Papers

Users of Workday's deployed LLM-based agents often request features which can be addressed by defining named procedures, also known as skills, in the LLM context, effectively augmenting agents' capabilities. However, as an agent's skills library grows in size, so does the agent's operational cost. Progressive disclosure (lazy-loading) of skills as needed may reduce operational costs, but its impact on overall latency and skill-retrieval quality remains unclear. In this report, we investigate the impact empirically and find that progressive disclosure improves skill-retrieval quality but marginally degrades overall latency.

Lax extensions (also called relators or relation liftings) are a categorical notion to reason about functors acting on functions and relations in a compatible way. They play a central role to develop sound proof principles for behavioral equivalence of state-based systems and are also important for establishing contextual equivalence for effectful programs. In this paper, we develop the theory of lax extensions for parametrized functors and monads and consider notions of behavioral preorders, equivalence relations or metrics which can now be modulated by additional parameters. From an operational viewpoint, we replace standard contextual equivalence where we quantify over all possible contexts by a refined notion of equivalence where the user can regulate the allowed contexts via chosen parameters.

Robots that learn from a few demonstrations often require two forms of generalization. Compositional generalization recombines skills to solve new tasks, and skill generalization lets the learned policy behind each skill work in new situations. The two depend on each other, yet information is lost between composition and the skills it calls. Where a skill works is determined by the structure its policy is trained with, while composition sees the skill only through a separate description, such as a name, an instruction, or a symbolic operator, that omits this structure. Our key idea is to use each policy's structural prior as part of the interface between composition and the skill. A structural prior states what a behavior depends on, for example that a grasp depends only on the gripper's pose relative to the object. Built into training, it shapes where the policy generalizes; stated in language, it tells composition where the policy applies. We instantiate this idea in Agent Priors-guided Policy Learning (APPL). A construction agent segments complete demonstrations into reusable skills, proposes several structural priors for each skill, and trains and verifies one policy per prior. A runtime agent then selects among these prior-specific policies and composes them toward new task goals using their interfaces. Across MetaWorld and long-horizon ManiSkill tasks, APPL improves out-of-distribution skill generalization and enables previously unseen skill compositions; ablating the interface information substantially reduces performance. These results support the use of training-time structural assumptions as a bridge between skill learning and skill composition.

Mechanistic interpretability (MI) aims to explain a model's behaviour through analyzing its internal computations; circuit-based explanations aim to isolate these computations with compact subnetworks validated by ablating the rest of the model. We show that circuits validated this way may fail to recover the underlying mechanism of the model's behaviour by closely reproducing its successful decisions while failing to account for most of its errors. Such explanations should account for the model's particular errors as well as its successes. We evaluate this requirement by measuring exact answer agreement separately on model successes and failures, across circuit sizes and ablation settings, on IOI, Docstring, and six model-task settings from the Mechanistic Interpretability Benchmark. We discover that many tested circuits closely replicate correct behaviour while missing most of the model's errors. On indirect object identification (IOI) for GPT-2 small, under mean ablation, the manual circuit and tested automated circuits, including one trained against the model's full output distribution, agree with the model on 97.3-99.5% of prompts it answers correctly but only 11.4-41.7% of errors. An IOI case study shows that lost errors are recoverable by restoring omitted attention-heads which raise error reproduction from 14.2% to 75.1% on a separate held-out set with 0.41 percentage point decrease on correct agreement, exceeding matched random extensions and scalar-biased control. Intervention traces show how omitted computations produce specific wrong answers for a reproducible subset of errors. In all, these findings show circuits can preserve task success without adequately explaining model's failures, and support exact error reproduction as a necessary, but not sufficient, test of circuit-based explanations of model behaviour.

Structured span annotations, such as quantities with their units, uncertainty modifiers, and event classes, are expensive to create and hard to keep trustworthy once language models enter the loop. We present QuanReview, an open-source system for auditing and correcting such annotation layers. QuanReview aligns two annotation streams over the same documents at character level, resolves unambiguous cases by an explicit and logged policy, and routes candidate conflicts to a browser-based adjudication interface where reviewers accept either side, build field-level hybrids, or flag items for re-annotation. A campaign manager assigns documents to multiple annotators with configurable redundancy, computes agreement at document and span level, auto-merges unanimous documents, and exports the corrected layer in the original file format, so that it can replace the original annotation files directly. Applied to a 4,457-record humanitarian benchmark and an LLM extraction stream, the system fully auto-merged 8% of documents, applied automatic policy decisions to a further 1,513 records, and concentrated human attention on 3,131 candidate conflicts, a mean of 5.4 per reviewed document.

