Statistical Mechanics
Quantum Mpemba effect from Stark localization
Nico Albert, Masudul Haque and Shovan Dutta
In classical systems, rugged potential energy landscapes provide a transparent mechanism for the celebrated Mpemba effect, in which hotter initial states cool down faster. This picture generally does not survive in
quantum systems. Here we show how to design a
quantum energy landscape with local dissipation leading to an Mpemba effect with parametrically separated timescales. Our approach uses Stark localization to design an energy landscape and localized incoherent hopping as cooling mechanism. The hops are triggered by rare "detection" events whose rate grows with energy, allowing hotter states to cool faster and producing super-exponentially separated cooling rates for localized initial states. We further show that the effect is dramatically enhanced by collective hopping of bound pairs in the presence of attractive on-site interactions. These findings have clear experimental signatures accessible in present-day setups.
Read the full abstract ▾
Cosmology and Nongalactic Astrophysics
Multi-Stage and Multi-Field Inflation in Random Inflationary Landscapes
Xingang Chen, Lucas Pinol, Zhong-Zhi Xianyu et al. (+1 more)
Generic potentials in an inflationary landscape are typically too steep to support a prolonged period of
cosmic inflation. Atypical flat regions of the landscape are required to support successful inflation. The occurrence of such regions leads broadly to two possibilities: a single prolonged period of slow-roll inflation (which we call single-stage inflation), or multiple shorter periods of inflation separated by transient departures from slow roll (which we call multi-stage inflation), giving rise to
primordial features and other departures from the behavior of the simplest slow-roll model. The former is the possibility most commonly assumed when deriving the
primordial density fluctuations for the Big Bang model. In this paper, using Gaussian random potentials as a simple model of the inflationary landscape, we study the relative occurrence of these two possibilities. We find a substantial fraction of multi-stage inflationary trajectories in landscapes with field-space dimensions one, two, and three, and this fraction increases with dimensionality at fixed typical landscape
curvature. In constructing the inflationary landscapes and identifying successful inflationary trajectories, we also define several relevant quantities and classify detailed properties of both multi-stage and multi-field inflationary trajectories.
Read the full abstract ▾
All authors Xingang Chen, Lucas Pinol, Zhong-Zhi Xianyu, Yisong Zhang
Distributed Computing
Who Pays for the KV Cache? Attributing Shared AI Inference Spend Across Kubernetes and LLM Provider Bills
Timothy Urista
Organizations pay for AI through disconnected ledgers: Kubernetes allocations for self-hosted inference, gateway logs, and per-
token bills from API providers. We present unalloc, an open-source tool that joins OpenCost, LiteLLM, OpenAI and Anthropic cost data into one exact ledger and reports the share of spend with no owner, and use it to study where attribution breaks at the seams between these systems. Five case studies run inference for real or simulate it: a vLLM-style serving simulator with paged KV
memory and prefix caching; a PyTorch
transformer serving a multi-tenant trace with a real KV cache; tensor- and pipeline-parallel inference on torch.distributed; the unmodified CLI against mock provider APIs; and four downstream use cases. At the seams, in a constructed multi-pod deployment scenario – one month of synthetic OpenCost allocations, not observed billing data – owner labels set only on LeaderWorkerSet leader pods leave 66% of that deployment's GPU bill unowned, and the natural fallback key assigns 61% of it to a Helm chart name while the headline unallocated share falls to 4%; enabling every source double counts all gateway spend; and reading one page of a billing API reports a quarter of spend. Inside a shared inference server the metering rule decides who pays: on an NVIDIA H100 running vLLM, a
token meter assigns a
retrieval-heavy tenant 12-14 percentage points more of the bill than an equal time-share meter at every load tested, while GPU utilization reads 97-99% across configured loads of 2 to 16 requests per second (3.7 to 26.9 completed requests per second; the configured rate counts session-initial arrivals only) and power draw tracks load. Neither meter is a ground truth; we position these results against recent Shapley-based energy attribution. Code, raw data, captured evidence, figures and the paper regenerate from the repository.
Read the full abstract ▾
High Energy Physics (Phenomenology)
Three-Flavoured Thermal Leptogenesis Without Tuning
Angus Spalding
We investigate thermal leptogenesis in an extension of the Type-I seesaw in which a real singlet scalar couples non-diagonally to the right-handed
neutrinos. The interactions $-y_{ij}\phi N_iN_j$, with $i\neq j$, introduce new sources of CP violation in heavy-
neutrino decays with a negligible enhancement of the standard leptonic washout. We derive an analytic upper bound on the scalar-induced CP asymmetry and present numerical results using fully flavoured Boltzmann equations and the density-
matrix formalism. We show that the observed
baryon asymmetry can be generated with a hierarchical right-handed-
neutrino spectrum at the electroweak scale. In the minimal Type-I seesaw, lowering the scale of decay leptogenesis typically requires either a tuned quasi-degeneracy of the right-handed-
neutrino masses or cancellations between the tree-level and one-loop contributions to the light-
neutrino masses, while hierarchical leptogenesis instead requires a high scale that aggravates the
Higgs hierarchy problem. This singlet-scalar extension appears to avoid these tensions, allowing successful leptogenesis with a hierarchical right-handed-
neutrino spectrum down to the electroweak scale.
