Papers

Self-regulation together with mutual promoter competition provides a simple mechanism for bistability in gene-regulatory models. We study a two-gene system with regulatory terms of Hill exponent one, allowing distinct basal production rates and distinct degradation rates. Each gene product, when bound to its own promoter, may enhance or reduce production relative to the basal rate, while the two products compete through promoter occupancy. We show that the system has at least one and at most three equilibria in the positive quadrant. Exactly two positive equilibria can occur only if one nullcline intersection is degenerate; consequently, a configuration in which all positive nullcline intersections are transverse has either one or three positive equilibria. If there are exactly three distinct positive equilibria, then all three are automatically hyperbolic: the two outer equilibria are asymptotically stable nodes and the middle equilibrium is a saddle. Moreover, every positive solution converges to an equilibrium. Hence the positive quadrant is the disjoint union of the basins of attraction of the two stable nodes and the one-dimensional stable manifold of the saddle, yielding global bistability.

Humanoid kicking requires coordinated whole-body motion and precise contact, while a banana kick demands contact mechanics that generate ball spin and aerodynamic curvature. Motion imitation provides a reliable ordinary-kick prior, but reinforcement learning may improve shot speed and placement accuracy without changing the underlying kicking technique. Adapting this prior to a qualitatively different contact-rich skill can fail even when the reward is dense and optimization remains stable. The failure occurs when the task objective is locally flat over the current policy's responses. We term this condition first-order learning starvation. To address it, we propose response-informed skill evolution (RISE), a closed-loop objective-continuation method for policy adaptation. RISE ranks bounded objective changes using response sensitivity estimated from cached rollouts and accepts updates only when they produce verified response progress while preserving kicking reliability. Our analysis shows that rescaling a saturated spin reward cannot recover first-order sensitivity at zero spin, whereas adapting coupled contact responses can provide a learnable path to spin generation. We integrate RISE into a humanoid kicking pipeline under calibrated contact and Magnus-force aerodynamics, and sim-to-real transfer. Experiments show that RISE evolves the ordinary kick into a high-spin curved kick with 11.55 rad/s mean ball spin, improves the mean evaluation score by 19.8% over a learning-progress curriculum, and raises joint target attainment from 15.2% to 50.9%. Ablations and response diagnostics support the mechanism, while 30 motion-capture-recorded physical trials demonstrate consistent hardware transfer of the learned curved kick. Project website: https://haozhang-thu.github.io/bananakick/

We study how determinant recurrences of banded Toeplitz matrices behave when their Laurent symbols are multiplied. Two classical structures underlie the problem. Clean banded Toeplitz determinants have a root-product description of their recurrences, while a product of two finite Toeplitz sections differs from the finite section of the product by finite-rank corner corrections. We connect these structures by identifying multiplication-induced boundary defects with exterior-degree sectors of the determinant recurrence. To each polynomial core we associate recurrence polynomials for all exterior degrees. Multiplication of cores becomes a graded convolution via composed products, while complementary degrees are related by a scaled reciprocal duality. On the matrix side, one-sided factors produce a defect filtration whose successive pieces are exactly these sectors. For general two-sided factors, independently weighting the two corner corrections gives an interval filtration: increasing either defect order adds one adjacent sector, and the cumulative recurrences are generically minimal. The multiplication-induced same-corner minors used in these filtrations lie in the standard row-column state module, whereas cross-corner imbalance corresponds instead to Laurent recentering. Opposite Laurent recenterings realize the individual sectors directly, and for several factors the admissible recenterings form an explicit lattice polytope. The pentadiagonal case gives the basic $1+4+1$ decomposition. The resulting framework gives a factor-level description of how finite-section boundary effects generate the recurrence spectrum of structured banded Toeplitz products.