With the recent advancements of neural compressors, explicitly incorporating perception constraints into the design of compression schemes has gained significant attention. Traditionally, these perception constraints ensure that the distribution of the reconstruction does not significantly deviate from the distribution of the source, thus attesting to the perceptual quality of the reconstruction. In this work, we uncover several perception constraints that are naturally present in task-aware compression. In particular, we consider a problem where the primary task is reconstruction and the secondary task is classification (i.e., a statistical test). We study this problem at varying levels of domain information available to us and discuss how to utilize the naturally emerging perception constraints to design rate-minimal compression schemes that also maximize the utility of our secondary task. We show that in this setting, if the decision boundaries of the classifier are ill-defined (mismatch) for our source distribution, then matching onto a target distribution enhances our classification accuracy.

We study numerical time integration for the motion of a charged particle in a strong magnetic field, focusing on the regime in which the fast gyration is not resolved by the time step. Standard Boris-type methods are structure preserving and accurate when the gyration is resolved, but their error constants deteriorate with increasing field strength; filtered Boris methods improve this behavior but may suffer from singularities and resonance-induced error blow-up. In this work we consider the case of a constant strong magnetic field and derive a broad family of one-step exponential integrators from the variation-of-constants formula, formulated in terms of matrix-valued filter functions. This framework includes existing filtered Boris variants and permits the construction of schemes with uniformly bounded filters. Using a discrete variation-of-constants representation together with summation-by-parts arguments, we establish error bounds whose constants are independent of the magnetic field strength and characterize the filter conditions required for first- and second-order accuracy. In particular, we show that bounded filters avoid resonance blow-up and allow arbitrary time steps, at the cost of a controlled order reduction near resonant frequencies. Numerical experiments confirm the predicted convergence behavior, the different accuracy of position and velocity components, and the role of the leading error terms across a wide range of step-size-field-strength products.

In reinforcement learning with verifiable rewards (RLVR), imperfect verifiers can reward incorrect responses, creating opportunities for reward hacking. Using gradient flow with a fixed verifier, we characterize the conditions under which reward rises while correctness falls. We then show that the observations available during RLVR are, in general, insufficient to detect or identify accepted errors, or to guarantee their reduction without sacrificing correct responses. To address this limit, we construct a correction using additional feedback about correctness from audits. This correction achieves selective control: at the current policy, it lowers the probability of accepted errors and raises that of correct responses, provided it outweighs the pressure toward errors from verifier reward. Experiments with log linear and neural contextual bandits and with a language model support the analysis and show that selective control under partial auditing reduces accepted errors while increasing correctness.

Of the approximately 4,500 Oracle Bone Inscription (OBI) characters discovered from the Shang dynasty, only about 1,600 have been deciphered. Many computational approaches compare OBI with glyphs from one historical period at a time. However, during the evolution of Chinese characters, significant structural or semantic changes often occur in uncertain dynasties. A single-period reference may be insufficient when relevant forms change substantially between observed eras. Therefore, we propose the Manifold-based Script Evolution Framework (MSEF), a framework that models the evolution series (OBI, Bronze, Seal, Clerical, Regular) of Chinese characters as the continual evolution of a manifold space. MSEF represents each character as an era-specific manifold point and learns continuous inter-era transition rules via Neural Ordinary Differential Equations. Both manifold space and transition dynamics can be trained end-to-end through character evolution pairs across any two eras.

