SEARCH · Search arXiv

Search Search arXiv

Search indexed arXiv papers on artificial intelligence, large language models, computer vision and robotics. Read source abstracts and follow links to arXiv.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,117 records · Page 62Linked to original sources

arXiv2026

Long-Horizon Investment with Delayed Information in a Mean-Reverting Market

We study long-term investment in an Ornstein-Uhlenbeck diffusion model with exponential utility. We solve the finite-horizon problem explicitly and show that information delay qualitatively alters long-horizon behavior. Under full information, the certainty equivalent generated by mean reversion grows quadratically with the investment horizon. With any fixed positive delay, it grows only linearly, at an asymptotic rate that we compute explicitly. We derive an explicit causal representation of the optimal delayed portfolio and show that projecting the full-information strategy onto delayed information can fail dramatically: the resulting exponential loss becomes infinite for sufficiently long horizons. Numerical examples illustrate the economic impact of information delay.

q-fin.MF↗

arXiv2026

V-SHARE: Provable Convergence Guarantees for Advantage-based Algorithms

In reinforcement learning, advantage-based value-learning methods, unlike standard Q-learning, share information across actions by decomposing the action-value function into a state-dependent value component and action-specific advantages. While this mechanism has proven effective in both tabular and deep reinforcement learning, its theoretical understanding remains incomplete. We introduce V-SHARE, an advantage-based value-learning algorithm that combines cross-action value sharing with an importance-weighted Bellman-residual correction. To our knowledge, we provide the first complete theoretical characterization of an advantage-based algorithm, including both asymptotic convergence and non-asymptotic finite-sample guarantees. We identify a regime in which V-SHARE mitigates the rare-action bottleneck of asynchronous Q-learning and uncover a contraction--variance tradeoff between stronger deterministic contraction and increased importance-weighting noise. Motivated by this tradeoff, we develop decreasing residual-preconditioning schedules that combine faster early contraction with lower asymptotic noise. We further develop a two-timescale analysis that removes the single-timescale residual-weight threshold while preserving convergence guarantees. Experiments on discrete Markov decision processes (MDPs) and deep reinforcement-learning benchmarks corroborate our results.

cs.LG↗

arXiv2026

Incentivizing Tool-Use Reinforcement Learning for Post-Arrival Emergency Care Prediction

Large language models (LLMs) can reason over both structured measurements and free-text records, which makes them attractive for emergency department (ED) prediction. However, most existing LLMs still fall short of conventional ML methods like trees on EHR-based prediction tasks. Our controlled analysis traces this gap to two weaknesses: LLMs do not select informative features, and they misjudge what numerical values imply for risk even when given the right features. We propose TIR-EC, a tool-integrated reinforcement learning framework that addresses both. TIR-EC exposes EHR feature groups as tools the LLM can call. It trains the LLM with rewards from a gradient-boosted tree teacher, which provides a reference set of informative tools and a graded risk target. TIR-EC also distills task-specific clinical skills into the model weights through a reverse curriculum that gradually removes the skills from the prompt during RL process. Evaluation on three clinical prediction tasks shows that our 4B agent achieves the highest macro F1 of all compared models, outperforms conventional ML baselines, and is competitive with much larger frontier LLMs. Further analysis shows that TIR-EC selects more useful evidence with fewer tool calls and combines measurements more effectively.

cs.AI↗

arXiv2026

X-Fly: Learning to Reach and Stop in Full-Chunk Aerial Vision-Language Navigation

Aerial vision-language navigation (VLN) agents built on large vision-language models are expensive to query. Executing a complete predicted action chunk before replanning amortizes this cost, reducing model inference from 195 to 42 ms per executed waypoint for our policy. It also moves the states at which the policy is queried away from expert demonstrations, which supervise neither recovery from these states nor termination at the learner's own arrival. On OpenFly test-seen, no-stop failures account for 47.78% of full-chunk behavior cloning (BC) outcomes. We present X-Fly, a staged training framework for full-chunk aerial VLN. Recovery DAgger labels learner-visited decision states with smooth chunks that reconnect to future expert segments and with STOP at the first in-region state. Flow-GRPO then jointly refines the chunked flow controller and a Bernoulli STOP policy from same-task closed-loop outcomes, supplemented by outcome-verified stopping supervision. With eight controls per decision, X-Fly reaches 74.18% SR on test-seen and 51.73% on test-unseen, compared with 16.45% and 18.43% for full-chunk BC and 32.04% and 29.46% for DreamFly.

