch sizes and context lengths. We pretrain Parallax at 0.6B and 1.7B scales and find consistent perplexity improvements throughout pretraining with gains that transfer to downstream benchmarks. The advantage persists under both parameter-matched and compute-matched controls, demonstrating a Pareto improvement. We perform careful pretraining ablations and identify a novel phenomenon whereby Muon unlocks the capacity of Parallax. To our knowledge, this is the first empirical demonstration of strong architecture-optimizer codesign for attention mechanisms in the architecture research literature.

Pointwise reward modeling offers critical signals for LLM post-training, yet struggles with absolute scoring in subjective, non-verifiable settings. Rubric-based methods address this by decomposing evaluation into explicit criteria, but existing approaches typically depend on frontier LLMs and suffer from ties caused by hard Boolean aggregation. We present RUBRIC-ARROW, an alternating framework that jointly trains a rubric generator and a rubric-conditioned judge, with its RL stage using only pairwise preference data. Our method couples a probability-based scoring rule that reduces ties with phase-specific preference-based rewards and an alternating GRPO scheme that together train the pointwise evaluator. Extensive experiments show that RUBRIC-ARROW achieves competitive reward-modeling accuracy and yields consistent gains for downstream policy post-training.

In the literature, actor-critic model predictive control (AC-MPC) integrates MPC with reinforcement learning to enable high-performance control of complex dynamical systems. However, its differentiable MPC layer requires repeatedly solving an optimization problem in both the forward and backward passes, leading to substantial training and inference latency. This paper tackles this bottleneck introducing a CUDA-accelerated variant that significantly reduces end-to-end execution time while preserving the control performance of the baseline formulation. Simulation results on an agile drone racing task show that our approach achieves state-of-the-art lap times and near-limit dynamic behaviour with markedly reduced training and inference time.

Neural networks trained under different hyperparameter settings can fall into distinct training "regimes," with consistent behavior within regimes and qualitative differences across regimes. In this paper, we study such multi-regime behavior in scientific machine learning (SciML) models through a regime-aware diagnostic framework that jointly analyzes performance, training dynamics, and loss-landscape geometry. We identify three key findings: (i) a consistent three-regime structure emerges across many standard SciML models, different constraint enforcements, and various optimizer designs; (ii) optimization effectiveness is regime-specific, with no single method performing well across all regimes; and (iii) SciML models can exhibit fine-grained failure modes that can challenge conventional interpretations of standard loss-landscape metrics. Our results provide an approach to establish a unified, task-oblivious perspective on failure modes in SciML and to inform regime-aware guidance for improving robustness. We validate these findings across widely-used SciML models, including physics-informed neural networks, neural operators, and neural ordinary differential equations, on benchmarks spanning representative ordinary and partial differential equations.

Randomly initialized neural networks induce a prior over functions, but the predictor used in practice is produced only after training. We ask how much of this initial bias survives the training pipeline. To make the question measurable, we introduce initialization memory: the dependence of the validation-selected predictor on the scale of the random initialization. We perform controlled CIFAR-10 experiments on ResNets where initialization memory already sharply separates training regimes. Low-learning-rate SGD can interpolate while still remembering its initialization: on ResNet-9 with batch size $b=128$, test accuracy varies by $26.5$ percentage points across initialization scales despite $\ge99.5\%$ training accuracy. This is not undertraining: extending the same low-learning-rate regime to $5{,}000$ epochs leaves the spread essentially unchanged. In contrast, Adam-family methods largely erase the dependence. SGD can also be made to forget when larger learning rates are paired with explicit $L_2$ norm control. We interpret these findings in terms of the time scale of forgetting: gradient-flow-like dynamics can preserve initialization memory, whereas stochastic finite-step effects, explicit norm decay, and adaptive preconditioning erase it on scales governed by the size of explicit or implicit regularization. The practical inductive bias of a trained network is therefore not the architectural prior alone, but the architectural prior after being filtered by the forgetting dynamics of the training pipeline; and the same regularizers that improve generalization are precisely those that erase memory of initialization.