Tool-based image editing (image retouching) is commonly formulated with autoregressive multimodal large language models (MLLMs) that sequentially generate reasoning, tool selections, and parameter values. In this work, we present a novel approach to tool-based image editing by framing the task as a flow matching problem. We introduce FlowTool, a framework that directly models the distribution of high-quality tool parameters conditioned on the input image and user instruction using conditional rectified flow. FlowTool combines a vision-language model backbone for multimodal understanding with a Diffusion Transformer parameter generator that transforms Gaussian noise into an editing plan. We train FlowTool with a two-stage supervised flow-matching curriculum, followed by reward-based post-training. Across MMArt-Bench, FlowTool-Eval, ArtEdit-Bench, and MIT-Adobe5K, FlowTool achieves significantly stronger reference-based performance than specialized MLLM editing agents and proprietary MLLMs, while remaining competitive with proprietary models under reference-free evaluation. Moreover, FlowTool significantly improves inference efficiency, reducing latency by at least $50\times$ while requiring nearly $2\times$ less memory than the compared baselines. These results demonstrate that tool-based image editing can be effectively modeled as conditional generation over structured continuous editing parameters, without autoregressive reasoning.

Music producers search libraries of isolated instrument recordings, called stems, for sounds resembling parts of an existing song. Neural retrieval systems address this by mapping audio to embeddings and ranking library stems by their similarity to the query. The leading method, Contrastive Instrument Retrieval (CIR), encodes the mixture as a single embedding, but it works best when a user specifies the target's instrument family. We introduce Stembed, which encodes a mixture as several slot embeddings representing candidate stems. During training, we construct mixtures from stems of the same song and match their slot embeddings to those of the isolated stems. The slot embeddings from mixtures inherit the stem identities of their assigned solo embedding, enabling a contrastive loss. On mixtures from held out MoisesDB artists, Stembed outperforms a CIR-style baseline when both search the full stem library. Even when predicting the stem count itself without family labels, Stembed exceeds the baseline's family-filtered R@1. Our website demonstrates how users can select a slot by inspecting the tags of its retrieved stems.

Mobile GUI agents operate through a perception–action loop: at each step they screenshot the device, invoke a vision–language model (VLM), and emit an action. It is slow, costly, and brittle, yet most of what it does is navigation—and everyday navigation is static, ordered, and endlessly repeated. We present PhoneCLI, which compiles an app's GUI navigation into callable commands, without any app-internal API, runtime instrumentation, or model training. Offline, PhoneCLI explores a target app from the outside and distills its screens, interactive elements, and navigation edges into a semantically annotated map; each screen yields one deterministic command: a replay sequence that reaches it. Online, the agent selects a command, verifies it before execution, and then executes it deterministically in sub-second time at zero VLM cost; open-ended interaction and every failure of the compiled path fall back to the embedded VLM interpreter, exactly the pure VLM agent, so compilation can only help. On AndroidLab, PhoneCLI improves the task success rate while reducing steps and token consumption, and it transfers to AndroidWorld's official M3A agent with consistent efficiency gains. What PhoneCLI compiles is the app's navigation rather than one run, so it serves new tasks, not only repeated ones.

We study the truthful and fair allocation of indivisible goods to $n$ strategic agents with additive valuations. Babaioff, Feige, and Manaker Morag [FOCS 2026] gave a randomized mechanism that uses only the agents' rankings of the goods, is truthful in expectation (TIE), and guarantees every agent $1/(H_{n-1}+2)=\Theta(1/\log n)$ of her maximin share (MMS) in every realized allocation, where $H_{n-1}$ is the $(n-1)$th harmonic number; this is nearly the best possible with rankings alone. They conjectured that cardinal information allows TIE mechanisms to achieve a constant ex-post MMS guarantee. We confirm this conjecture: our TIE mechanism guarantees every agent at least $1/7$ of her MMS in every realized allocation; moreover, the mechanism is ex-ante envy-free and can be implemented in polynomial time. Our mechanism has two key technical ingredients, both of which may be of independent interest. The first is a truthful fractional allocation rule specifying each agent's probability of receiving each good: it favors each agent on her top $n-1$ goods and reduces her probability of receiving a good for each other agent who also ranks it among her top $n-1$ goods. The second is the balanced edge coloring: we decompose these probabilities into equally likely matchings from agents to high-value goods, those that alone meet an agent's guarantee, and balance these matchings in a fine-grained way without changing any marginal probability, so that every agent who receives no high-value good can obtain sufficient value from the remaining goods without over-allocating any good.

