Searcharxiv⌕ Search

arXiv subjects

Bo Zhang

Publications and source records attributed to Bo Zhang.

At least 19 recordsLinked to original sources
arXiv2026

FreeForm Harness: A Mobile-Native Agent Harness for Open and Composable Extensions

Mobile phones support many everyday agent tasks, but current mobile agents bundle capabilities into monolithic pipelines with no open recombination point. Users and third parties cannot publish, install, and compose new behaviors, leaving personalized daily-life requests underserved. We argue that the missing piece is a substrate: although phones are the Web's dominant access point, open publication and composition under decentralized authorship stop at the agent harness. We present FreeForm Harness (FFH), a mobile-native agent harness for open, composable extensions. A frozen base and a permanence-aware required layer handle mobile constraints, including location, proactive triggering, and scarce channels. A uniform plugin contract makes independently authored extensions composable, checkable, and replaceable over a frozen socket set. FFH also supports end-user authorship through a tiered commissioned-authoring path. When no existing extension matches a user's intent, a Web coding agent generates a candidate artifact, which is validated and admitted on device under the same contract as other extensions. An Android prototype spans device, market, and cloud. Evaluation shows that architectural boundaries are enforced at runtime; realistic daily-life requests fit the declared composition space without widening it; and user intent reaches on-device execution through retrieval or commissioned authoring, with final trust decisions remaining device-local. FFH advances the agentic Web from "everything is a plugin" toward "everyone is a developer" on mobile.

cs.SE↗
arXiv2026

OneTrans-GSR: A Generative Slate Ranking Paradigm Unifying Reranking and Mix-Ranking with One Autoregressive Transformer

Industrial recommender systems allocate substantial computation to pointwise candidate scoring, yet separate reranking and mix-ranking stages still assemble the final exposure slate using hand-crafted rules and specialized low-capacity models. Mix-ranking combines product cards, short videos, livestreams, and advertising cards supplied by distinct upstream business channels. Although these rules and models encode valuable domain knowledge, they are inherently limited in leveraging additional data and computation. They constrain the combinatorial space of candidate slates and determine objective trade-offs through manual tuning. To enable scalable and effective slate generation, we reformulate reranking and mix-ranking as a unified learning problem whose performance can improve with computation. Specifically, we instantiate this formulation across the reranking and mix-ranking stages and propose OneTrans-GSR, a single decoder-only policy that extends our deployed OneTrans backbone to generate the exposure slate end-to-end. During training, a reward model evaluates complete slates and guides reinforcement learning toward slate-level and long-term objectives. During decoding, business constraints are explicitly enforced. Offline evaluation shows that slate quality consistently improves as reward-model and policy compute increase. Deployed in a large-scale industrial recommendation system, OneTrans-GSR mixes heterogeneous content types within a unified slate. In online A/B tests, it improves GMV/u by 2.63%, advertiser value by 2.06%, and repurchase rate by 2.03%, while reducing end-to-end latency across the reranking and mix-ranking stages by 13%. These results mark a paradigm shift in slate generation: moving beyond human-engineered heuristics toward a general learned policy creates a scalable path for further algorithmic optimization and compute-driven improvement.

cs.IR↗
arXiv2026

Zero Variance Portfolio

When the number of assets is larger than the sample size, the minimum variance portfolio interpolates the training data, delivering pathological zero in-sample variance. We show that if the weights of the zero variance portfolio are learned by a novel ``Ridgelet'' estimator, in a new test data this portfolio enjoys out-of-sample generalizability. It exhibits the double descent phenomenon and can achieve optimal risk in the overparametrized regime when the number of assets dominates the sample size. In contrast, a ``Ridgeless'' estimator which invokes the pseudoinverse fails in-sample interpolation and diverges away from out-of-sample optimality. Extensive simulations and empirical studies demonstrate that the Ridgelet method performs competitively in high-dimensional portfolio optimization.

stat.ME↗
arXiv2026

RoboSeed: Multidimensional Evaluation of Vision-Language-Action and World-Action Models: Capability, Resource Affordability, and Research Evidence