We study stochastic decision-theoretic online learning with full information and event-level pure differential privacy. A COLT open problem of Hu and Mehta asks to determine the optimal gap-dependent regret rate for stochastic decision-theoretic online learning under pure event-level differential privacy. For $K$ actions, losses in $[0,1]$, and a unique best action separated from the second-best action by gap $Δ_{\min}$, the known lower bound is of order $ \frac{\log K}{\min\{Δ_{\min},\varepsilon\}}, $ or equivalently, up to universal constants, of order \[ \frac{\log K}{Δ_{\min}}+\frac{\log K}{\varepsilon}. \] We give a horizon-free pure-DP algorithm and prove the explicit regret bound \[ \operatorname{Reg}_T \le 1000 \cdot \left(\frac{\log K}{Δ_{\min}}+\frac{\log K}{\varepsilon}\right) \] for every horizon $T$. The numerical constant is not optimized. The algorithm partitions time into blocks of exponentially increasing size, plays a single action throughout each block, and chooses the next action by an exponential mechanism applied to a data-independent random prefix of the previous block. The random prefix converts block regret into a sum, over all prefix lengths, of softmax selection errors. A single entropy-potential argument controls all privacy-dominated large-gap actions at cost $\log K/\varepsilon$.

Robotics Wire Arc Additive Manufacturing (WAAM) is governed by complex and nonlinear process dynamics coupling thermal field to the build geometry. The process may be regarded as a multi-input/multi-output dynamical system with welding torch speed and wire feed rate as inputs and weld bead deposition height and width as outputs. In this paper, we use the input/output data to learn a data-driven model and use it for weld planning and control. We show that a simple recurrent neural network architecture and one-step-ahead predictive control can improve the process performance in terms of height and width consistency. To account for the changing thermal conditions during the printing process, we update the learning model using prediction error from the previous layer. This adaptation step further improves the prediction accuracy and controller performance. Experiments on a robotic WAAM testbed with integrated line-scanner feedback significant improvements in height and width consistency compared to constant input and static model baselines. The proposed learning and adaptation framework provides a practical pathway toward robust, data-driven regulation of additive manufacturing processes.

Latency-accuracy tradeoffs are fundamental in real-time applications of deep neural networks (DNNs) for cyber-physical systems. In autonomous driving, in particular, safety depends on both prediction quality and the end-to-end delay from sensing to actuation. We observe that (1) when latency is accounted for, the latency-optimal network configuration varies with scene context and compute availability; and (2) a single fixed-resolution model becomes suboptimal as conditions change. We present a multi-resolution, end-to-end deep neural network for the CARLA urban driving challenge using monocular camera input. Our approach employs a convolutional neural network (CNN) that supports multiple input resolutions through per-resolution batch normalization, enabling runtime selection of an ideal input scale under a latency budget, as well as resolution retargeting, which allows multi-resolution training without access to the original training dataset. We implement and evaluate our multi-resolution end-to-end CNN in CARLA to explore the latency-safety frontier. Results show consistent improvements in per-route safety metrics - lane invasions, red-light infractions, and collisions - relative to fixed-resolution baselines.

We introduce a probabilistic splat-based radiance field framework that retains the fast rasterization and test-time efficiency of 3D Gaussian Splatting (3DGS) while replacing heuristic primitive manipulation with gradient-based optimization of a volumetric probability density. Rather than relocating, splitting, or culling Gaussians via hand-tuned densification (e.g., ADC), we treat primitive locations as samples drawn from a persistent, learnable density. We instantiate this density using a novel, memory-efficient multi-scale hierarchical grid that enables end-to-end gradient-based optimization. To stabilize the optimization, we derive an unbiased gradient estimator with control variates that markedly reduces variance. By allowing probability mass to flow to where the loss demands, our framework eliminates brittle priors and naturally explores the volume, achieving state-of-the-art reconstruction quality on mip-NeRF 360 while preserving 3DGS-level rendering speed.

Agentic AI systems are increasingly being explored as production infrastructure: they reason over multiple steps, call tools, act through workflows, and adapt through memory and feedback. These systems create governance challenges that are not fully captured by traditional software or predictive ML technical debt. We define Agentic Technical Debt as the accumulated liability created when prompts, memory, tool schemas, orchestration graphs, control policies, and observability routines are patched together faster than they can be validated, standardized, and governed. We define Stochastic Tax as the recurring operating burden of keeping probabilistic agent behavior within acceptable bounds. The distinction matters: debt is a stock of design and governance liability, while the tax is a flow of operating cost that arises because stochastic agents act through tools and workflows. We outline how managers can make both visible through lightweight dashboards and governance controls.