Read the full abstract ▾
Cosmology and Nongalactic Astrophysics
Tangled $\tau$: Planck CMB Constraints on Flash Reionization
Param Upadhyay, Elisa G. M. Ferreira and Evan McDonough
Recent cosmological tensions have renewed interest in the role of reionization in parameter inference, and flash reionization has been proposed as a transient high-
redshift ionization episode that can increase the Thomson optical depth while remaining compatible with CMB polarization data. In this work, we present the first Markov Chain Monte Carlo analysis of the flash reionization model fit to Planck
cosmic microwave background temperature, polarization, and
lensing data. We perform this analysis using Planck PR4 data in $\Lambda$CDM with flash reionization. We find that CMB data allow a wide range for the
redshift of the flash, $z_{\rm flash}$, but with constraints that are tightly correlated with the peak ionization fraction $x_{\rm flash}$. We find constraints $x_{\rm flash}=0.24 ^{+0.13} _{-0.07}$ and $x_{\rm flash}= 0.15 ^{+0.07} _{ - 0.05}$ for the
benchmark flash reionization scenarios with fixed $z_{\rm flash}=20$ and $25$ respectively. The standard $\Lambda$CDM parameters do not exhibit any significant shifts, and in particular, the total optical depth to reionization is not appreciably changed, with $\tau=0.061 \pm 0.007$ in flash reionization as compared to $\tau=0.059\pm 0.006$ in the standard tanh parametrization. We repeat this analysis for Planck PR3 data in place of PR4, and find that the mild preference for a flash is replaced by 95% CL upper bounds, given by $x_{\rm flash}<0.28$ and $x_{\rm flash}<0.18$ for $z_{\rm flash}=20$ and $25$, respectively, from Planck PR3 data.
Read the full abstract ▾
High Energy Physics (Phenomenology)
Interpreting the LZ 248 keV Event using Dark QCD
Francesco Sannino and Jessica Turner
The LUX-ZEPLIN (LZ) Collaboration has reported a
nuclear-
recoil candidate with reconstructed energy $248\pm23_{\rm stat}\pm23_{\rm sys}\,\mathrm{keV}$. We investigate an interpretation in terms of composite inelastic dark matter, in which a dimension-ten operator splits the lightest
baryon of a confining $SU(3)_{\rm D}$ theory into two Majorana states. Endothermic scattering is mediated by a leptophobic vector coupled to
baryon number. For a dark-
baryon mass of $1\,\mathrm{TeV}$ and a splitting of $300\,\mathrm{keV}$, a reference nucleon cross section of $6.5\times10^{-43}\,\mathrm{cm^2}$ lies within LZ's published two-sided 90% confidence interval. This
benchmark fixes $M_V/g_B\simeq9.9\,\mathrm{TeV}$, where $M_V$ and $g_B$ are the mediator mass and
baryon-number gauge coupling, and is compatible with
solar neutrino and dijet constraints.