CXL-enabled memory expands server memory capacity, but introduces a page-placement problem: the operating system must decide which pages should reside in DRAM and which should reside on slower CXL memory. Existing systems make this tradeoff in one of two ways. Userspace controllers support flexible policies, but expose placement decisions to scheduler jitter and kernel-userspace crossing overhead. Kernel-space systems avoid this latency, but rely on fixed heuristics that must generalize across workloads. We present xTier, a kernel-resident learned memory-tiering system. xTier attaches eBPF programs to PEBS events and uses a compact quantized MLP to score sampled pages inside the kernel at microsecond-scale latency. Rather than reacting to every candidate, xTier converges to a low-churn placement for the current workload phase, reduces sampling cost after convergence, and returns to a higher sampling cadence when the workload shifts. We evaluate xTier on six memory-bound workloads at DRAM:CXL ratios from 1:5 to 1:25. The advantage grows as the DRAM budget tightens. At 1:15 and beyond, xTier is the fastest system in 14 of 18 configurations. Where it is not fastest, it trails the best baseline by 3.9% on average. It reaches this performance while moving 13% fewer pages in geometric mean, and 22% fewer at the tighter ratios. When a workload changes phase, xTier rebuilds its hot set in DRAM faster and more completely than any baseline.

Secondary use of educational records is increasingly mediated by platforms that share a differentially private synthetic version of a dataset and validate specific findings against the real data on request. The synthetic version is evaluated by comparing summary statistics of each variable, yet reported confirmation rates suggest that such comparisons do not predict which findings survive. We propose a structural check: the number of connected components of a weekly proximity graph over learners, tracked across a term. Across four annual cohorts of lower-secondary study-habit logs, the synthetic versions reproduced the level of this quantity and the shape of the weekly partition, but its variation across the term was between 2.6 and 4.9 times smaller than in the real data at a common working point, without exception, and those changes fell in different weeks: the synthetic cohorts single out the term's examination weeks and the real cohorts do not. We also show that a routine rule for setting the graph threshold makes naive comparisons between two datasets invalid, and illustrate this with an error of our own. The real curves are also distinguishable from marginal-preserving surrogates of themselves in all four cohorts, where three of the four synthetic ones are not, a comparison that needs no real data; these differences trace to what the generator was given.

We present GaussPDE, a framework that injects physically structured partial differential equation (PDE) dynamics into pretrained 3D Gaussian scenes without mesh extraction, voxelization, or retraining. Our key observation is that PDE rendering requires not only accurate appearance, but also a reliable discrete computational domain. We therefore first introduce camera-aware regularization during 3DGS reconstruction to suppress camera-near floaters and oversized primitives that would create unstable graph topology. We then construct an active Gaussian graph using covariance-aware distances and opacity, appearance, and boundary-aware conductance, enabling mass-weighted graph Laplacian PDE evolution directly over Gaussian primitives. The evolving scalar PDE state is coupled back to rendering by modifying the direct-current spherical harmonic color coefficients while preserving geometry, opacity, and view-dependent rendering behavior. Experiments on real and synthetic scenes show that GaussPDE produces stable, controllable, and spatially coherent dynamic visualizations, with reduced cross-boundary leakage compared with baselines.

Agentic workflows shift software development from prompting AI agents for individual tasks to defining recurring work that agents execute automatically. GitHub Agentic Workflows (gh-aw) enables this approach through Markdown files that combine natural-language instructions with configuration and compile into executable GitHub Actions workflows. Unlike conventional workflows that primarily prescribe scripted operations, these files delegate tasks requiring interpretation to AI agents. They also couple agent instructions with execution triggers, making those instructions operational specifications for repeated repository activities. However, how developers structure and maintain these specifications, and which execution requirements and safeguards they express, remains insufficiently understood. In this paper, we examine the structure, evolution, and instruction content of gh-aw Markdown files to inform how practitioners define and maintain agent-run work. We analyze 1,248 files from 276 repositories, 20,841 commit-file events, and 288 resolved instruction-label sets from a sample of 294 files. Our results show that workflow instructions extend beyond short prompts, with a median of 556.5 words and code blocks in 62.1% of files. Among files with at least 120 days of observed activity, 78.2% still receive updates in month 4, while size-normalized churn decreases after the first month. Tasks, outputs, constraints, and process instructions each appear in over 93% of labeled workflows, yet only 9.4% explicitly address prompt-injection defense. LLM classification achieves an F1 score of 0.818 and Cohen's Kappa of 0.715 against the resolved human labels. These findings suggest that developers should account for the evolution of workflows copied or referenced across repositories and consider adding prompt-injection defenses, resource budgets, and evidencecredibility checks where applicable.