cs.RO↗

arXiv2026

PET Quantification-Preserving Registration for Longitudinal Whole-Body PSMA PET/CT

We adapt the diffeomorphic coarse-to-fine LapIRN architecture for longitudinal whole-body PSMA PET/CT registration in the Learn2Reg 2026 PSMAReg challenge. The method combines affine pre-registration, multi-modal deformable registration, and test-time instance optimization. We further introduce PET biomarker preservation losses for metabolic tumor volume (MTV) and total lesion glycolysis (TLG), including Jacobian-based constraints, and a per-structure rigidity regularization for skeletal structures. The submitted configuration achieved 81.0% DSC and 6.17 mm HD95, with MTV and TLG errors of 3.3% and 3.3%, respectively, on our evaluation protocol. Code is available at https://github.com/koegl/psmareg.

cs.CV↗

arXiv2026

Full-Duplex Interaction with Speech, Expressions and Gestures

Human face-to-face conversation requires coordinating speech, gestures and expressions across both speaking and listening. We present ELLSA 2, an end-to-end full-duplex dialogue model that jointly optimizes speech interaction and streaming gesture-expression generation. An autoregressive, chunk-wise synthesizer combines pretrained acoustic features with distinct listening and speaking hidden states from the dialogue backbone and speech synthesizer, enabling continuous non-verbal generation even when the model is verbally silent. To translate dialogue understanding into vivid semantic gestures, we introduce contextual semantic calls and template guidance, separating whether to invoke an available semantic gesture from how to realize it. We also establish a curated data and evaluation pipeline for dialog-aware co-speech gesture generation and investigate reference-based multimodal large language models (MLLMs) as automatic judges. Experiments show that ELLSA 2 remains highly competitive with open-source full-duplex speech interaction models and outperforms cascaded baselines across dialog-aware gesture and expression metrics. Semantic template guidance further significantly improves semantic gesture quality. Strong system-level correlations with human judgments support these reference-based scores as useful preliminary metrics for full-duplex and semantic gesture assessment. Together, these results advance dialogue modeling from spoken responses toward coordinated multimodal interaction.

eess.AS↗

arXiv2026

CarryOn: A Dual-Loop Harness for Personalized and Reliable Robot Assistance

While generalist vision-language-action (VLA) models have enabled diverse open-world manipulation, real-world robotic assistance requires two critical capabilities they currently lack: personalization and reliable execution. First, executing an instruction like "make me a sandwich" requires grounding subtask decomposition in persistent user preferences rather than relying on a single generic solution. Second, reliable deployment demands explainable safety assurance that proactively prevents hazards and gracefully recovers from occasional physical failures. To address these gaps, we present CarryOn, a dual-loop harness designed to augment a frozen VLA policy. The planning loop maintains persistent user-level and scene-level context that is continually grounded against new observations and revised through interaction, enabling personalized task selection and execution grounding across conversations and changing scenes. The execution loop actively supervises policy rollouts, shielding actions from predicted hazardous contact. When execution fails, the system invokes reusable control-based recovery skills that restore task progress and are refined through reflection on earlier failures. On a real robot and in simulation, CarryOn dispatches preference-consistent subtasks across users and scenes, reduces the violation rate of an already safety-finetuned VLA from 52.6% to 6.1%, and increases recovery success from 13% to 80% when refined skills are reused.

cs.RO↗

arXiv2026

Curvature-Distortion Numbers of Graphs: Prescribed Lin--Lu--Yau Curvature Lower Bounds

We study an optimization problem for weighted Lin--Lu--Yau curvature. Given a graph $G=(V,E)$ and a prescribed lower curvature bound $k\ge0$, we ask for the smallest multiplicative distortion \[ \Dist(w):=\frac{\sup_{e\in E}w_e}{\inf_{e\in E}w_e} \] of the edge weights needed to make every edge have curvature at least $k$. This leads to the family \[ \DN_{\LLY}^{k}(G) := \inf\bigl\{ \Dist(w): κ_{\LLY}^{w}(e)\ge k \text{ for every }e\in E(G) \bigr\}, \] which extends the nonnegative-curvature invariant introduced in our preceding paper \cite{Xia2026InvariantI}. The edge weights change the transition probabilities, while the combinatorial graph distance remains fixed. Positive prescribed curvature leads to a fundamental distinction from the case $k=0$. If $G$ is a connected locally finite infinite graph and $k>0$, then \[ \DN_{\LLY}^{k}(G)=\infty. \] Thus no finite multiplicative distortion of the edge weights can impose a uniform positive curvature lower bound on such a graph. For a finite tree $T$ with $m$ edges, such weights exist exactly when $k\le2/m$. We prove that the minimum distortion is attained by a weight unique up to scaling, and that the normalized minimizer and its distortion depend continuously and nondecreasingly on $k$. We then extend the finite-tree theory to general connected graphs. As an application, we characterize the graphs of girth at least six for which finite $k$-distortion is possible.