Choosing a robot action model requires evidence about task capability, adaptation cost, deployment requirements, and the reliability of reported conclusions. Success rates alone offer limited guidance to researchers with different computing budgets and goals. We propose RoboSeed, a multidimensional evaluation framework for vision-language-action (VLA) and world-action models (WAM), within a continuing program to train, evaluate, and openly release robot models with verifiable evidence for model selection. The framework defines five dimensions: capability, downstream training affordability, deployment affordability, locally verified research value, and reproduction evidence. It maps empirical measurements and audit criteria to 0-100 profiles using fixed normalization anchors, with balanced, training-constrained, and deployment-constrained preference settings. Interface and protocol checks establish when comparisons are appropriate. Missing measurements remain explicit, incomplete aggregate scores are withheld, and weighted input coverage identifies evidence gaps. We apply the framework to representative VLA and WAM models, drawing on local training, released-weight evaluation, task-level audits, and an evaluation-system study. The results reveal budget-dependent backbone choices, sensitivity to adaptation recipes, and differences between language and visual task interfaces. Reproducible scoring code and machine-readable evidence support extension to additional models. The study provides dimension-level assessments and priorities for further measurement, while incomplete serving traces and heterogeneous protocols preclude overall numerical rankings.

cs.RO↗
arXiv2026

Analytic Model of Trans-axial Sensitivity in Cylindrical PET Systems Based on Solid Angle

In positron emission tomography (PET), a clear theoretical model describing how system sensitivity varies as a source is moved trans-axially within the field of view (FOV) is lacking. The current understanding and practical intuition often suggest that sensitivity is maximum at the center of the FOV, an assumption reflected in standardized protocols. In this work, we derive an analytic model for the trans-axial-plane sensitivity distribution in a cylindrical PET scanner based on solid angle. The model, formulated as a function of trans-axial offset from the center, is validated through both Monte Carlo simulations and physical experiments on a representative system. We find that the derived theoretical distribution is essentially consistent with simulation and experimental results, revealing a non-intuitive feature: sensitivity increases with trans-axial offset, peaks at the edge of the FOV, and drops off sharply beyond it. This study provides the first closed-form model of trans-axial geometric sensitivity in cylindrical PET scanners, offering a vital benchmark for isolating detector technology improvements and revealing a non-intuitive, offset-dependent sensitivity profile that enables new protocol optimization strategies.

physics.med-ph↗
arXiv2026

AI Cosmos: An Agentic Infra for Situated Learning

Large language models (LLMs) can improve task performance during learning, yet these gains do not necessarily reflect learners' ability to thereafter apply the knowledge and skills independently. We present AI Cosmos, an agentic infrastructure for situated learning that organizes learning around two complementary principles: scenario-based participation and cumulative assets of practice. AI Cosmos organizes learners and role-specialized agents within learning scenarios that integrate task-relevant knowledge and, when needed, executable workbenches. An asset-memory loop preserves the processes and outcomes of practice as reusable assets with explicit provenance and dependencies. We evaluate scenario-based interaction through two preregistered randomized controlled studies examining the acquisition of legal reasoning and programming competence, holding the same underlying LLM across conditions. AI Cosmos achieves greater learning gains than a conventional LLM in legal reasoning and higher mastery of unseen programming problems than an LLM-as-Teacher baseline.

cs.CL↗
arXiv2026

STAPO: Stabilizing Reinforcement Learning for LLMs by Silencing Rare Spurious Tokens

Reinforcement Learning (RL) has significantly improved large language model reasoning, but existing RL fine-tuning methods rely heavily on heuristic techniques such as entropy regularization and reweighting to maintain stability. In practice, they often suffer from late-stage performance collapse, leading to degraded reasoning quality and unstable training. We identify a key factor behind this instability: a small fraction of tokens, termed spurious tokens (around 0.01%), which contribute little to the reasoning outcome but receive disproportionately amplified gradient updates due to inheriting the full sequence-level reward. We present a unified framework for evaluating token-level optimization impacts across spurious risk, gradient norms, and entropy changes. Building on the analysis of token characteristics that severely disrupt optimization, we propose the Silencing Spurious Tokens (S2T) mechanism to efficiently suppress their gradient perturbations. Incorporating this mechanism into a group-based objective, we propose Spurious-Token-Aware Policy Optimization (STAPO), which promotes stable and effective large-scale model refinement. Across six mathematical benchmarks and three model scales, STAPO consistently demonstrates superior entropy stability and achieves average accuracy improvements over baselines including GRPO, 20-Entropy, and JustRL under two widely-adopted evaluation settings.

cs.CL↗
arXiv2026

Past, Future, All at Once: Mitigating Stability-Plasticity Dilemma via Post-hoc JANUS Rectification