We determine the optimal query complexity of ground-state preparation to trace-distance error $\varepsilon$ when an energy threshold in the spectral gap is known. Let $U_H$ be an $\alpha$-block-encoding of a Hamiltonian with unique ground state $|\psi_0\rangle$, and suppose $|\langle\psi_0|U_I|0\rangle|\ge\gamma$ for a state-preparation oracle $U_I$. The threshold lies at least $\Delta/2$ above the ground-state energy and at least $\Delta/2$ below every excited-state energy. We give two algorithms that prepare a state within trace distance $\varepsilon$ of the ground state. One uses $O((\alpha/\Delta)(\gamma^{-1}+\log(1/\varepsilon)))$ calls to $U_H$ in expectation; the other uses $O((\alpha/(\gamma\Delta))\log(1/\varepsilon))$ calls to $U_H$ in the worst case. We prove a lower bound matching the expected query count; the corresponding worst-case lower bound follows from Somma and de Wolf [SdW26]. The respective bounds on calls to $U_I$ are $O(1/\gamma)$ in expectation and $O(\gamma^{-1}\log(1/\varepsilon))$ in the worst case. On $(N+1)$-dimensional systems, these $U_I$ bounds are also optimal when the expected or worst-case count of $U_H$ calls, respectively, is $o((\alpha/\Delta)\sqrt N)$. Both algorithms use a constant-accuracy spectral filter to construct a purifier, which we then sequentially compose during amplitude amplification to prepare a state with constant overlap with the ground state. The expected-query algorithm repeats the preparation followed by one high-accuracy spectral filter until success. The worst-case algorithm uses filters of increasing accuracy and limits the total number of queries.

Interior-point methods (IPMs) are among the most widely used algorithms for constrained optimization, yet their Newton-based search directions require costly second-order information and large linear-system solves. Learning to optimize offers cheaper updates learned from data, but the singular behavior of logarithmic barriers near constraint boundaries makes IPMs highly sensitive to perturbations, complicating both warm starting and learning reliable updates. We introduce pdLIP, an IPM for smooth nonlinear programs that integrates learned preconditioning with pdProj, an all-shifted primal-dual projected-search IPM. A shared coordinate-wise recurrent network predicts a positive diagonal preconditioner that scales the right-hand side of the reduced Newton system for the primal step, and the remaining slack and multiplier directions are recovered analytically. The learned iterations avoid Hessian evaluations and Newton-system solves, using only first-order and coordinate-wise operations amenable to GPU parallelization. Training is self-supervised, with a loss based on a penalty-barrier merit function and the residual of perturbed optimality conditions, requiring neither target directions nor precomputed solutions. Primal and dual shifts mitigate the barrier's sensitivity to perturbations near constraint boundaries, enabling effective warm starting. Across four classes of 200-dimensional convex and nonconvex constrained problems, pdLIP warm starts reduce pdProj refinement iterations by 63-67% compared with cold starts at the same KKT residual tolerance of $10^{-8}$, with negligible warm-start generation cost relative to the subsequent pdProj solve. Improvements persist on box-constrained QPs with 1000 variables and extend to applications including portfolio optimization, support vector machines, and a nonlinear control example.

Gated Linear Attention (GLA) Transformers advance linear recurrent models through data-dependent gating, but face a core limitation: the fixed-capacity memory matrices across all heads operate at a single temporal resolution, where each token is processed individually, forcing them to simultaneously encode local syntactic patterns and long-range semantic structure, creating a representational bottleneck that gating alone is insufficient to resolve. We introduce Multi-Scale Gated Linear Attention (MS-GLA), which addresses this by distributing attention heads across multiple temporal resolutions. Coarser resolutions pool longer token spans naturally specializing toward long-range dependencies, while finer head groups retain sensitivity to local syntactic structure. A learnable, input-dependent fusion layer dynamically recombines head group outputs at each timestep, expanding effective memory capacity without increasing per-head state size. This multi-resolution decomposition draws on principles from Multi-Scale State-Space Models (MS-SSM), adapting them to the gated linear attention setting. We evaluate MS-GLA on language modeling, recall-intensive tasks, and long-context generalization. Across all settings, MS-GLA consistently achieves higher accuracy and lower perplexity than GLA at matched parameter counts, with up to 18.9% improvement on recall-intensive tasks and 9.5% lower average perplexity on language modeling benchmarks, validating multi-temporal resolution decomposition as a principled and effective extension of Gated Linear Attention.