math.CO↗

arXiv2026

An Alternating Minimization Algorithm for Federated Sparse Orthogonal Nonnegative Matrix Factorization

In this paper, we develop an alternating minimization algorithm for federated sparse orthogonal nonnegative matrix factorization (ONMF) with emphases on controlling client drift. To reduce communication frequency, clients perform multiple local updates between communication rounds, biasing updates toward their respective local objectives and giving rise to client drift, particularly under data heterogeneity. In federated sparse ONMF, global nonsmooth regularization further complicates drift control, as proximal mappings generally do not commute with averaging. To address these challenges, we develop an effective optimization algorithm that introduces variable splitting to decouple proximal operator evaluation from communication, allowing clients to handle joint sparsity regularization locally. In addition, the algorithm incorporates a correction term into the local gradients to adjust the update directions and mitigate client drift. We theoretically establish that the proposed algorithm achieves sublinear convergence to a bounded residual error under some mild conditions. Numerical experiments demonstrate favorable convergence behavior on a synthetic dataset and competitive clustering performance on real-world multi-view datasets.

math.OC↗

arXiv2026

Translation Circuit Discovery and Targeted Optimization for Low-Resource Multilingual Translation

Recent multilingual large language models achieve strong translation performance, yet the internal mechanisms responsible for low-resource multilingual translation remain insufficiently understood. In this paper, we study translation circuits in multilingual decoder-only LLMs for Hindi-to-Urdu and Hindi-to-Sindhi (Devanagari). We propose TITAN, a mechanistic framework that combines causal intervention, sparse feature decomposition, and targeted preference optimization to identify and optimize crucial translation components in the transformer. Unlike conventional full-parameter fine-tuning, TITAN restricts optimization to sparse, causally important circuits, enabling scalable and parameter-efficient adaptation while updating fewer than 5% of model parameters. Experimental results show that TITAN achieves BLEU improvements of 33.75% on Hindi-to-Urdu and 45.86% on Hindi-to-Sindhi over the strongest multilingual baselines. Human expert evaluation further confirms substantial gains in adequacy and fluency across both language pairs. Additional analyses reveal that multilingual translation is implemented through sparse, structured circuits operating over a shared semantic subspace rather than through an English-centric pivot representation. Overall, our findings demonstrate that mechanistic circuit discovery can directly improve low-resource multilingual translation while providing interpretable insights into multilingual transformer behavior.

cs.CL↗

arXiv2026

Benchmarking Post-Quantum Key Encapsulation in S3-Compatible Object Storage under Client Concurrency

Data encrypted today with classical public-key algorithms could be decrypted once large quantum computers exist. Cloud object stores keep data for years, so the client-side encryption that protects it must move to post-quantum algorithms such as ML-KEM, which adds a key-encapsulation step and a larger ciphertext to every stored object. So far, the cost of ML-KEM has been measured for the algorithm on its own and for securing network connections, but not for encrypting each object in a storage system that serves many clients at once. It has therefore been unclear whether ML-KEM slows such a system, under which conditions, and why. We address this problem by measuring per-object ML-KEM against classical key exchange on an S3-compatible object store, with two client implementations, two processor types and up to 500 concurrent workers, using pre-registered experiments and statistical equivalence tests. ML-KEM reduced throughput only when the client's processor was at or near full load, with one unexplained exception, and then by 4.0 to 4.9% in our main client. Controlled experiments on one processor type traced most of that loss to the way the software library was called, and a second, native client showed that the size of the loss depends on the implementation. The lasting cost is storage: 1,056 more bytes per object than with classical key exchange. The results cover two clients, one single-node store and one cloud provider.