Read the full abstract ▾
Astrophysics of Galaxies
EPOCHS-DR2 I: Expanded data release and properties of $6.5<z<16.5$ photometrically selected galaxies across NIRCam deep fields covering $\sim700\,\rm arcmin^2$ with the galfind software
Duncan Austin, Christopher J. Conselice, Thomas Harvey et al. (+14 more)
We present the \epochstwo\ sample of $2452$
star forming
galaxies at $6.5<z_{\rm phot}<16.5$ produced using the \galfind\ photometric pipeline across $\sim650-740~\mathrm{arcmin}^2$ in six of the most widely used deep, blank field \jwst/NIRCam imaging surveys. $293$
galaxies are spectroscopically confirmed and a spectroscopic–photometric
redshift comparison yields a normalized median absolute deviation, $\sigma_{\rm NMAD}=0.017$, an $\eta=4.3\%$ outlier rate, and a mean offset $\langle\Delta z\rangle=-0.022$. We find $34$ little red dots (LRDs) with “V–shaped” SEDs, as well as $69$ L–type, $36$ T–type, and $2$ compact Y–type brown
dwarf candidates, which we remove from our sample. The fiducial sample yields a flattening $\beta_{\rm UV}-M_{\rm UV}$ relation with only moderately blue median $\beta_{\rm UV}(M_{\rm UV}=-19)=-2.32^{+0.08}_{-0.07}$ at $z\simeq12.5$, implying
dust-polluted
stellar populations at the highest
redshifts. A “gold” sample of $62$ sources are identified at $z_{\rm phot}>10.5$, $\sim85\%$ of which are without NIRSpec follow-up, including $15$ candidates at $z_{\rm phot}>13.5$ across a diverse range of proper sizes $30\,\rm pc\lesssim r_{\rm c}\lesssim1.5\,\rm kpc$ and UV slopes $-4.0\lesssim\beta_{\rm UV}\lesssim-1.5$. Two very promising candidates stand out, GS-W-4608 and GN-Med-1881, with $\beta_{\rm UV}\lesssim-3$, $\gtrsim1.6\,\rm mag$ \lya\ break colors, and F335M and F410M detections which improve their high-
redshift solution robustness. Complementary to our high-
redshift sample, this public data release provides the community with optical/NIR catalogs, empirical PSFs, and \eazypy, \bagpipes, and \bdfinder\ SED fitting results, enhancing the legacy value of photometric JWST data in the upcoming era of wide-field imaging
datasets with Euclid and Roman.
Read the full abstract ▾
All authors Duncan Austin, Christopher J. Conselice, Thomas Harvey, Nathan J. Adams, Louis Quilley, Jordan C. J. D'Silva, Vadim Rusakov, Qiong Li, Lewi Westcott, Caio Goolsby, James Arcidiacono, Kai Madgwick, William J. Roper, Brenda Frye, Norman A. Grogin, Rafael Ortiz III, Rogier Windhorst
Human-Computer Interaction
Who Does What in AI Auditing? Designing Human-AI Collaboration for Auditing Generative AI
Eunkyu Park, Markelle Roesti, Wesley Hanwen Deng et al. (+5 more)
AI
auditing increasingly incorporates AI
agents to expand the scale and breadth of audit coverage, yet little is known about how
auditing work should be divided without displacing human judgment. We introduce Human-
Agent Audit Collaboration (HAAC), a workflow and system for structuring human-AI collaboration in AI
auditing. Drawing on prior work and formative consultations with AI
auditing practitioners, HAAC specifies how
agents can support exploration, assessment, reporting, and review while preserving human oversight where contextual judgment is critical. We instantiate HAAC for conversational shopping
agents and evaluate it through two studies. With 71 auditors, AI assistance increased attack success and broadened exploration, while also shaping later attacks and increasing auditors' reliance on AI-generated assessments and reports. Interviews with Responsible AI practitioners showed that actionable audits require visibility into coverage, reproducible attack trajectories, and evaluation of the
auditing agents themselves. Our findings identify design considerations for effective and accountable human-AI
auditing.
Read the full abstract ▾
All authors Eunkyu Park, Markelle Roesti, Wesley Hanwen Deng, Renata Barreto, Mohammad Tahaei, Kenneth Holstein, Jason Hong, Motahhare Eslami
Machine Learning
Critical-State RL: Diagnosing Trainable States for Multi-Turn Tool Use
Zixiang Chen, Wenting Zhao, Zhepeng Cen et al. (+9 more)
Multi-turn tool-use failures can hinge on a single model call, yet reward variation alone does not reveal which call would benefit from training. When rewards depend on later interactions, their variation can reflect downstream randomness rather than differences between the current actions. We introduce Critical-State RL to identify trainable states in multi-turn interactions. Given task-defined candidate calls and local rewards, the method assesses whether each reward captures the action's effect on task success and whether improvement over a reference policy is possible. It then uses nested sampling to separate action-dependent reward variation from continuation noise and optimizes the policy at the selected states using contextual-bandit training. Experiments on the Berkeley Function Calling Leaderboard (BFCL) v4 compare training at diagnostic-selected states with training at alternative states. For missing-function tasks, the diagnostic selects the response after the tool becomes available; for missing-argument tasks, it selects the response before the missing argument is supplied. Training the selected responses improves performance, including about 14 percentage points on the missing-function task, while training the alternatives leaves performance flat or worse. We further apply the recipe across models and tasks, including logged repeat-call avoidance and
memory management.