A linear probe can decode a representation almost perfectly and yet be completely irrelevant to how the model uses it. On calendar-date duration reasoning in language models, a $\sin$/$\cos$ probe recovers day-of-year from a layer's activations, yet ablating its direction has no effect on the model's answers -- while ablating a four-dimensional subspace found by Distributed Alignment Search (DAS) at the same layer collapses performance entirely. We measure the angle between these two subspaces -- the \emph{readout-mediator angle} -- and find it indistinguishable from the angle between two random subspaces (the Haar-uniform null), meaning the probe has learned a direction orthogonal to the model's actual computation. Reverse-engineering the circuit reveals why: attention heads route month-grained context through learned QK offsets at ${\pm}30$ and ${\pm}61$ days, and MLPs then convert \emph{when} (absolute date) into \emph{how long} (duration) -- all downstream of the causal subspace the probe never touches. Sparse-autoencoder decomposition confirms the split: probe-aligned and DAS-aligned features encode semantically disjoint concepts with negligible causal overlap. The dissociation replicates across four scales ($1.5$-$9\,$B) and two model families, with preliminary evidence on two further domains (spatial displacement, symbolic arithmetic), suggesting that readout-mediator orthogonality is a general failure mode of probe-based interpretability. This directly undermines proposals to deploy probes as runtime safety monitors: the probe can report high confidence on a direction the model has silently abandoned.

Masked diffusion language models (MDMs) uniquely support any-order generation, with confidence-based decoding currently serving as the de facto standard inference policy. To optimize for this, recent training schemes attempt to align training mask patterns directly with those observed during generation. However, we argue that confidence-based decoding is inherently misaligned with the logical-flow trajectories required for complex reasoning, and that confidence-aligned training actively entrenches this misalignment. We make this concrete using multi-digit addition, where the decoding strategy prematurely predicts locally easy digits before resolving their long-range dependencies, producing high-confidence errors on challenging inputs. While traditional random masking keeps the failure rate low on this challenging tail, confidence-aligned training amplifies the error rate by an order of magnitude. Across five distinct reasoning tasks, this same pattern emerges with task-dependent severity: confidence-based decoding induces failures on highly complex inputs, and confidence-aligned training exacerbates them. In contrast, random masking -- despite its perceived inefficiency -- robustly preserves the reasoning-trajectory conditionals essential for solving the challenging tail.

It is often desirable to generalize medical imaging AI models trained with dense annotations to data acquired from different ultrasound scanners or clinical sites; however, retraining these models with new annotations is often difficult and costly. We examine this challenge in pediatric wrist fracture assessment using point-of-care ultrasound (POCUS), where fractures are common and can be effectively triaged via ultrasound. AI has shown radiologist-level performance for fracture detection, often aided by high-quality bony structure segmentation. However, due to significant domain shifts, models perform poorly on data from other centers or probes, and obtaining segmentation labels across devices is impractical due to manual annotation effort and data privacy concerns. To address this, we propose a target-informed self-supervised pretraining and model-ensemble strategy. Specifically, our approach combines masked image modeling (MIM) and contrastive learning to learn target-domain structural representations without labels, and introduces a confidence-aware infusion head to adaptively integrate predictions. The source dataset, collected with a Philips Lumify probe, contained dense labels, while the target dataset, acquired with a TeleMED portable probe, was unlabeled. The datasets were kept strictly separate throughout the entire process. Our method used labeled source data for supervised training and leveraged target-domain pretraining to improve generalization. On 318 images from 62 pediatric POCUS videos, this approach significantly improved cross-device performance, achieving over 6% Dice improvement on the target domain versus the baseline. These results demonstrate a label-efficient and privacy-preserving approach for cross-device-robust ultrasound AI, offering a framework that can be extended to multi-center studies or federated learning setups.