Production customer-support systems often require LLMs to support multiple skills, such as intent classification, question answering, summarization, or tool-use decisions. A central deployment question is whether these skills should be handled by separate task-specialist models or by a single model trained through multi-task training, sequential updates, or model merging. We study this question using thirteen models spanning five families (Qwen3, Qwen3.5, Gemma-3, Llama-3.1, and Mistral) from 0.6B to 32B parameters across eight customer-support datasets, spanning four public and four proprietary datasets with approximately 74.5k training and 8.7k evaluation samples. Under a fixed training protocol, we train more than 200 checkpoints. Our experiments reveal that multi-task full fine-tuning is the strongest operational default at every model size we test. Specialist models are strong on their target tasks but often degrade sharply off-task, making reliable routing important. Sequential Low-Rank Adaptation (LoRA) preserves earlier skills better than sequential full fine-tuning, while merging a specialist with its base model improves off-task robustness with limited same-task loss for larger models. We conclude with practical guidelines for selecting fine-tuning strategies in real-world settings.

This paper investigates the anti-localization uplink communication in a satellite-terrestrial system, where a ground transmitter Alice communicates with a legitimate satellite receiver Bob in the presence of multiple cooperative adversarial satellites attempting to localize Alice with the time difference of arrival (TDOA) technique. Specifically, we propose a cooperative jamming-based scheme for such anti-localization communication,in which Alice exploits the superposition coding with power allocation to simultaneously transmit information/jamming signals for communication with Bob and for confusing signal detection/TDOA measurement at adversarial satellites, while Bob employs the combining vector technique to enhance the desired information signal and also suppress the jamming. We define a localization error probability (LEP) metric to jointly depict both the impacts of signal detection and TDOA measurement on localization performance, and then develop a theoretical framework for the LEP modeling under the proposed scheme. We further explore the joint optimal design of jamming coding and power for LEP maximization, subject to the constraints of AliceBob communication reliability and Alice's transmit power. An effective sample average approximation method is also provided to tackle this non-convex optimization problem. Finally, extensive numerical results are illustrated to validate our theoretical models and demonstrate how the cooperative jamming helps to provide an anti-localization guarantee while ensuring communication reliability

Enterprise data agents must preserve organization specific semantics, not just translate questions into queries. We present ChinaUnicom DataAgent (UniDataAgent), an ontology grounded system for reusable question-to-report analysis that separates semantic acquisition from online execution. Ontology Acquisition and Validation stage (OAV) builds versioned enterprise ontologies from metadata, business knowledge, and supporting materials through expert authored business skills, constrained generation, question verification, and selected expert review. Question-to-Report Execution (QRE) stage retrieves semantic contracts for each question, coordinates skills and data tools, validates results, and produces evidence linked reports. Across 27 enterprise tables and roughly thousands of metric types, ontology construction took a few hours instead of about one week manually. It took just a few minutes to generate the reports, instead of several working days. Ontology grounding achieved 95.0% strict accuracy on real business questions, versus 72.5% for document RAG, especially on structured and compositional tasks. The system has already been deployed to generate cost savings and has the potential to be replicated in other enterprises.

The Platonic Representation Hypothesis suggests that increasingly capable models converge toward shared representations. Recent work narrows this claim to shared local neighborhood relationships, finding that capacity-dependent trends in several global similarity measures largely disappear after calibration. We challenge this interpretation by showing that prior local-global comparisons confound structural scale (local versus global) with what is compared: relational structure, defined by which samples are related, versus metric geometry, characterized by quantitative relations such as distances, similarities, or correlations. To disentangle these factors, we construct a controlled $2\times2$ framework that evaluates both relational structure and metric geometry at local and global scales. We introduce $H_0$ skeleton overlap as a global counterpart to mutual $k$-nearest neighbors, together with matched distance-aware variants. Across vision-language models, relational structure exhibits robust representational convergence at both scales after calibration, whereas increasingly stringent distance agreement substantially weakens alignment and progressively flattens the capacity-dependent trend. We further extend the analysis beyond ambient Euclidean geometry by evaluating distance agreement under a Riemannian metric approximation and recover the same structure-geometry pattern. The pattern is also reproduced in video-text representations. Together, these results show that relational convergence extends beyond local neighborhoods to global spanning structure, whereas metric geometry exhibits substantially weaker convergence.