Read the full abstract ▾
All authors Zixiang Chen, Wenting Zhao, Zhepeng Cen, Akshara Prabhakar, Jielin Qiu, Jianguo Zhang, Zhiwei Liu, Tulika Manoj Awalgaonkar, Liangwei Yang, Shelby Heinecke, Silvio Savarese, Huan Wang
Computer Vision
WorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory
Wangbo Yu, Kunhao Liu, Wenbo Hu et al. (+8 more)
Video world models enable interactive exploration of dynamic environments, yet struggle to respect prior observations over long horizons and across viewpoints. We present WorldCrafter, a
video world model that learns a camera-queryable implicit 3D-aware
memory for this purpose. The key insight is to let the requested viewpoint shape how multi-view evidence is compressed into the
video generator's limited
token budget. Trained jointly with the
video generator, a
memory encoder and pose-conditioned readout module integrate historical observations into a fixed set of target view-specific
tokens before denoising, without explicit depth-based correspondences. By combining this
memory with recent temporal context and few-step
distillation, WorldCrafter enables streaming scene exploration from a single input image or text
prompt. Experiments across static and dynamic scenes show substantial gains in long-horizon consistency and camera-control accuracy while preserving visual quality during minute-scale exploration.
Read the full abstract ▾
All authors Wangbo Yu, Kunhao Liu, Wenbo Hu, Shenghai Yuan, Chaoran Feng, Haiyang Zhou, Yukun Huang, Yiran Wang, Wang Zhao, Yingmin Luo, Ying Shan
Computation and Language
onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction
Lei Yang, Mengyin Liu, Jia Wang et al. (+9 more)
We present onPanda, an interactive tool for efficiently annotating LLM
alignment data and
agent trajectories. onPanda adopts
token-level correction as its core interaction: while reading a model response, the annotator locates the first inappropriate
token and either picks a substitute from the model's candidate
tokens or types the correct text via free-form editing. The system then truncates everything after that position and continues generation from the corrected prefix, repeating this locate-correct-continue loop until a satisfactory response is obtained. This mechanism lets annotators precisely steer model outputs at low cost: a small controlled study suggests that onPanda reduces median annotation time by 52% over manual post-editing. Since the vast majority of
tokens in the final response are generated by the model itself, the resulting data largely preserves the model's sampling distribution and is well suited for constructing on-policy SFT and preference data. Furthermore, the
token-level corrections recorded during annotation provide fine-grained supervision with precise positions and naturally paired positive–negative samples. onPanda also connects to external tools and harnesses, enabling interactive trajectory annotation in realistic environments. In addition, we release Panda-CVL, a
dataset annotated with onPanda, together with a
benchmark for
token-level correction.
Read the full abstract ▾
All authors Lei Yang, Mengyin Liu, Jia Wang, Hangyu Guo, Liang Zhao, Zheng Ge, Kang An, Binxing Jiao, Qi Han, Daxin Jiang, Siqi Shen, Xiangyu Zhang
High Energy Physics (Phenomenology)
A closer look at the LZ 248 keV event through the lens of cosmic-ray boosted dark matter
Bhavesh Chauhan, Soham Sahasrabuddhe and Manibrata Sen
We investigate whether
cosmic ray boosted dark matter (CRDM) can explain the 248 keV
nuclear recoil event observed by the LUX-ZEPLIN (LZ) collaboration. We consider
spin-dependent interactions, mediated by a pseudoscalar, and incorporate relativistic kinematics in CRDM production, propagation, and detection. We find that
spin-dependent interactions can produce sufficient number of
nuclear recoils in the observed region of interest, without populating events at lower energies, which have not been observed. However, for a
benchmark pseudoscalar interaction corresponding to the non-relativistic effective operator $\mathbb{O}_6$, the coupling required to produce the observed event corresponds to an energy scale comparable to the momentum transfer scales, thus making contact interaction approach challenging. Instead, we find that a pseudoscalar mediator with mass $\sim 300\,$MeV and $O(1)$ couplings can provide a promising solution to this unexplained event.
Read the full abstract ▾
Computer Vision
GAE: Learning a Geometry-Native Latent Space for 3D-Consistent World Generation
Jiahao Lu, Minghao Yin, Wenbo Hu et al. (+5 more)
We present a compact geometry-native latent space as a shared foundation for perception and generation. Visual generators can produce photorealistic frames without preserving a consistent 3D scene. We argue that this is not only a modeling problem but also a representation problem: generators typically evolve appearance-centric latents, while perception models recover geometry in a semantically rich space that encodes cross-view structure. Rather than adding geometry as another output, we reparameterize a geometry foundation model's features into a compact latent space for generation. We realize this shift with the geometry-native autoencoder (GAE), whose latent is jointly decodable to appearance, depth, cameras, and point maps. With this state, a standard conditional flow supports diverse generation tasks. In controlled comparisons that hold the generator and training protocol fixed, replacing the latent with GAE improves both visual quality and independently measured 3D coherence: FVD falls by $12.7\%$ and $23.1\%$ on RealEstate10K and DL3DV, and camera-trajectory error is halved on RealEstate10K. Together, these results show that the latent space is central to geometry-consistent generation and can serve as a shared interface between perception and generation.