Fine-tuning foundation models on new tasks inevitably suffer from catastrophic forgetting. While existing works attempt to mitigate this on the basis of parameter-efficient fine-tuning methods, they adopted an overly restrictive Subspace Orthogonality condition. In this paper, we introduce a purely post-hoc and tuning-agnostic weight rectification framework that achieves Parameter Space Orthogonality, which is the necessary and sufficient condition for preserving historical performance to the first order. By projecting parameter updates into the JAcobian NUll Space (JANUS), our method significantly recovers compromised historical knowledge without interfering with the underlying fine-tuning process. To overcome the local validity of the Jacobian approximation, we further propose a Multi-step Adaptive Rectification mechanism that utilizes the JANUS shift to dynamically verify the valid trust region and adjust step sizes. Coupled with our proposed ghost projection, ghost orientation comparison, and sequence-level singular value decomposition compression techniques, JANUS also achieves great temporal and spatial efficiency. Experiments demonstrate that JANUS seamlessly integrates with various fine-tuning methods, significantly mitigating the stability-plasticity dilemma by recovering historical knowledge while preserving downstream task adaptation.

cs.LG↗
arXiv2026

HoloBoxing: A Simple Recipe for Humanoid Boxing with 7 Minutes of Single-GPU Training

HoloBoxing presents a simple recipe for learning humanoid boxing by building on a pretrained whole-body controller. The recipe separates reusable motor skills from task decisions: a compact planner requests future body-point targets, while the frozen controller handles joint-level execution. A learned decoder maps 32-dimensional actions to structured target windows, and measured-state imitation initializes the planner before reinforcement learning. Task learning then refines interaction behavior while reusing the decoder and controller, avoiding the need to relearn whole-body control. This combination keeps task optimization focused on a compact upper policy and supplies an initial behavior before reward-driven exploration. Starting from that available controller, the recipe accounts for about seven minutes of additional training, including reused representation fitting and a single-GPU task run. This excludes controller pretraining, data preparation, evaluation, and search. Simulated boxing experiments and component controls evaluate the resulting behavior and the roles of imitation initialization and the target representation.

cs.RO↗
arXiv2026

Voevodsky's motives and zeroth slice of motivic sphere

Let $S$ be a noetherian scheme of finite Krull dimension and let $R$ be a commutative ring in which every positive residue characteristic of $S$ is invertible. We prove that Voevodsky's stable Nisnevich category of motives with $R$-coefficients is canonically equivalent to modules over the zeroth slice of the motivic sphere with derived $R$-coefficients. The comparison identifies the corresponding commutative coefficient algebras and is compatible with arbitrary pullback. The theorem applies to singular and nonreduced schemes, including mixed-characteristic bases, and requires no excellence or finite-type hypothesis. Its special cases include rational coefficients over arbitrary bases, arbitrary coefficients over $\mathbb Q$-schemes, and $\mathbb Z[1/p]$-algebra coefficients over $\mathbb Z_{(p)}$-schemes. The proof establishes base change for finite-cycle spectra using permutation quotients, families of cycles, and arithmetic power homotopies, then passes to general bases by universal homeomorphism invariance and continuity.

math.AG↗
arXiv2026

RoboDuo: Attempt, Check and Atomic Recovery through Code for Vision-Language-Action Models

Vision-Language-Action (VLA) models map a language instruction and the current observation directly to actions, without checking whether the instruction has been completed. For long-horizon tasks, a single failed step therefore goes unnoticed and leaves the whole task unfinished. Such failed episodes, however, often end with the task partially complete. Moreover, the unfinished steps are reusable atomic skills that recur across demonstrations. A failed attempt can thus be recovered by composing with atomic skills. We propose RoboDuo, a closed-loop dual-system framework that uses code as the interface between a coding agent and two VLAs. The agent writes a program that lets a full-task VLA attempt the whole task and checks whether it is complete; on failure, an atomic VLA finishes the remaining steps. Since the agent emits no actions and only decides which VLA for execution, RoboDuo wraps an existing full-task VLA without updating its weights. To train the atomic VLA and the coding agent with the completion-checking ability, we build LIBERO-Atomic, LIBERO-Code, and LIBERO-Judge from a single set of LIBERO demonstrations. RoboDuo consistently improves X-VLA and $π_{0.5}$ on LIBERO and LIBERO-PRO, with the largest gains on long-horizon tasks, and recovers failed rollouts of a frozen full-task VLA on a real robot.

cs.RO↗
arXiv2026

EXE: Egocentric Cross-Embodiment Pretraining via Factorized Latent Actions for VLA Models