The Schrödinger bridge owes its computational power to a single structural fact: by Girsanov's theorem the controlled problem is a Kullback–Leibler (KL) projection onto a fixed reference measure, solvable by alternating projections. This letter shows that the fact does not survive risk sensitivity. When the expected path cost is replaced by the entropic risk measure and both endpoint marginals are kept as hard constraints, the resulting fixed-point bridge value $J_\theta$ (the soft-problem value at the multiplier that enforces the terminal constraint) admits no representation as a constrained KL minimum against any fixed path-space reference with a regular endpoint law (a class strictly larger than the uniformly elliptic diffusion references: no Markov property is required), even allowing an additive normalisation depending on the initial marginal. Moreover, no single reference generates the one-parameter family in the risk parameter. The obstruction is computed in closed form: the Gaussian bridge value violates, by exactly $\theta/2$, a heat equation that any Gaussian smoothing of a fixed endpoint density must obey. In place of the projection, the theory rests on a terminal-multiplier fixed point and an asymmetric factorisation penalising the score energy of the backward factor.

Edge AI model deployment is a multi-stage engineering process involving model conversion, operator compatibility handling, runtime integration, and precision verification. While prior work has demonstrated agent-based automation for Qualcomm AI Runtime, the broader edge inference runtime ecosystem, including Intel OpenVINO, Rockchip RKNN, NVIDIA TensorRT, and ONNX Runtime, presents distinct toolchains and optimization strategies. This paper extends AIPC (AI Porting Conversion, an LLM agent-driven methodology for AI model deployment automation previously demonstrated on Qualcomm AI Runtime) to multi-runtime scenarios, proposing an LLM agent-driven approach for automated single-model-to-single-runtime deployment across heterogeneous inference backends, such as Intel OpenVINO, Rockchip RKNN, NVIDIA TensorRT, and ONNX Runtime. We decompose the edge AI deployment into standardized, verifiable stages, and inject runtime-specific domain knowledge into the agent execution flow through agent skills, auxiliary scripts, and staged verification loops. Using representative vision models, we demonstrate that agent-based deployment can complete the conversion from a PyTorch model to its executable inference, targeting OpenVINO for x86/NPU, RKNN for RK3588, TensorRT for NVIDIA GPU, and ONNX Runtime for Qualcomm NPU with a focus on FP16 precision deployment feasibility verification. The contributions of this paper primarily lie in providing multi-runtime deployment engineering practice experience, toolchain mapping analysis, a layout-adaptation and inference-replacement layer that removes manual transpose insertion from the agent's repair burden, and an empirical characterization of agent deviation behavior under structured knowledge injection, rather than large-scale systematic benchmarking or cross-runtime operator repair strategy comparison.

Next-token prediction has enabled highly fluent autoregressive language models, but it represents global structure only indirectly through sequential factorization. In contrast, high-fidelity autoencoders have become a standard primitive in image generation, enabling generative models to operate over continuous latent spaces; text lacks a comparably faithful continuous representation. We propose LLMAE, a method for repurposing a pretrained decoder-only language model as a continuous text autoencoder by exposing the activations of an intermediate layer as a fixed-length latent bottleneck. LLMAE achieves this interface with structured attention masks and LoRA adaptation, leveraging the generative prior of the original LLM. Across two backbones (270M Gemma 3 and 0.5B Qwen2.5), LLMAE reconstructs sequences up to 1024 tokens with high accuracy, reproducing up to 97% of documents verbatim. We demonstrate downstream utility by training a lightweight diffusion model that generates detailed image captions directly in the frozen LLMAE latent space. By mapping text into a fixed-length continuous latent space, our approach provides an effective substrate for downstream adaptation while benefiting from the fluency of the original LLM. Code and models are available at https://github.com/arkanath/LLMAE.