Read the full abstract ▾
All authors Jiahao Lu, Minghao Yin, Wenbo Hu, Hengyu Liu, Wang Zhao, Sai-Kit Yeung, Ying Shan, Yuan Liu
Cryptography and Security
Residual Community Prototypes Under-Reject Held-Out Malware Families in FCG-MFD
Junru Zhu, Yixin Yang, Xiaoqing Ding et al. (+1 more)
Open-set malware-family recognition must classify known families while rejecting families absent from training. We test whether Louvain-community summaries add rejection information beyond a graph neural network
embedding and dimension-matched generic
topology. The study uses a deduplicated, conflict-audited FCG-MFD corpus, five held-out families, and three optimization seeds. Community features are residualized against generic
topology using known-family training data before nearest-prototype scoring. Residual community does not produce stable held-out-family rejection. Ranking effects reverse across families, the false-positive rate at 95 percent unknown recall worsens for every held-out family, and a validation-fitted threshold rejects only 4.48 percent of unknown samples. Accepted-known macro F1 improves in every family, but with five independent family units the exact two-sided sign-flip p-value is 0.0625, the smallest attainable value. The score remains associated with graph scale, while simple classifier uncertainty performs better on ranking, high-recall rejection, and OSCR. In this GIN/FCG-MFD setting, community-enriched prototypes change known-class geometry without creating a stable unknown margin. Graph open-set evaluations should pair structural features with matched
topology controls, operational thresholds, and held-out-family analysis.
Read the full abstract ▾
All authors Junru Zhu, Yixin Yang, Xiaoqing Ding, Ruoyu Qi
Machine Learning
LoRA-generating hypernetworks for efficient on-device LLM generative personalization
Sean Augenstein, Li Ding, Jihwan Lee et al. (+2 more)
On-device large
language models (`LLMs'), e.g. running on mobile phones, are ripe for improvement via personalization. The limited compute resources of mobile devices impose limits on model scale and thus model quality, making any realizable quality gains highly impactful. At the same time, their personal nature (i.e., the close coupling to a particular user) means that a given on-device LLM tends to be used in similar, predictable patterns over the course of time. This paper presents a novel method for personalizing on-device LLMs. It trains a hypernetwork to map a user's context
tokens to a low-rank adaptation (`LoRA') well-suited to that user. Once the trained common artifacts are deployed to users' devices, each user uses the hypernetwork to synthesize (entirely on device) a personalized LoRA. This approach blends the benefits while avoiding the drawbacks of two existing approaches to LLM customization: in-context learning (`ICL') and parameter-efficient fine-tuning (`PEFT'). Like ICL (and unlike PEFT), the on-device phase of our approach is computationally feasible, requiring only forward passes through neural networks. Like PEFT (and unlike ICL), our approach modifies the `target' base LLM via weights (the LoRA), avoiding negative consequences (e.g. increased latency) associated with extending the input sequence. Our approach is particularly well-suited to the mobile device regime. Apart from the on-device compute and latency benefits mentioned, it also requires minimal additional storage, as internally its architecture partly leverages the same LLM weights as belong to the target LLM to be personalized. We demonstrate the benefits of LoRA-generating hypernetworks on several representative personalization
datasets, comparing against baselines like ICL and PEFT. Of note, our personalization experiments focus on more challenging and less studied long-form text generation tasks.
Read the full abstract ▾
All authors Sean Augenstein, Li Ding, Jihwan Lee, Keith Rush, Andrey Zhmoginov
Accelerator Physics
Bayesian Optimization of The Relativistic Heavy Ion Collider Luminosity via $s^*$ Control
Xiaofeng Gu, Benjamin Coe, Guillaume Robert-Demolaize et al. (+7 more)
Maximizing luminosity at the interaction point (IP) requires the collision location $s_{IP}$ to coincide with the longitudinal position of the minimum beta function, $s^$. Accurate optics measurements and control of $s^$ are therefore essential for luminosity optimization. At the Relativistic Heavy Ion
Collider (RHIC), average horizontal beta-beat measurements between operating IPs are approximately $20%$, with significant variation in measured $s^*$. Precise control of the beam waist position $s^$ is particularly challenging for modern high-energy
colliders with short bunch lengths and large crossing angles. We present an online
Bayesian optimization (BO) application using the GPTune framework to optimize sPHENIX luminosity through $s^$ control at RHIC. GPTune was first validated at the RHIC Electron Beam Ion Source (EBIS), where it achieved up to a $70%$ increase in beam intensity over the baseline, although experienced operators could reach similar performance with longer manual tuning. The framework was subsequently deployed during sPHENIX operations. Using an intensity-normalized Zero-Degree Calorimeter (ZDC) signal as the optimization objective, due to the unavailability of the live sPHENIX MVTX signal, GPTune identified local luminosity maxima, recovered from intentionally degraded $s^$ configurations, and revealed residual horizontal and vertical waist offsets in the interaction region. These results demonstrate the robustness and efficiency of
Bayesian optimization for real-time
collider tuning under noisy, time-varying conditions. The $s^$ control methodology provides a promising tool for precision luminosity optimization and is particularly relevant to next-generation short-bunch
colliders such as the Electron-Ion
Collider (EIC).