Large language models (LLMs) reliably perform entity copying, in which a model copies tokens referring to an entity, termed entity tokens, from the prompt into its output to answer a question. Although entity copying is straightforward for most LLMs, existing research does not provide a systematic account of which layers specialize in this fundamental task or how other tokens in the same sequence, termed context tokens, influence the model's ability to copy the entity tokens. To address these questions, we conduct experiments on Qwen3-8B using two novel methods: genie-in-a-bottle, which controls exactly which layers can participate in an entity-copying task, and attention lobotomy, which cuts off specific tokens' attention to entity tokens without affecting the remaining attention distribution. We find that two distinct groups of layers in the second half of the model are both necessary and sufficient for entity copying. Moreover, in addition to the decoding position's attention to entity tokens, context tokens' attention to entity tokens also proves necessary for copying the exact tokens, even though context tokens do not store entity information themselves unless they satisfy particular semantic properties. Our findings establish the critical role of late layers in entity copying under the guidance of context tokens, calling for future work on how models propagate and consume entity information.

The wakeup problem captures a fundamental symmetry-breaking challenge among devices sharing a communication channel. We study the dynamic setting, where packets become active at arbitrary times on a time-slotted multiple access channel. In each slot, a transmission succeeds if and only if exactly one packet transmits; two or more simultaneous transmissions cause a collision. The goal is to obtain a successful transmission quickly. Prior work on wakeup has largely focused on the number of slots until the first success, referred to as the latency. However, a collision may incur substantial additional delay, represented by a per-collision cost $C$. We therefore seek to control both latency and the collision cost of an execution, defined as $C$ times its number of collisions. We design and analyze a randomized algorithm for dynamic wakeup, Lowball, without collision detection or knowledge of the number of packets, $n$. Fix a constant $0<\epsilon\le 1/2$. There is a constant $K>0$ such that, when $C\ge K\lg^{1/\epsilon} n$, Lowball has expected latency $O(C^{1/2+\epsilon}\ln C)$ and expected collision cost $O(\sqrt{C})$. Below this threshold, both expectations are $O(n\log^{\Theta(1/\epsilon)} n)$. These guarantees hold against an adaptive, non-anticipating adversary, and the algorithm succeeds with probability 1. For algorithms in which each packet's transmission probability depends only on $C$ and the packet's local age, with packets activated together using the same probability schedule, we prove that the maximum of expected latency and expected collision cost is $\Omega(\sqrt{C})$.

AI agents run shell commands, edit repositories, send email and move money with their operators' credentials, yet the record of what they did is usually a log the agent, or whoever runs it, can edit. We present Tracekit (open source, MIT, version 1.4.1), which turns each agent run into evidence a third party can check offline. A signer that the agent cannot reach outside dev mode decides every tool call against its own policy before it runs, and signs the call, decision and result as Ed25519 records in a per-run hash chain inside an RFC 6962 Merkle tree whose checkpoints independent witnesses can cosign. A run exports as one .tkb bundle that two independent verifiers (Python, and TypeScript, also as a browser page) check against keys the verifier pins, reporting integrity and assurance separately. Disagreements between capture layers become signed records. Tracekit also records the person's request and what the model said before each call, holds calls that break a limit the request stated, compares what was asked, said and done, and finds which input caused an action by replay. Of 21 experiments: a signed log alone detects 6 of 8 tamper classes and a witness the other two; every insider attack in two suites is caught; the signer round trip is 0.45 ms median; no acknowledged record was lost in 60 kill -9 rounds; reconciliation flags 12 of 12 seeded discrepancies; the verifiers reject all 33 forgeries. Policy rules are a tripwire, not the boundary: they stop 570 of 570 harmful calls in our corpus but 211 of 420 (50.2%) written blind to them; the protection is a signer and credentials the agent cannot reach. On a held-out pool, a failed test claimed as passing is found in 86% of sessions with 5 false findings in 1,812 honest calls, and replay names the planted cause in 6 of 6 confirmed harmful runs with a 7B model.