A robot returning a block to its origin tray may encounter two task-consistent pasts that reconverge to the same current input but warrant different actions. Yet memory-policy evaluations often rely on task success or action change under memory perturbation, neither of which establishes that memory guides the decision. We propose the Counterfactual Memory Audit (CMA), an evaluation protocol that crosses two histories at a verified-identical present, queries a frozen policy under common randomness, and evaluates each saved action under both pasts. This separates memory sensitivity, warranted choice, matched-world physical value, and per-pair reliability. On Mem-0, every audited Put Back pair changes action, but only $20/64$ pairs are fully reliable; at a later Swap decision, all paired actions change while both memories select the same branch. Native interventions further show closed-loop influence: replacing the history bank redirects behavior toward the replaced content, while restoring a 4096-byte protected anchor recovers $38.9$ points of Swap success lost to injected bank faults. On a dual-arm physical platform, memory changes saved actions, yet five of nine completed Put Back manipulations reach the wrong target. These results show that a robot can remember and react without reliably using memory to choose the behavior its past warrants. CMA provides a decision-level audit for distinguishing these cases.

As embodied agents take on increasingly social and relational roles in VR, visual realism and embodiment alone may be insufficient; users must also perceive these agents as emotionally attuned, supportive, and humanlike. Prior work suggests that verbal attunement and nonverbal mimicry can each improve users' social evaluations of embodied agents. However, behavioral mimicry has largely been studied outside of real-time, conversational AI interactions, leaving limited understanding of how users respond when an agent simultaneously generates contextually responsive dialogue and adapts its nonverbal behavior during an immersive conversation. To address this gap, we developed an embodied AI counselor that combines conversational AI with real-time facial-expression and posture mimicry, while producing either verbally attuned or neutral responses. We evaluated the system in a 2 X 2 within-subjects study with 20 participants, manipulating verbal attunement and behavioral mimicry. Results showed that verbal attunement was the most reliable driver of perceived empathy. Behavioral mimicry showed a marginal relationship with perceived humanness, while greater mimicry exposure showed preliminary, exploratory positive associations with empathy, positivity, and humanness, particularly among female participants. Together, these findings show that multimodal synchrony is not a simple additive strategy for designing empathic conversational agents in VR and underscore the need to consider how verbal and nonverbal behaviors are combined during real-time interaction.

A neural network's layers can be treated as time steps of a state-space model, turning Bayesian training into a smoothing problem: a forward pass propagates Gaussian moments through the network, and a backward Rauch–Tung–Striebel pass updates the weight posteriors in closed form. Such methods learn from each observation in a single pass, in an uncertainty-aware manner, and without gradient-based iterations or replay, which makes them well suited for online adaptation and data-efficient learning. Existing smoothing-based methods, however, are restricted to diagonal covariances across activations, discarding correlations between neurons. We overcome this limitation via a cross-covariance identity that enables full-covariance propagation through a network's nonlinear activations. We derive a one-step-per-layer smoother that approximates as Gaussian only each layer's affine output, and that applies both to deterministic systems with noisy observations and to stochastic systems described by output statistics. We demonstrate this method in non-stationary classification, online dynamics learning, and policy adaptation of a vision-language-action model, and find that it is generally more accurate than other smoothing-based methods.

Variational data assimilation computes a Bayesian update through optimization on a fixed posterior geometry, whose conditioning is determined by the prior covariance and the observation operator. In strongly anisotropic, weakly observed, or highly informative regimes, this geometry can become severely ill conditioned, leading to substantial sensitivity to solver design and preconditioning. This work investigates an alternative computational route based on flow-based transport, using the ensemble score filter as a training-free realization. Rather than repeatedly optimizing over the full posterior geometry, the flow-based update transports the predictive distribution through a sequence of intermediate distributions whose covariance and curvature are progressively regularized before reaching the target posterior. In the linear-Gaussian setting, we characterize this mechanism through the evolution of the diffused posterior covariance and the associated time-dependent curvature, revealing a continuation from a near-isotropic reference geometry to the ill-conditioned posterior geometry targeted by variational assimilation. Numerical stress tests show that this geometric regularization makes the flow-based method substantially less sensitive to severe ill-conditioning than the variational method, while a heterogeneous $10^4$-dimensional Lorenz-96 benchmark demonstrates improved robustness and lower delivered computational cost. These results identify posterior-geometry regularization as a key mechanism distinguishing transport-based from optimization-based data assimilation.