Computer use agents (CUAs) have shown strong potential for automating complex digital workflows, yet their training remains constrained by costly live environment interaction and limited high-quality supervision. Existing filtered behavior cloning pipelines suffer from imitation bottlenecks, including distribution shift from the expert demonstration and the absence of negative learning signals. Meanwhile, standard trajectory-level reinforcement learning struggles with sparse rewards, ambiguous credit assignment, and high infrastructure costs for long-horizon GUI interaction. In this work, we propose PRO-CUA, a process-reward optimization framework for training CUAs with iterative step-level reinforcement learning. PRO-CUA decouples on-policy environment interaction from policy optimization: the current policy collects states through live rollouts, generates diverse candidate actions for each state, receives step-level feedback from a process reward model (PRM), and is optimized with group-relative advantages. This design enables dense and flexible credit assignment without relying on golden answers or offline expert trajectories, while reducing distribution shift by training on the agent's own execution states. Experiments on live web benchmarks demonstrate the effectiveness of PRO-CUA and the reliability of PRM-guided step-level training.

When multiple LLM agents solve the same problem, standard practice compresses each agent's reasoning into a majority vote or layered synthesis, treating agreement as the finish line. We show this is unnecessarily lossy: an LLM aggregator that reads complete reasoning traces recovers correct solutions even when agents unanimously agree, with beneficial corrections consistently outweighing harmful ones -- the \emph{aggregation paradox}. Majority voting has a ceiling that perturbation diversity does not raise (error correlations are identical); the aggregator's gain comes from trace-level complementarity, assembling correct intermediate steps from minority chains that voting discards. These findings motivate Self-Consistent Mixture of Agents which generates trace diversity through semantic-preserving input perturbations, safeguards the majority via anchored refinement with provable non-degradation guarantees, and always synthesizes -- never gates on consensus. A single model with perturbation-induced trace variation outperforms heterogeneous model pools across structured reasoning, PhD-level science, competition mathematics, and competitive programming. The unit of aggregation should be the reasoning trace, not the answer.

Selecting efficient multi-step synthetic routes is a central challenge in organic synthesis, particularly in medicinal and process chemistry, where route choice directly impacts feasibility, cost, and development efficiency. Data-driven assessment systems often oversimplify the multi-objective nature of synthesis design and rely on proxy datasets, such as patent routes, rather than universally grounded criteria. To address this, we introduce an expert-augmented, data-driven scoring framework that integrates machine learning with chemists' domain knowledge for both numerical and explainable route assessment. A DeepSets-based model is trained using tree edit distance between reference and machine-generated routes, and then fine-tuned with expert evaluations to produce both quantitative scores and interpretable qualitative categories: Good, Plausible, and Bad. The resulting system achieves a Spearman correlation coefficient of 0.78 and a Pearson correlation of 0.77 for category assessment prediction, and 60.2% top-1 ranking accuracy for score prediction, substantially outperforming the previous baseline of 17.5%.

Model merging combines fine-tuned checkpoints into a single multi-task model without retraining. Existing methods - such as task arithmetic, model soups, TIES, and DARE - are computationally efficient and empirically successful, but rely on heuristic design choices and lack formal optimality guarantees. We show that merging can be formulated as a convex quadratic programme over residual updates, yielding weights that minimise a squared-output calibration objective using calibration inputs and fine-tuned model outputs, and subsuming existing methods as special cases. Our framework yields a closed-form diagnostic - the fraction of residual energy captured by a chosen basis - that predicts downstream merge quality using only the calibration set. Empirically, the QP matches or outperforms existing methods in the single-layer setting, and we characterise when the optimal basis provides significant gains over the cheaper diagonal QP. We extend to multi-layer merging via a sequential layer-wise algorithm and demonstrate consistent gains across language and vision benchmarks.