Real-world capture is heterogeneous: perspective, fisheye, and $360^\circ$ panoramic images can coexist within a single reconstruction task, yet most feed-forward 3D reconstruction models assume perspective imagery and a uniform input representation. Recent models handling several camera types are either informed of the camera type for each view or reconstruct one image pair at a time. No single-pass method reconstructs mixed-camera tuples containing full panoramas from images alone. We present MEOW, a feed-forward system that jointly reconstructs metric pointmaps and camera poses from one N-view tuple mixing perspective, fisheye and full-panorama images, in a single forward pass from images alone: no calibration, distortion parameters, camera-type labels or poses are supplied for any view. Our guiding design philosophy is to treat heterogeneous-camera reconstruction as a data-adaptation problem rather than an architectural redesign. MEOW retains a perspective-pretrained backbone and learns heterogeneous cameras entirely from a procedural data engine, which renders each scene across a continuous manifold of camera models with exact rays and depth, and certifies covisibility for every camera-sampled training tuple. Trained on synthetic tuples only, MEOW transfers zero-shot to real captures: on heterogeneous 2D3DS tuples it achieves 79.9 mAA@30 against 53.8 for Wid3R given the camera type of every view; on our laser-scanned mixed-camera benchmark it registers every four-view mixed tuple with 79.4 AUC@30. The data engine, benchmark, and complete evaluation pipeline will be released.

Population-based optimization methods often use previous search information through successful solutions, parameter adaptation, or operator performance, but they rarely retain the context in which a search behavior succeeded or failed. We introduce Cognitive Memory-Driven Optimization (CMDO), a derivative-free population-based optimizer that represents experience as the relationship between search context, search behavior, and observed outcome. CMDO organizes these experiences across working, episodic, and consolidated memory, retrieves them according to similarity with the current search state, and uses both positive and negative evidence to guide subsequent search. Retrieved experience does not replay previous candidate locations; instead, it selects search recipes that are reconstructed from the current population through exploratory, directed, and local search behaviors with adaptive search geometry. We evaluate CMDO on selected Blackbox Optimization Benchmarking test suite on COCO (BBOB/COCO) and Congress on Evolutionary Computation 2017 (CEC2017) problems against DE, CMA-ES, SHADE, GWO, HHO, and ORCA, and further study its application to seven-parameter photovoltaic model estimation using measured current–voltage data. The results show problem-dependent but competitive optimization performance, including the lowest median error among the compared methods on CEC2017 F10. More importantly, analysis of the search traces shows that context-dependent recall changes the distribution of executed search behaviors, while unsuccessful experiences remain available as negative evidence for later decisions, showing that accumulated experience directly influences subsequent search behavior. These results support the use of explicit context–behavior–outcome memory as an active mechanism for controlling population-based search.

Mobile manipulation extends robot interaction beyond a fixed kinematic workspace by making the reachable region itself controllable. This flexibility introduces two central challenges: spatially grounded perception under continuous ego-motion and coordinated control of heterogeneous arm and base actions. Existing approaches strengthen geometry through explicit 3D representations or predictive world models, and often decouple mobility and manipulation into separate action streams. We argue that effective mobile manipulation requires not only decoupling, but also representations that support efficient cross-stream collaboration. We present MM-ABC, a foundation model built around Seeing, Coordinating, and Imagining Arm-Base Collaboration. MM-ABC combines sparse multi-level VLM features for spatial perception; a training-only future branch that uses world imagination and geometric intent as extra supervision, strengthening perception and manipulation-intent prediction and improving the overall learning signal; and MM-APT, which coordinates separate manipulation and mobility streams through masked joint attention and clean-action x-prediction. In controlled ablations, replacing clean-action prediction with velocity prediction lowers success on RoboCasa365 composite-seen tasks from 32.8% to 29.2%, and removing future supervision or multilevel conditioning causes larger drops. We pretrain MM-ABC on 5,000+ hours of heterogeneous robot data spanning 400K+ episodes, 12 datasets, and 17 embodiments. Experiments cover EBench, RoboCasa365, ManiSkill-HAB, LIBERO, LIBERO-Plus, and real-world mobile manipulation. MM-ABC achieves 44.71% success on EBench, 61.2% on RoboCasa365, 99.1% on LIBERO, 82.8% on LIBERO-Plus without perturbation training, and 83% mean success on five real-world tasks.