LLMs can generate programs for ARC-AGI-2 tasks, but the provided compute only allows a small number of attempts to generate, debug and validate solutions. Genetic algorithms can search and test many more programs, but random search rarely starts in a useful neighborhood of the solution space. We combine the two methods through a compact domain specific language (DSL). First, a quantized Qwen3.5-4B LLM generates an initial set of programs for each ARCAGI-2 task. Then, we use those programs to seed an initial population of starting programs, and use genetic algorithms to evolve these programs towards a solution to the given task. The DSL is designed such that every mutated program remains valid and can be executed. The initial programs proposed by the LLM solve 2 (3.3%) of the first 60 tasks of the ARC-2 public evaluation set. The genetic algorithm solves an additional 4, giving 6 correct test outputs in total (10.0%). If we try using evolving solutions without this LLM seeding, we do not arrive at any solutions at all. The results show that genetic search can improve programs generated by LLMs and produce additional correct solutions.

This work revisits a fundamental question in active learning: how powerful is the ability to synthesize arbitrary queries? Compared to pool-based active learning, where the learner only selects queries from a given unlabeled pool, it has been observed that this seemingly mild change in query ability may dramatically alter the difficulty of statistical learning. In particular some hypothesis classes, such as two dimensional halfspaces, permit only polynomial error decay in the number of samples when the learner must draw its labels from the pool, yet become exponentially learnable once synthesized queries are allowed. This striking gap suggests that membership query synthesis induces a fundamentally different mode of learning, one that is not adequately captured by existing active learning theory and calls for new analytical tools to characterize its sample complexity for general hypothesis classes. Motivated by this phenomenon, we develop several sufficient conditions, present intriguing examples, and propose a conjectural perspective toward understanding which hypothesis classes admit efficient learning through synthesized queries.

In this paper, we present the Berkeley QUAD (Quasi-direct-drive, Underactuated, Asymmetric Design) Hand, a four-finger anthropomorphic robotic hand with 11 degrees of freedom and 8 degrees of actuation. The design utilizes QDD actuation at the base of each finger, enabling high force transparency for dexterous, adaptive performance. However, the low torque density of these actuators traditionally presents major issues with size, weight, and thermal limits. We overcome this by applying bio-inspired asymmetry, delegating dexterity to the radial fingers through individual QDD actuation, and strength to the ulnar finger through an underactuated, compliantly coupled transmission driven by a larger QDD motor. A novel preloaded, linkage-based transmission permits this ulnar coupling in a way that preserves human-like workspace reachability. Under light loads, the ulnar motor drives the third (middle) finger directly for dexterity while the fourth (ring) finger mirrors its motion. However, under larger loads, the middle finger complies while the motor drives the ring finger further downwards and inwards toward the center of the grasp to apply better closure forces. Hardware evaluations validate this architecture, demonstrating that the hand achieves 29 out of 33 Feix taxonomy grasps and exhibits backdrive forces as low as 50 g for delicate interactions. Additionally, the underactuated fourth finger improves grasp closure and provides the spatial efficiency necessary for larger actuation, yielding up to a 96-fold reduction in heat generation during sustained loading. Webpage: https://benudavis.github.io/berkeley-quadhand/

Oriented ship detection in very high resolution (VHR) remote sensing imagery remains challenging due to elongated hull geometry and dense target distributions in complex port scenes. Existing methods typically address geometric representation and duplicate suppression separately. To jointly tackle these issues, we propose an oriented ship detector with two complementary components. The C3k2_Strip module employs orthogonal strip convolutions to better capture elongated hull structures, while the Class-Aware Direction-Aware Exclusion Loss (CA-DAEL) suppresses redundant predictions using class, direction, and confidence cues. Experiments on HRSC2016 and DIOR-R achieve 78.45% and 53.71% mAP50:95, respectively, with only 2.91M parameters. On HRSC2016, the proposed method improves mAP50:95 by 6.32 percentage points over the YOLOv11-OBB baseline, demonstrating its effectiveness for accurate oriented ship detection.

We present CoBranchMR, a mixed reality (MR) system that enables distributed collaborators to work in parallel from different locations on the same digital representation of a physical object. CoBranchMR lets users branch an object into editable virtual copies, customize them independently, and then merge their work back into a shared object. When merging copies, the system displays potential conflicts on the object's surface and provides several resolution options. By adopting branch-and-merge workflows for embodied spatial collaboration, CoBranchMR introduces a new collaborative interaction model that supports parallel design, conflict resolution, and negotiation in remote creative work.