Reconstructing object geometry from radio frequency (RF) signals is fundamentally challenging due to the lensless imaging nature of RF sensing, which leads to low spatial resolution and high noise. Unlike light signals, RF signals can penetrate occlusions and thus capture information about hidden scenes. Existing Non-Line-of-Sight (NLoS) 3D neural reconstruction methods can recover coarse surfaces inside enclosed environments but often suffer from unstable optimization, noisy surface geometry, and surface ambiguity, failing to produce accurate zero-level sets from the signed distance field (SDF). These limitations largely stem from neglecting the role of Line-of-Sight (LoS) geometry outside the enclosed region, which provides valuable physical constraints for modeling signal propagation. In this paper, we introduce a Unified LoS and NLoS neural geometry reconstruction framework GeRaF 2.0 that leverages the outside LoS geometry to model and guide RF propagation from the LoS region into the NLoS region. By integrating visual LoS priors into the neural field formulation, GeRaF 2.0 achieves stable training and physically consistent reconstruction of both visible and hidden geometry, setting a new state-of-the-art in RF-based geometry reconstruction.

GeRaF is the first method to use neural implicit learning for near-range 3D geometry reconstruction from radio frequency (RF) signals. Unlike RGB or LiDAR-based methods, RF sensing can see through occlusion but suffers from low resolution and noise due to its lensless imaging nature. While lenses in RGB imaging constrain sampling to 1D rays, RF signals propagate through the entire space, introducing significant noise and leading to cubic complexity in volumetric rendering. Moreover, RF signals interact with surfaces via specular reflections, requiring fundamentally different modeling. To address these challenges, GeRaF (1) introduces filter-based rendering to suppress irrelevant signals, (2) implements a physics-based RF volumetric rendering pipeline, and (3) proposes a novel lensless sampling and lensless alpha blending strategy that makes full-space sampling feasible during training. By learning signed distance functions, reflectiveness, and signal power through MLPs and trainable parameters, GeRaF takes the first step towards reconstructing millimeter-level geometry from RF signals in real-world settings.

This paper examines records retrieved from the ClinicalTrials.gov registry to characterize temporal trends in AI terminology and the geographical distribution of AI trials. The work also reports on an exploratory hybrid human-AI approach to analyzing human-AI interaction trends in registered clinical trials. The hybrid workflow comprised a frontier generative AI model (GPT-5.5) and human review to screen and categorize records returned by an AI-focused search. The findings indicate a marked increase in AI-related trials over time, with recent growth in references to machine learning, deep learning, chatbots, GPTs, and large language models. Geographically, China and the United States accounted for the largest numbers of AI-related trials, with notable recent increases in several other countries including Italy, France, Spain, the UK and Turkey (Türkiye). In a random sample of 100 records, human and AI classifiers showed good agreement in identifying studies not substantively using AI, but lower agreement in classifying human-AI interaction, particularly where health professional interaction was ambiguous or insufficiently described. Overall, the results suggest that hybrid human-AI screening of clinical trial records is potentially viable, but clearer trial reporting and more precise interaction definitions will benefit the process.

Current face video forgery detectors use wide or dual-stream backbones. We show that a single, lightweight fusion of two handcrafted cues can achieve higher accuracy with a much smaller model. Based on the Xception baseline model (21.9 million parameters), we build two detectors: LFWS, which adds a 1x1 convolution to combine a low-frequency Wavelet-Denoised Feature (WDF) with a phase-spectrum channel derived from Spatial-Phase Shallow Learning (SPSL), and LFWL, which merges WDF with Local Binary Patterns (LBP) in the same way. This extra module adds only 292 parameters, keeping the total at 21.9 million, smaller than F3Net (22.5 million) and less than half the size of SRM (55.3 million). Even with this minimal overhead, the fused models increase the average area under the curve (AUC) from 74.8% to 78.6% on FaceForensics++ and from 70.5% to 74.9% on DFDC-Preview, gains of 3.8% and 4.4% over the Xception baseline. They also consistently outperform F3Net, SRM, and SPSL in eight public benchmarks, without extra data or test-time augmentation. These results show that carefully paired, handcrafted features, combined through the lightweight fusion block, can provide competitive robustness at a significantly lower cost than comparable frequency-based detectors. Our findings suggest a need to reevaluate scale-driven design choices in face video forgery detection.