Read the full abstract ▾
All authors Xiaofeng Gu, Benjamin Coe, Guillaume Robert-Demolaize, Takeshi Kanesue, Masahiro Okamura, William Fung, Yue Hao, Ji Qiang, Xiaoye S. Li, Yang Liu
Nuclear Theory
Accessing zero-point fluctuations in high-energy nuclear collisions
Jean-Paul Blaizot and Jean-Yves Ollitrault
Ultracentral high-energy collisions between identical
nuclei are very sensitive to the size and shape of the frozen configuration of nucleons at the instant of the collision, an information that is encoded in the ground state wave functions of the colliding
nuclei. The relative standard deviation of the size of the colliding system soon after the collision has recently been inferred from event-by-event fluctuations of the momentum per particle in collisions of $^{208}$Pb
nuclei at the LHC, and since a single collision event produces thousands of particles, this information is very precise. We connect this extracted value to the magnitude of the zero-point fluctuations of the radius of the lead
nucleus. The value that is extracted from the data is compatible with estimates based on
nuclear sum rules, but is overestimated by classical Glauber-type modeling of high-energy collisions. The role of the Pauli principle in suppressing the fluctuations in closed-shell
nuclei is underlined. It leads to a distinctive $A$ dependence that could be checked in collisions of smaller systems.
Read the full abstract ▾
Robotics
DexTacWAM: A Visuo-Tactile World-Action Model for Dexterous Manipulation
Haoran Yuan, Zekai Wang, Boning Shao et al. (+4 more)
Dexterous manipulation depends on contact dynamics that are often only partially observable from
vision. Recent World-Action Models (WAMs) couple predictive
video world modeling with action generation, but remain largely
vision-centric and therefore cannot directly model these contact dynamics. We present DexTacWAM, a visuo-tactile WAM that encodes each fingertip independently, aggregates the resulting features through a finger- and pose-aware tactile compressor, and injects the tactile latent into a
video diffusion world model for joint visuo-tactile world modeling. Across six contact-rich dexterous manipulation tasks on a 22-DoF bimanual platform, DexTacWAM achieves the highest score on every task, averaging 70.6 versus 38.0 for the strongest baseline. Ablations attribute the gain to modeling contact
evolution as part of the predicted world state rather than tactile conditioning alone: removing tactile world modeling reduces the four-task mean from 74.7 to 26.6 while keeping the same tactile features and action expert. After four hours of tactile-encoder adaptation with a frozen pretrained
vision VAE, our continual
vision-to-touch learning extends the pretrained
video model to touch using roughly 100 demonstrations per task without tactile midtraining, while retaining visual prediction quality within 0.5 dB of
vision-only counterparts. The compressor retains 89.4% of pre-fusion contact recall while enabling 2.26x faster training and 1.29x faster inference. Together, these results show that pretrained
video priors can be extended to distributed multi-finger contact dynamics in a data- and compute-efficient manner.
Read the full abstract ▾
All authors Haoran Yuan, Zekai Wang, Boning Shao, Haoran Lu, Trevor Darrell, Ismini Lourentzou, Wei Zhan
Mesoscale and Nanoscale Physics
Collective Coordinate Dynamics of Antiferromagnet: Gravity by Quantum Metric
Kuangyin Deng and Ran Cheng
The collective coordinate dynamics of
antiferromagnetic and ferrimagnetic textures emerges as a geodesic equation governed by the
quantum geometry of the CP$^1$ state associated with the Néel vector, rendering an effective Riemannian
gravity and a Lorentz force in the texture motion. This unveils a physical scenario fundamentally distinct from the Thiele equation widely used for
ferromagnetic textures, which is inertialess without any
gravity analogy. Applying our framework to an optically driven ultrafast
antiferromagnetic domain wall, we can explain the helicity-dependent asymmetric motion observed experimentally and predict a vibrantly breathing wall width.