AI virtual cells aim to predict cellular responses to specified interventions, yet held-out predictive performance alone does not establish use of the supplied perturbation information. This prediction-claim gap matters in agentic model discovery, where language-model agents generate and revise predictors using score-based feedback. We introduce CELLAUDIT, which audits input-use claims by asking whether an input can enter the cited computation, whether fitted predictions depend on it, and whether that dependence improves prediction of observed response. On a paired morphology-transcriptomics perturbation benchmark (BBBC047), an agent-selected predictor attains a mean held-out Global Pearson correlation coefficient (PCC) of 0.3153 but remains invariant to compound replacement; a control-profile-only predictor reaches 0.3142. Source inspection identifies a compound-query pathway blocked by singleton key-value attention, and the invariance persists after refitting with disjoint control wells. In a stratified audit of 48 candidates across two linked tasks, 47 change predictions under compound replacement on both held-out folds, but only 20 show target-loss gains with intervals above zero on both folds. On BBBC047, falsification-guided revisions recover positive mean compound contributions while retaining gains over the control-profile-only baseline. In matched sci-Plex searches, audit-enriched feedback yields higher held-out performance and larger mean compound and dose contributions across five trajectories, although paired intervals span zero. Refitting fixed designs on an independently acquired cohort shows predictive generalization need not imply generalization of input-use claims: dose contribution persists, whereas support for compound identity does not. CELLAUDIT adds a falsification layer to agentic model discovery, moving from generate-score-revise toward discover-falsify-revise.

Transformer language models process sequences token by token in an autoregressive manner, making growing contexts increasingly expensive. Yet many adjacent token spans are highly predictable or frequently occur as stable units, suggesting that their representations may be compressible. We introduce a method for distilling sequential computation by replacing spans of input tokens with collapsed representations, computed on the fly by a lightweight merge module. This module generates a single surrogate embedding from a sequence of static token embeddings that captures the functional role of the multiple tokens, allowing pretrained models to operate on compressed inputs without architectural changes or re-training. We apply this approach during inference to compress both prompts and intermediate decoding steps, using a rollback mechanism to substitute stored multi-token KV cache entries with their single-step surrogates. Experiments across diverse models show that the merge module can be used to reduce effective sequence length by up to 40% with minimal accuracy degradation across language modeling evaluations and downstream tasks, including question answering, summarization, commonsense reasoning, and long-form mathematical reasoning. Additional lightweight adaptation of the merge module further improves the accuracy-compression trade-off in selected settings. These results demonstrate that sequential token computation in Transformers can be effectively approximated through condensed surrogate representations that approximate the original behavior without model updating.

Scaling laws have guided large-model development in computer vision and natural language processing, but the relationships among data, model size, and compute remain unclear for functional magnetic resonance imaging (fMRI) foundation models. Here, we conduct a controlled empirical study using pretraining data from more than 200 source datasets and over 10,000 GPU-hours of experiments. Holding the pretraining framework and downstream protocol fixed, we vary pretraining data size, model size, and training duration. Downstream performance generally improves with compute, yet models using similar compute can perform substantially differently. Additional pretraining data bring larger gains at larger model sizes, suggesting that data and model size should be scaled together. At matched compute, increasing pretraining data benefits more tasks than increasing model size, although the pattern varies across tasks. We then use in-distribution (ID) downstream performance to select the combination of pretraining data size, model size, and training duration at two fixed compute budgets. The resulting models are locked before out-of-distribution (OOD) evaluation. They achieve the highest average performance across the evaluated OOD tasks among the compared fMRI foundation models while using less pretraining compute. Overall, our results show that compute alone does not characterize fMRI scaling: performance depends on how pretraining data, model size, and training duration are combined.