Swarm and field robotics face significant barriers to real-world validation due to the high cost and development time to deploy hardware. This paper introduces the ``Bionic Swarm,'' a novel system that lowers these barriers by abstracting away many of the tasks that are difficult to implement on robots but which do not contribute to the overall algorithm evaluation, giving these tasks to human users. These human users take directions from a smartphone web-app that takes measurements from Bluetooth-connected sensors and relays them to a centralized server. This server runs the swarm algorithm and directs actions to the human users. We evaluate this system through the experimental validation of a geotechnically-focused search algorithm named Score-Biased-Search, which functions by assigning a ``score'' to each location on a reconstructed map, then biases search patterns through areas of higher expected scores, and which exhibits superlinear map reconstruction relative to the number of search agents. After presenting simulation results for the algorithm, we then apply the algorithm on the Bionic Swarm platform to validate its function in a real-world, outdoor setting. This work demonstrates that this human-in-the-loop approach significantly lowers the barrier to entry for field and swarm robotics research.

Recent reinforcement learning (RL) post-training approaches primarily optimize the final output policy using sparse outcome-level rewards, while largely overlooking predictive signals encoded in intermediate representations. In this paper, we introduce a new paradigm called on-policy internal self-distillation and propose the OISD framework, which improves reasoning by transferring on-policy predictive signals from the final layer to intermediate representations. During rollout and Group Relative Policy Optimization (GRPO) optimization, the final layer acts as both the policy and a detached internal teacher for selected intermediate layers, which are guided to align with it through two complementary mechanisms: logit alignment, which transfers high-level reasoning behaviors (how to think), and attention alignment, which enforces consistent attention patterns (where to look) from the final layer to the selected intermediate layer, both without requiring external privileged information. Our OISD, together with GRPO, employs signed advantage-weighted Jensen--Shannon alignment to distill informative intermediate representations while preserving policy consistency under a unified acting policy. Experimental results demonstrate the effectiveness of OISD, with substantial and consistent improvements over strong reasoning RL baselines across four mathematical reasoning tasks. The code will be released at https://github.com/THE-MALT-LAB/OISD

Synthetic Aperture Radar (SAR) imagery enables all-weather, day-and-night Earth observation; however, it remains difficult to interpret due to speckle noise and other intrinsic imaging artifacts. Sentinel-1 (S1) constitutes one of the most widely used spaceborne SAR missions, offering systematic global coverage, high temporal resolution, dual-polarization imaging, and free data availability. Among S1 modes, Stripmap (SM) provides the highest resolution, yet speckle noise and spatial constraints often hinder applications requiring finer spatial detail. This motivates the need for effective image enhancement strategies. In this work, we propose a self-supervised enhancement framework for S1 SM imagery based on azimuth subaperture decomposition. The method exploits the physical consistency between subaperture reconstructions and the corresponding full-aperture image to generate paired training data without external sensors, simulated ground truth, or multi-temporal stacks. The proposed framework integrates single- and multi-frame learning and incorporates an iterative inference scheme that progressively refines image quality. Experiments on real S1 SM data show that the proposed approach consistently outperforms the widely adopted self-supervised deep learning baseline MERLIN, in terms of PSNR and SSIM, while MERLIN attains higher ENL, highlighting a trade-off between structural fidelity and speckle smoothing. Overall, the results demonstrate that subaperture-based supervision provides a physically grounded, reproducible, and operationally viable approach for SAR image enhancement using S1 data. It is worth noting that the proposed approach can be extended to other SAR platforms, polarizations, and acquisition modes.

Reasoning models are evaluated on single-turn benchmarks but deployed in multi-turn dialogue, where users push back on correct answers. Under sustained adversarial pressure we find a previously undocumented failure mode: the chain-of-thought stays factually correct from first turn to last while the emitted answer flips wrong. We call this unfaithful capitulation (UC) and isolate it with a $2\times 2$ latent-versus-behavioral framework that flip-rate metrics and single-turn faithfulness probes both miss. Across three datasets (MT-Consistency, MMLU-Pro, GSM8K), the latent-correct rate at the behavioral flip clusters near 50% in think mode and collapses to 11-15% under no_think -- paired, within-model causal evidence that reasoning creates the gap. Across models the effect tracks the reasoning channel (high in Qwen3-32B and GPT-OSS-20B, low in inline-CoT Gemma-4-31B-it). An independent GPT-4o judge corroborates $86\%$ of UC labels; a token-level probe shows the answer-slot argmax is correct in $84\%$ of UC cells; and a naive trace-anchored defense backfires. We release all trajectories, traces, and judge labels.