Read the full abstract ▾
Artificial Intelligence
Harness-Zero: Harness Distillation via Agent-as-Harness
Haoran Ye, Yuxing Lu, Haonan Dong et al. (+2 more)
Agent harnesses, the external systems that mediate model-environment interaction, can substantially improve
agent performance, but their gains remain tied to the harness at deployment. Because the best harness varies across domains, instances, and models, a general-purpose
agent must either settle for a suboptimal shared harness or route among an ever-growing set of specialized ones. We therefore study
agent harness
distillation: using a domain- or instance-optimized harness as training-time guidance and transferring the behaviors it induces into model weights, so that its gains survive under a single fixed target harness. The challenge is that the two harnesses differ in action space and available information, so guidance from the optimized harness cannot serve directly as supervision for the target one. We introduce Harness-Zero, which enables harness
distillation through
agent-as-harness. Guided by the optimized harness, a harnessing
agent corrects student responses before execution in the target harness's action space, turning harness guidance into training demonstrations. Fine-tuning on the resulting trajectories internalizes harness-induced behavior into the model, so the specialized harness can be removed at deployment. Our experiments spanning knowledge work, tool use, and science domains show that: (1) For frontier LLMs using the same evolved harness,
agent-as-harness outperforms code-as-harness. (2) With the specialized harness removed at deployment, Harness-Zero improves the base model's macro-average task success from 23.3% to 44.3%, even exceeding the 41.7% it reaches with that harness still attached. (3) Harness-Zero recovers harness-induced behaviors absent from the base model, with 82.3% average recovery across 28 patterns in the three domains.
Read the full abstract ▾
All authors Haoran Ye, Yuxing Lu, Haonan Dong, Zhaochen Su, Guojie Song
Combinatorics
Extremal subtrees of critical beta-splitting trees
Anna Brandenberger, Byron Chin and Elchanan Mossel
We determine the most and least likely shapes for an instance of the critical beta-splitting tree via a connection to data compression and Huffman's minimum redundancy codes. This allows us to answer combinatorial questions about the distribution of clades posed by Aldous and Janson, stated as problem 7 in arXiv:2303.02529.
Read the full abstract ▾
Machine Learning
RRSI: Regularized Recursive Self-Improvement of Agent Harnesses
Peng Xia, Rujun Han, Zifeng Wang et al. (+11 more)
An LLM
agent's capability is largely magnified by its harness, namely the
prompts, control flow, tooling,
memory, and context management surrounding the frozen backbone model. Recent methods increasingly automate this process by iteratively proposing and selecting component-wise edits of an
agent harness, practically establishing a form of recursive self-improvement (RSI) at the
agent-system level. However, such recursive
evolution may overfit by memorizing the training tasks, showing large in-distribution gains that shrink or even vanish on out-of-distribution
benchmarks. We introduce Regularized Recursive Self-Improvement of
Agent Harnesses (RRSI), which incorporates the principles of regularizations into harness self-improvement by constraining the
evolution candidate proposal and selection. The proposer operates with a temporally annealed budget, limiting how many edits a candidate can bundle, and it encourages unexplored trajectories based on
evolution history. The selector is equipped with a critic and a pruner: the critic screens
benchmark-specific proposals, while the pruner, removes changes that are too small, too expensive, or no longer useful. Together these constraints favor reusable
agent mechanisms over
benchmark-specific ones or even noises. Across eight
benchmarks spanning coding, agentic workspace and engineering design tasks, RRSI gains up to 6.0 points on the split it evolves against and up to 4.3 points on the five out-of-distribution
benchmarks, while producing a harness that runs on 30% fewer policy
tokens than the unregularized
evolution. Code is available at https://github.com/google-research/rrsi and project page is https://regularized-rsi.com/.
Read the full abstract ▾
All authors Peng Xia, Rujun Han, Zifeng Wang, Yanfei Chen, Yufan Zhuang, Yoonho Lee, Chengsong Huang, Han Yu, Zhongying CuiZhu, Yifei Ming, Huaxiu Yao, Burak Gokturk, Tomas Pfister, Chen-Yu Lee
Computation and Language
DolphinBench: Mapping the Pareto Frontier of Agent Memory
Soumil Rathi, Deshraj Yadav and Taranjeet Singh
Agents today often take real-world actions that depend on long-term
memory and context recall over time. However, most current
memory benchmarks are built for a conversational question-answer format, where the question itself signals that some fact must be retrieved, and often which one. Moreover,
benchmarks rarely require anything beyond accuracy from submissions, allowing
memory systems to make unreasonable cost/time tradeoffs to achieve higher scores. We present DolphinBench, a
benchmark that evaluates
memory directly through an
agent's task completion. DolphinBench includes three knowledge-work personas with roughly 500k
tokens of user messages per persona and evaluates
agents on tasks that depend on information from that history. We verify all 200 tasks per persona by running an
agent with and without the relevant history, requiring success with it and failure without it. Finally, we require all evaluations to report total cost and latency alongside accuracy, which enables us to evaluate
agent memory systems holistically. No existing
memory benchmark combines all three. The
dataset and evaluation code are available at https://dolphinbench.ai.