Multi-robot trajectory planning is a fundamental problem in multi-robot coordination but remains computationally challenging due to its nonconvex, multimodal, and high-dimensional nature. This work builds upon D4orm, a dynamics-aware diffusion-denoising framework, and develops a family of planning architectures for diverse operational requirements. Unlike conventional numerical optimization methods, D4orm employs sampling-based optimization to generate solution trajectories through massively parallel sampling, leveraging modern computing architectures such as GPUs. Its diffusion-denoising structure iteratively optimizes deformations to candidate control trajectories, providing an efficient and versatile paradigm for generating kinodynamically feasible and conflict-free trajectories. Using D4orm as the building block for advanced planners, we present a decoupled planner for improved scalability, an online receding-horizon planner with feedback control, and a distributed planner for resource-constrained settings. Evaluations with differential-drive and holonomic robots in 2D and 3D environments demonstrate that D4orm-based approaches find high-quality solutions faster and more reliably than other sampling-based optimization methods, such as MPPI, as well as a learned diffusion-model-based method. We further demonstrate zero-shot deployment on ten real quadrotors with obstacles, large-scale deconfliction with 100 simulated robots, and fully onboard distributed `lifelong' operation with six ground robots. Overall, these results establish diffusion denoising as a scalable and reliable framework for multi-robot coordination. Code and video: https://github.com/proroklab/d4orm

Tandem mass spectrum prediction supports compound identification across metabolomics, natural-product discovery, and environmental analysis. However, pretrained predictors often degrade under shifts in chemical space and acquisition conditions, while retraining domain-specific models from scratch is costly. We introduce SPARC, a retrieval-guided test-time specialization framework that adapts a pretrained predictor using a spectral reference library without accessing test-query spectra. For each target query, SPARC retrieves chemically related reference spectra to recalibrate fragment intensities within the learned fragmentation space. During Transfer, SPARC combines reference-guided spectral adaptation with reliability-aware consistency, using reconstruction behavior on retrieved spectra to selectively preserve trustworthy predictions during continual specialization. Across MassSpecGym, NPLIB1 and application-specific GNPS libraries, SPARC improves spectral prediction under multiple transfer settings. These results establish retrieval-guided test-time specialization as a practical strategy for extending pretrained MS/MS predictors to specific chemical and acquisition domains, with continual test-time training providing further refinement during deployment.

We study the problem of learning a sparse quantum state, an $n$-qubit quantum state whose density matrix has at most $s$ nonzero matrix entries in an unknown product basis. While such states admit compact classical descriptions, they can carry long-range entanglement that prevents reconstruction from local reduced density matrices alone. Therefore, previous learning approaches addressed such long-range-entangled states using many entangling gates to extract the necessary information. In this work, we show that sparse states can nevertheless be efficiently learned using only single-qubit measurements. Specifically, when the sparsity $s$ is constant, our algorithm can learn sparse states from single-qubit measurements with polynomial sample complexity and classical computational complexity. When $s$ grows polynomially with $n$, sparse states can still be learned from single-qubit measurements with polynomial sample complexity, although efficient classical computation is not guaranteed in general. In this regime, however, the classical computational complexity becomes quasipolynomial when the state is sparse in an unknown basis that is a product of a known fixed finite set of single-qubit bases (e.g., eigenbases of Pauli operators). These results establish efficient learning of sparse states with long-range entanglement without entangling gates, and the single-qubit measurement requirements make our algorithms compatible with current quantum devices.

While structure-preserving methods for the forward simulation of Hamiltonian systems are well established, the mathematical foundations of their data assimilation and filtering counterparts remain comparatively less developed, particularly regarding rigorous accuracy guarantees. To address this gap, we propose an online filtering algorithm to reconstruct an unknown function $u^\dagger$ from finitely many measurements, assuming that $u^\dagger$ solves a parametric Hamiltonian PDE with unknown inputs. The method uses low-dimensional symplectic approximation spaces that evolve in time according to the PDE model and informed by the measurement data. The method can also be coupled with a dynamical sensor-placement strategy designed to optimize reconstruction stability. We derive error bounds in terms of stability and best approximation, and show that the reconstruction remains in a symplectic space and preserves energy up to the approximation error. Numerical experiments demonstrate the effectiveness of the method and the benefits of dynamic over static sensor placement.