A retrieval-augmented generation (RAG) system deployed over a multi-author institutional corpus can give a different answer to the same question depending on which source it retrieves -- a failure mode the dominant single-gold-answer paradigm cannot diagnose. We argue that source-dependence is a missing axis of NLP evaluation, and that auditing it means shifting the unit of evaluation from answer correctness to the inter-source relationship. We make this concrete in transplant patient education, where institutional sources demonstrably disagree, releasing three artefacts: TransplantQA, a benchmark of real patient questions, each answered by grounding generation in multiple institutional handbooks as candidate sources; HERO-QA, a hierarchical retrieval strategy that grounds and audits each answer; and a structured-output judge that scores inter-source relationships on a validated 5-label taxonomy. At scale, better retrieval reveals far more disagreement than prior estimates suggested -- understating its prevalence, not its intensity. The framework is domain-agnostic and transfers to legal and educational RAG: measuring source-dependence is a responsibility for deployed multi-source NLP generally.

AI agents are increasingly expected to operate as digital employees: accessing enterprise data, making decisions, and taking actions autonomously. But agents are simultaneously less predictable than humans -- prone to hallucination, misinterpretation, and adversarial manipulation -- and more technically capable: with deep system knowledge and high-throughput interfaces cascading damage at machine speed. This combination makes it unsafe to rely on agents to faithfully interpret or propagate security-critical metadata such as access policies, data classifications, and behavioral constraints. We present the Redpanda Agentic Data Plane (ADP), an architecture built around out-of-band metadata channels: infrastructure pathways that carry security context, policy signals, and audit trails deterministically, entirely outside the agent's read and write path and across heterogeneous infrastructure. These channels enforce governance at every stage of the agent lifecycle -- scoping data access on the way in, constraining actions during execution, and capturing tamper-proof transcripts on the way out. We demonstrate ADP with a multi-agent portfolio rebalancing system in which autonomous agents monitor markets, make trade decisions, and execute orders across isolated client accounts -- with per-client data scoping, trade approval thresholds, and tamper-proof audit trails all enforced by out-of-band channels the agents can neither see nor bypass.

Event-driven scheduling policies are increasingly deployed in industrial environments, where decisions are made under asynchronous and partially observed system states. As a result, decision states are not temporally consistent, action admissibility is not explicitly defined, and the origin of execution errors remains ambiguous. These issues limit both reliability and interpretability. To address this gap, a policy-neutral execution and measurement layer is proposed to mediate between scheduling policies and the industrial execution environment. The layer constructs decision-valid snapshots from asynchronous event streams, defines a standardized execution contract with explicit action admissibility, and records outcomes as divergences between policy intent, transactional outcomes, physical execution, and human intervention. This enables a separation between decision semantics and execution behavior and makes deployment mismatch observable and structurally attributable. The proposed framework is evaluated using a discrete-event simulation. The results show analytical benefits across all observation lag regimes, as undifferentiated execution failures are transformed into structured, typed outcomes with full attribution coverage. Operational benefits are strongest under low observation lag, where avoidable execution errors can be prevented before commitment. Overall, the layer turns execution uncertainty into supervisory data for evaluation and policy refinement.

LLMs encode both general capabilities and domain-specific knowledge in a single set of parameters. We ask whether this capacity can be reorganized: keeping broadly useful computation in a shared backbone, while moving specialized knowledge into external memory modules. We propose \emph{knowledge offloading} (KOFF), a framework for decomposing a pretrained LLM into a sparse shared backbone and domain-specific memories. Starting from a frozen base model, we jointly learn a structured pruning mask and lightweight recovery modules, implemented as LoRA adapters and learned key-value caches. Across Llama and Qwen models from 3B to 8B, we find that non-trivial capacity can be moved out of the shared backbone without a large loss in model ability. At around 12\% global sparsity, KOFF preserves much of the unpruned model's performance, while pruning the same frozen model without memories degrades sharply. Ablations show that LoRA and learned KV memories are complementary, and specialization analyses suggest that the learned decomposition is meaningful: language-specific neurons are preferentially removed while language-general neurons largely remain in the backbone. These results suggest that knowledge can be reallocated between a shared core and swappable external memories.