cs.CR↗

arXiv2026

What Survives Self-Evolution: Provenance Marks in Copied Agent Harnesses

Agent harnesses, the prompt and tool code around a base model, leak. Claude Code's 512,000 lines were mirrored within hours of a packaging error. A plagiarist can pass a leaked harness through a published self-evolver and ship the result as its own. We define this as a covert rewriting attack. No one targets a provenance mark, yet a large language model (LLM) proposer rewrites the file that carries it and a score keeps the better rewrites. We give the first measurement of marks under this attack, with three evolvers, five proposers and two environments. Published marks fail. KGW in the prompt survives GEPA's rewrites in 0 of 12 copies, AgentMark as released in 5 of 12. The proposer's habits decide survival, by two orders of magnitude across models, and selection protects a mark only when each decision sees enough games. We then build a mark for this attack. Its key is the order of near-tied entries of a table of where objects are found, built from public games, so deleting the table costs success, rewrites that keep its rows keep the order, and a customer reads the key from the agent's actions alone. On held-out games it is detected in 62 of 76 copies whose lineage rewrote its file. A scheme-aware plagiarist reshuffles the ties at no measurable cost, and a plagiarist who asks the proposer for more general code removes the key in 75-100% of rewrites.

cs.CR↗

arXiv2026

Can Visual Generation Expand What Unified Multimodal Models Know?

Unified multimodal models (UMMs) receive image-generation supervision that exposes them to visual information not fully documented in text. Can learning to generate images expand what pretrained models know? We introduce Knowledge Transfer Probing (KTP), a controlled protocol that supplies new knowledge through image-generation training and tests its accessibility through text-only question answering. Across seven knowledge categories and three UMMs, we observe direct but uneven transfer from visual supervision to language. Motivated by these observations, we successively introduce Textual Alignment and Interleaved Image--Text Alignment, which connect generation and answering through shared knowledge and interleaved responses, respectively. Guidance on selected entities can improve knowledge access on other entities without QA supervision, with additional gains from interleaved alignment. Further analysis shows that knowledge correctly reproduced in images is often more accessible through text, although substantial gaps remain between the two forms of access. These findings support visual generation as a source of knowledge beyond textual supervision, and motivate joint investigation of training data design and model architecture to make this knowledge more broadly accessible.

cs.CV↗

arXiv2026

Dust at cosmic dawn: unveiling the onset of dust and obscured star formation at $z\gtrsim8$

Although less than 1 percent of the baryonic mass budget of a galaxy, cosmic dust profoundly alters the multi-wavelength appearance of galaxies across cosmic time. The (sub)millimeter wavelength regime is uniquely capable of weighing these dust reservoirs, and revealing the total extent of galaxy evolution that remains hidden behind them. With current facilities, however, doing so at $z > 8$ is largely out of reach. ALMA2040, a next-generation (sub)millimeter interferometer, is set to change this. Through an order-of-magnitude increase in sensitivity compared to ALMA, ALMA2040 is a transformational facility that will be able to routinely detect dust emission from galaxies at $z > 8$ and, through simultaneous multi-band characterization, accurately measure their dust masses, temperatures, and infrared luminosities. Combined with ancillary metallicity measurements -- including from the far-infrared lines detected by ALMA2040 itself -- this will establish the relative roles of supernovae and grain growth in building up dust in the earliest galaxies. Simultaneously, ALMA2040 will measure the fraction of star formation that is obscured by dust at cosmic dawn, and thereby the extent to which rest-UV and optical surveys miss early galaxy build-up. Without a 10$\times$ upgrade to ALMA -- comparable to the leap from the VLT to the ELT -- we will not be able to capitalize on the momentous advances enabled by JWST, and will remain blind to the cold and obscured component of galaxy formation in the $z > 8$ Universe.

astro-ph.IM↗

arXiv2026

Descending Everest with First-Order Oracles: Oracle Complexity of Convex Optimization under $(L_0,L_1)$-Smoothness