Many language tasks have no single answer that can be checked automatically. Rubrics provide criteria for judging responses to these tasks. For reinforcement learning, the resulting verdicts must be combined into a scalar reward. A common approach sums the points assigned to satisfied criteria. Distinct verdict patterns can thus receive the same reward, and the fixed points encode how much each criterion should count, not how strongly its verdict distinguishes the current rollouts. Beyond this aggregation problem, judging the full rubric needs more judge requests as the criterion count grows. To address these limitations, Rubric Response Theory (RRT) measures quality and selects criteria when rubric criteria are monotone indicators of a shared target. Rather than adding assigned points, RRT uses a two parameter item response model that treats the verdict pattern as evidence about scalar quality specific to the rubric. Under this model, its likelihood score maximizes the local signal-to-noise ratio for quality. Its Response Parameter Network (RPN) reads the prompt and criterion text to predict criterion difficulty and discrimination. As the policy distribution changes during training, RRT uses online expectation maximization to update the RPN from current rollout verdicts. With Qwen3.5-4B as the policy, RRT's macro criterion score across Medical, Science, Rubrics as Rewards Science, and RubricBench is 1.7 points above that of group relative policy optimization (GRPO). On hard and very hard criteria in Medical and Science, RRT gains 2.8 to 5.6 points over GRPO. At half the criterion budget, adaptive Fisher selection with a frozen RPN keeps the macro criterion score across four datasets within 0.1 points of GRPO with full judging. These results show RRT can reduce judge requests while remaining competitive with GRPO.

Code-switching (CS), a seamless alternation between languages within a single utterance, remains a critical challenge in automatic speech recognition (ASR). While prior works focus on conversational CS-ASR, enterprise settings demand evaluation of operational impact beyond edit-distance errors: how code-switching transcription errors propagate to downstream voice agent task failures. In this work, we propose (1) a CS-ASR synthetic benchmark and multidimensional evaluation framework tailored to enterprise domains, (2) systematic evaluation of frontier ASR systems across 5 language pairs, (3) diagnostic analysis of the additional transcription errors that code-switching introduces across language pairs and models. We release COSE-E to support enterprise-focused CSASR evaluation for multilingual voice agents in enterprise deployment.

Network Digital Twins (NDTs) are important enablers of 6G and future networks. However, there is a lack of studies regarding practical aspects, such as the impact of simultaneously changing twinning rate, fidelity, and other NDT operational parameters. For instance, works often consider the impact of operational parameters in isolation or with physical twin (PTwin) implementations relying on simulations. Therefore, the main contribution of this work is to provide an in-depth study on the performance impacts of both twinning rate and fidelity, using PTwins that rely on measurements obtained from hardware. We also investigate the optimization of these two operational parameters to minimize energy consumption, exploring a Bayesian Optimization (BO) method and what-if analysis. In this work, the NDT models an indoor propagation environment to optimize beam management. The virtual twin (VTwin) was implemented with the Sionna ray tracing (RT) simulator and the PTwin with our in-house setup composed of customized Wi-Fi radios with 32 antenna elements operating at 60 GHz. The study reveals important aspects of wireless channel NDTs, suggesting that fidelity levels vary throughout the experiment and with the what-if difficulty. Moreover, the twinning rate can be adjusted using sample-efficient methods, such as BO, even at lower fidelity levels.

Large Language Models can perform multi-step reasoning and improve task performance through different forms of intermediate computation, from token-based traces to computation carried out in latent space. However, a question remains open: do these different forms of thinking rely on the same underlying mechanism? To address this, we train and compare five variants of the same GPTNeoX backbone from scratch on an extended multi-hop reasoning task (ProsQA-Ext): a vanilla model, a Chain-of-Thought (CoT) model, a Pause Token model, and two latent-reasoning models that are optimized end-to-end without intermediate reasoning traces. We find that, strong in-distribution (ID) performance does not guarantee depth generalization. Vanilla, CoT, and Pause Token models solve ID problems well, but rely largely on local graph features and generalize poorly to out-of-distribution (OOD) problems with longer hops. In contrast, latent variants generalize better and show internal dynamics consistent with forward reachability propagation on the graph. Causal interventions and circuit analysis localize this computation to a sparse recurrent search circuit in the bottleneck latent model: an attention head retrieves graph relations, an MLP and the residual stream update the reachability state across recurrent steps, while multiple attention heads together then do the candidate matching. Together, these results show that different thinking mechanisms can learn distinct computational solutions, even at similar ID performance. In this setting, latent recurrence supports a reusable forward-search algorithm that generalizes beyond the training depth.