Vision-Language-Action (VLA) models are bottlenecked by the scarcity of robot data. Egocentric human manipulation videos offer an abundant, diverse, and semantically-rich alternative that inherently captures manipulation intent. However, the embodiment gap makes it difficult to extract action-relevant representations from these videos. From the perspective of action execution, human hands and robots share analogous manipulation primitives---reaching, grasping, placing---yet exhibit distinct kinematics tied to their embodiments. This might motivate our key approach: egocentric video pretraining can yield shared action representations capturing transferable manipulation skills, while human-specific motion patterns should be disentangled to avoid interfering with robot policy learning. To this end, we propose EXE, an Egocentric Cross(X)-Embodiment pretraining framework that learns a factorized latent action representation from egocentric human videos, separating shared manipulation priors from embodiment-specific signals. By leveraging the shared action space, the robot policy acquires generalizable manipulation semantics from diverse human demonstrations without being biased by human-specific kinematics. This shared knowledge transfers across embodiments with different kinematic structures, enabling the same pretraining to benefit both parallel grippers and dexterous hands. EXE yields consistent gains over baseline without pretraining on RoboTwin 2.0, RoboCasa GR1, and real-world manipulation tasks.

cs.RO↗
arXiv2026

Étale motives with and without transfers

We compare stable étale motives defined with and without finite transfers. For every noetherian scheme $S$ of finite Krull dimension and every commutative ring $R$, the canonical functor $\mathrm{DA}_{\mathrm{\acute{e}t}}(S,R)\to\mathrm{DM}_{\mathrm{\acute{e}t}}(S,R)$ is an equivalence of symmetric monoidal stable $\infty$-categories. No excellence, geometric unibranchness, or restriction on the residue characteristics is imposed on $S$, and no flatness or finiteness condition is imposed on $R$. The rational comparison is proved by relating divided effective-cycle monoids on split tori to geometric symmetric powers. For wild permutation actions, coordinate power maps and invariant symmetric polynomials give a vanishing theorem after passage to arithmetic neighbourhoods and sequential telescopes. A homotopy of roots-of-unity cycles removes these telescopes on the divided cycle spaces. Seminormal descent and weak continuity extend the comparison to general bases. Bachmann's rigidity theorem and the rigidity theorem with transfers then give the comparison with finite-field coefficients. Compatibility of the forgetful right adjoint with rationalization allows these comparisons to be assembled integrally, after which extension of coefficients gives the result for $R$. As a consequence, rational motivic cohomology defined by finite correspondences is the rational plus sphere and commutes with arbitrary base change in this class of schemes.

math.AG↗
arXiv2026

When Agents Fail: Model or Harness--Who's to Blame?

LLM agents combine models with execution harnesses, so agentic benchmark performance reflects their joint behavior. We argue that harnesses should therefore be evaluated through their interactions with underlying models. We develop a framework combining five complementary measures: (i) Valid Completion Rate (VCR), how reliably agents finish normally and deliver evaluable results, irrespective of task correctness; (ii) Model-Harness Fit, how well a harness supports task achievement through valid completion for a given model; (iii) Cross-Model Fit, a harness's overall suitability across evaluated models; (iv) Interaction Residuals, pairing-specific execution strengths and weaknesses hidden by component-level summaries; and (v) Token Load, the token demands of agent execution, including failed attempts. We evaluate four general-purpose harnesses (OpenCode, Codex CLI, Pi CLI, and DeepAgents) with four core models, plus two fixed native model-harness systems, on six benchmarks covering context management, tool use, long-context planning, multimodal interaction, and artifact handoff. Including native references and stress conditions, 111 evaluation settings yield 52,168 trajectories. Harness rankings vary across models and benchmarks; task performance and valid completion can favor different pairings, and similar task scores can accompany distinct execution profiles. OpenCode ranks first in cross-model performance, while DeepSeek V4 Flash Preview maintains consistently high Mean VCR across harnesses on the shared four-benchmark panel. These findings support evaluating model-harness interactions, training transferable harness-use capabilities in models, and harness designs adapted to the model and task.

cs.AI↗
arXiv2026

Sci-CoE: Co-evolving Scientific Reasoning LLMs via Geometric Consensus with Sparse Supervision

Large language models (LLMs) have demonstrated exceptional reasoning capabilities, and co-evolving paradigms have shown promising results in domains with verifiable reward signals, such as code and math. However, in unverifiable domains, such as scientific reasoning, these methods remain fragile due to unreliable solution evaluation and limited diversity in verification strategies. In this work, we propose Sci-CoE, a two-stage co-evolving framework for domains without verifiable feedback. The framework enables models to self-evolve as both solvers and verifiers by transitioning from sparse supervision to unsupervised learning. In the first stage, the model uses a small set of annotated data to establish fundamental correctness judgment anchors for the Verifier. In the second stage, we introduce a geometric reward mechanism that jointly considers consensus, reliability, and diversity, driving large-scale self-iteration on unlabeled data. Experiments on several general scientific benchmarks demonstrate that Sci-CoE enhances complex reasoning capabilities and exhibits strong scalability, facilitating the construction of more robust and diverse evaluation systems. Codes are available at https://github.com/InternScience/Sci-CoE.