For convex $(L_0,L_1)$-smooth functions, the curvature may grow together with the gradient norm. Therefore, even when $\|x^0-x^*\|\le R$, the gradient and the optimality gap at the starting point may be exponentially large in $L_1R$, and the bound $\sqrt{L_0R^2/\varepsilon}$, known for the accelerated method on functions with Lipschitz gradient, no longer describes the complexity of the problem. We determine the oracle complexity of this class for arbitrary deterministic first-order methods: for $L_0R^2/\varepsilon\ge1$, the optimal number of oracle calls $N$ satisfies $1+N=\widetildeΘ(\sqrt{L_0R^2/\varepsilon}+L_1R)$. The upper bound reads $O(\sqrt{L_0R^2/\varepsilon}+(1+L_1R)\log(2+L_1R))$: the method first passes through the large-gradient regime in $O(1+L_1R)$ iterations with an inexact line search, and then the standard accelerated method with a constant stepsize takes over. The lower bound is obtained by means of chains and the orthogonal invariance principle on functions conjugate to strongly convex functions on a polytope. If the initial optimality gap is at most $4\varepsilon$, the optimal bound becomes $1+N=\widetildeΘ(\sqrt{L_0R^2/\varepsilon}+(L_1R)^{2/3})$.

math.OC↗

arXiv2026

Semiclassical Band Splitting in the Infinite Chern Band Model

We study the two lowest Bloch bands of the infinite Chern band model in the semiclassical limit. When the Berry curvature parameter is an odd multiple of $π$, we derive an explicit asymptotic formula for their exponentially small splitting, uniformly over the Brillouin torus. The proof combines Agmon estimates and WKB approximations for a matrix Schrödinger operator whose lowest potential eigenvalue has two wells. Geometric properties of the Agmon distance allow us to compute the leading tunneling coefficients by the method of stationary phase. We provide numerical computations to illustrate the asymptotic formula and support the conjecture that the two bands remain separated for other values of the Berry curvature parameter.

math-ph↗

arXiv2026

Trajectory-Level Speculation with Semantic Verification for Tool-Using Agents

Large language model agents solve tasks through long sequences of tool calls, and the prevailing design has a single strong model generate every step, although many steps are routine enough for a much smaller model. Existing ways of sharing this work keep the strong model as the generator: routers and collaborative agents decide which model handles each query or step, and speculative methods check a small model's drafts one step at a time, usually against the strong model's own action. We introduce SpecAgent, a collaboration protocol that turns the strong model from the author of each step into a reviewer of steps drafted by a weak model. The weak model executes read-only tool calls as it drafts, so that every step is grounded in real observations, and holds the first action that would modify the environment. The strong model then verifies the drafted segment in a single prefill pass, commits the longest prefix it judges semantically valid, and repairs the first rejected step. Because verification replaces autoregressive decoding with parallel prefill, the strong model decodes only short verdicts and occasional repairs rather than every step. On $τ^2$-Bench, SpecAgent comes within one point of the strong model's success rate while roughly halving memory traffic and wall time, and on AutomationBench and VitaBench it recovers 44-75% of the quality gap while lowering both. In an edge-cloud deployment on MCP-Atlas with three cloud models, it closes 53-75% of the weak-strong quality gap while the strong model generates at most half as many tokens. These results suggest that a strong model need not author every step of an agent: reviewing grounded drafts retains most of its quality while cutting both latency and memory traffic.

cs.CL↗

arXiv2026

Poisoning the Neighborhood: A Topology-Aware Topic-Level Attack on Large-Scale HNSW-based RAG Systems

Retrieval-augmented generation (RAG) mitigates hallucinations and knowledge staleness in large language models using external knowledge. However, ingestion from public and third-party sources exposes RAG systems to poisoning attacks. Existing attacks are designed and evaluated under exact retrieval, where poisoned documents participate in global ranking. Deployment-scale systems instead rely on approximate nearest-neighbor (ANN) indexes such as hierarchical navigable small world (HNSW), where limited search budgets, graph-search reachability, and local benign-document competition invalidate this assumption and degrade existing attacks. We present a topology-aware topic-level poisoning attack against large-scale HNSW-RAG that generalizes to unseen topic queries. We organize poisoned documents into a Hub-Satellite structure: Satellites provide competitive entry points across semantic regions, while Hubs strengthen local cohesion and facilitate cascading retrieval. For generation, we propose an entity-claim promotion attack combining evidence anchoring and entity-claim coupling to promote the target entity via a target claim across different query regions. On 2M-document Wikipedia, our attack maintains over 72% retrieval success and reaches 31.29% attack success, substantially outperforming baselines; domain-specific experiments further support its effectiveness. Structural analyses show that the Hub-Satellite topology forms in HNSW, indicating that graph structure contributes to improved poisoned-document retrieval. Existing poisoning defenses cannot mitigate the attack, exposing graph-structural risks in HNSW-RAG systems.

cs.CR↗