When designing algorithms for geometric graphs, exploiting structural parameters can lead to significantly improved bounds. Two prominent parameters in this context are $c$-packedness and $\lambda$-low density, both of which locally restrict graph complexity. Parameterized algorithms based on these parameters have been developed for computing well-separated pair decompositions, balanced separators, as well as distance oracles. Nevertheless the practical applicability of algorithms parameterized by $c$ or $\lambda$ remains unclear. While $c$-packed and $\lambda$-low-density graphs have been proposed as realistic models for road networks, the actual parameter values of large real-world instances have so far remained unknown, and existing theoretical guarantees are partially too loose for practical usage. In this paper we first devise scalable implementations for the approximate computation of $c$ and the exact computation of $\lambda$. Our experiments on road networks with millions of edges reveals a significant gap between the two parameters. On the theoretical side we prove that $c\in O(\lambda\sqrt n)$ which complements the known result that $\lambda\in O(c)$. Furthermore we present improved parameterized algorithms for balanced separator computation that reduce the separator size in theory and practice. We also show how to compute a tree decomposition with a width linear in the respective parameterized balanced separator size in polynomial time. This structural result yields a variety of new algorithmic consequences. Among them is an exact distance oracle with query time $O(c)$ for $c$-packed graphs after polynomial-time preprocessing, which improves upon the previous $O(c\log n)$ bound. Our experiments show that the proposed techniques efficiently produce small balanced separators and enable the construction of concise exact distance oracles on large road networks.

Objective assessment of robotic surgery uses instrument kinematics, which must be reconstructed when only video is available. We introduce a kinematic reconstruction network for estimating instrument position, orientation and jaw angle from monocular video. Our visual representation combines global attention pooling of frozen DINOv3 features with local pooling at instrument landmarks from fine-tuned SAM 3.1 masks. Our shared Transformer encoder and temporal convolutional heads integrate this representation with mask geometry, monocular depth and visual state estimates from arm-specific multilayer regression networks. Our position branch predicts displacement magnitude and direction separately to preserve traveled distance. We fit trajectories to predicted state observations and motion increments by differentiable weighted least squares, expressing quaternion observations relative to cumulative predicted rotations to obtain a quadratic orientation objective. We evaluate reconstruction across 2,802 Open-H episodes. Compared with LiveMAE on the main Open-H benchmark, our method reduces path-length mean absolute error from 0.45 to 0.34\,cm and increases temporal mean average precision for motion segmentation from 44.54% to 54.44%.

Probabilistic question-answering systems – whether large language models (LLMs) themselves, retrieval-augmented generation (RAG), or trained multi-hop retrievers – conflate "what is known" and "how to reason" into a single probabilistic computation: hallucination cannot be eradicated, evidence chains cannot be audited, and the system answers even when it does not know. We present LatWeave, which organizes knowledge into a multidimensional knowledge lattice and compiles multi-hop QA into three deterministic operators – meet (constraint intersection), compare (lattice-order comparison), and abstain (structural abstention); LLMs appear only on the construction side (one-shot extraction) and the query-planning side, while the answer-generation path is zero-LLM, zero-task-training, and auditable end to end – so that question answering over Web-published knowledge becomes reproducible item by item. Rather than claiming across-the-board SOTA, we characterize the operating envelope of this paradigm on six public benchmarks: when knowledge is complete (MetaQA, 39,093 questions) meet chains are near-lossless over three hops (any-hit 0.9975, on par with fully supervised KBQA); on templated multi-hop home ground (2WikiMultihopQA held-out n=1,258) EM 0.865, well above published structure-augmented RAG reproductions; on open-text deep composition (MuSiQue) and extraction-coverage gaps (HotpotQA) we report degradation honestly and attribute it to causes outside the lattice-algebra layer; and when information is incomplete (IIRC) we achieve structural abstention with abstain accuracy 0.971 and leak rate 0.029. Within the operating envelope, deterministic execution pays no performance penalty, and every step on the answer path can be recomputed – precisely the source of end-to-end auditability.

Converter terminal characteristics are central to both dynamic interactions with the network and steady-state power sharing in inverter-based microgrids. Conventional additive virtual impedance (VI) shapes these characteristics through a single virtual branch. This paper proposes virtual two-port control, which uses four coordinated transfer channels to reconstruct the terminal impedance. The connected network consequently observes the original converter through a virtually inserted twoport interface, with conventional virtual impedance recovered as a degenerate case. The proposed control transforms the original impedance through a matrix linear-fractional map. We also derive a necessary and sufficient condition for the reconstruction to be stable and proper. Among its various potential applications in microgrids, we develop three representative ones in detail: passivation with reduced control effort, uncertainty compression, and seriesshunt power-flow regulation. Simulations of VSG-controlled converters demonstrate the advantages of the proposed method over conventional VI in these applications.