cs.AI↗
arXiv2026

MLEvolve: A Self-Evolving Framework for Automated Machine Learning Algorithm Discovery

Large language model (LLM) agents are increasingly applied to long-horizon tasks such as scientific discovery and machine learning engineering (MLE), where sustained self-evolution becomes a key capability. However, existing MLE agents suffer from isolated exploration, inefficient experience reuse, and lack of hierarchical control, which together hinder long-horizon optimization. We present MLEvolve, an LLM-based self-evolving multi-agent framework for end-to-end machine learning algorithm discovery. By extending tree search to Progressive MCGS, MLEvolve enables cross-branch information flow through graph-based reference edges and gradually shifts the search from broad exploration to focused exploitation with an entropy-inspired progressive schedule. To allow the agent to evolve with accumulated experience, we introduce Retrospective Memory, which combines a cold-start domain knowledge base with a dynamic global memory for task-specific experience retrieval and reuse. For stable long-horizon iteration, we further decouple strategic planning from code generation with adaptive coding modes. Evaluation on MLE-Bench shows that MLEvolve achieves state-of-the-art performance across multiple dimensions including average medal rate and valid submission rate under a 12-hour budget (half the standard runtime). Moreover, MLEvolve also outperforms specialized algorithm discovery methods including AlphaEvolve on mathematical algorithm optimization tasks, demonstrating strong cross-domain generalization. Code is available at https://github.com/InternScience/MLEvolve.

cs.AI↗
arXiv2026

Liouville Theorems and Sharp Homogeneity Defects for Weighted Potential Systems

In this paper, we consider \[ \begin{cases} u=K*(|x|^αa_0(x)v^p)&\quad\hbox{in }\mathbb R^n,\\ v=K*(|x|^βb_0(x)u^q)&\quad\hbox{in }\mathbb R^n, \end{cases} \] where $n\ge2$, $0<ρ 0$, and $α,β>-ρ$. The radial kernel is nonincreasing and satisfies $K(r)\asymp r^{ρ-n}$. The coefficients satisfy $0 n-ρ+\varepsilon_K, \] where the defects bound the radial logarithmic derivatives. No symmetry, finite-energy condition, or prescribed decay is assumed. The proof combines a localized virial identity with a source-layer pressure estimate whose constants are uniform under centered dilations, including the low-order range $0<ρ\le1$. We also identify the necessary scaling range for existence. For radial weighted Newton systems with unequal powers, we prove that the combined coefficient defect has a sharp, unattained infimum equal to the subcritical gap. A fixed bound on the coefficient ratios forces a strictly larger defect, whereas suitable coefficient--solution pairs approach the threshold with excess at most $C/\log M$. In the scalar Newton case, the threshold is explicit and the approximating family converges, after normalization, to an explicit critical weighted profile. The sharpness results concern the Newton kernel; no corresponding construction is asserted for general fractional kernels.

math.AP↗
arXiv2026

Selective strain tuning of nonequilibrium multiferroic dynamics in BiFeO$_\text3$

Quantum materials are characterized by intertwined orders, and a fundamental goal in condensed matter physics is to achieve their selective control in such a way that one order parameter can be tuned while leaving others largely unaffected. Although significant progress has been made in thermal equilibrium, realizing such control out of equilibrium remains highly challenging. Here we employed in-situ tensile strain to selectively manipulate the photoinduced lattice dynamics associated with ferroelectric and magnetic orders in multiferroic BiFeO$_3$. By applying MeV ultrafast electron diffraction to freestanding BiFeO$_3$ membranes under tunable strain, we showed that tensile strain markedly suppresses the ultrafast photoinduced reduction of the ferroelectric displacement. By contrast, the photoinduced dynamics of the antiferrodistortive rotation of the oxygen octahedra, which modulates the magnetic order, remains insensitive to strain. Not only do these findings reveal distinct microscopic pathways underlying nonequilibrium multiferroic dynamics, they also establish tunable strain as an effective route for engineering ultrafast phase control in correlated materials.

cond-mat.mtrl-sci↗