Read the full abstract ▾
Solar and Stellar Astrophysics
Elemental Abundances of Cool Stars: A Spectroscopic Framework Applied to Five Planet-Host Stars
Nicole Gromek, Drew Weisserman, Ryan Cloutier et al. (+2 more)
Stellar elemental abundances provide important constraints on the chemical environments from which
planets form. M and late-K
dwarfs are
prime targets for the characterization of small
exoplanets, yet their complex spectra make accurate elemental abundance measurements challenging. In this work, we present a spectral synthesis-based framework to derive detailed abundances for cool
stars from high-resolution, near-infrared spectra. We apply our framework to SPIRou YJHK spectra of five
planet-host
stars–GJ 9827, K2-18, TOI-1201, TOI-1685, and Barnard's
Star–and derive abundances of C, O, Na, Mg, Al, Si, K, Ca, Ti, Cr, Mn, and Fe for all five
stars, plus Sc, V, and Ni for a subset, with typical uncertainties of $\lesssim 0.1$ dex. Our recovered
metallicities and elemental abundances are broadly consistent with expected trends and previous literature analyses of these
stars. We derive
stellar ratios of C/O, Mg/Si, Fe/Mg, and Fe/O and find that all five systems are broadly consistent with
solar ratios, with the exception of the metal-poor Barnard's
Star ([M/H] $= -0.33 \pm 0.06$ dex), which is substantially iron-poor ([Fe/H] $= -0.722 \pm 0.085$ dex). Our results establish
stellar chemical baselines for interior structure and atmospheric studies of these planetary systems with the Hubble Space
Telescope (HST) and James Webb Space
Telescope (JWST), while highlighting the remaining model-dependent limitations of precise abundance analyses for cool
stars.
Read the full abstract ▾
All authors Nicole Gromek, Drew Weisserman, Ryan Cloutier, Alexandrine L'Heureux, Charles Cadieux
Machine Learning
Rare Event Estimation via Iterative Unalignment
Hanming Yang, Daksh Mittal, Jing Dong et al. (+1 more)
As
agents are deployed with increased autonomy, even extremely rare events along their
stochastic output trajectories can occur and prove catastrophic. Safe deployment therefore does not depend on whether these events can occur, but on how often they might. We study the problem of estimating the probability of rare events that arise from
stochastic variation in the
agent's own actions. Estimating this type of risk requires searching over the combinatorially vast space of trajectories. Naive Monte Carlo is computationally prohibitive in this regime, and constructing effective importance sampling (IS) proposals requires coordinated changes to a context-dependent chain of conditional distributions. We develop a new IS method that perturbs the original model's weights to construct the proposal. The proposal is itself a differentiably parameterized
language model, enabling gradient-based search over weight space. We formulate an objective that combines a differentiable surrogate for event amplification and an adaptive regularization scheme that dynamically balances amplification against
estimator stability. We evaluate our approach on $\sim$120M and $\sim$2.6B models across three event families spanning 300+ rare events as rare as $10^{-9}$, all with independently estimated reference probabilities. In our most verifiable settings, we observe that our IS
estimator achieves over $800\times$ compute-weighted efficiency gains on average over naive Monte Carlo for events with probabilities lower than $10^{-7}$. Our implementation is available at https://github.com/namkoong-lab/iterative-unalignment.
Read the full abstract ▾
All authors Hanming Yang, Daksh Mittal, Jing Dong, Hongseok Namkoong
Algebraic Geometry
Boundedness of geometrically integral $\epsilon$-klt del Pezzo surfaces
Shikha Bhutani and Sudipta Das
We prove the remaining positive characteristic cases of the Borisov-Alexeev-Borisov
conjecture for geometrically integral $\varepsilon$-klt del Pezzo surfaces over arbitrary fields. Bernasconi and Martin established boundedness outside characteristics $2$, $3$, and $5$, and reduced the remaining cases to a uniform bound for the irregularity. We prove this bound, thereby completing the boundedness
theorem in all positive characteristics.
Read the full abstract ▾
Artificial Intelligence
Emergent Collusion in Long-Horizon LLM Agent Interaction
Xinrui Shi, Yanzhe Zhang and Diyi Yang
LLM
agents are increasingly deployed in collaborative settings, yet long-term interaction may give rise to undesirable coordination. We study the emergence of collusion in a long-horizon multi-
agent environment: two
agents repeatedly complete individual tasks, share task logs, verify each other's work, and receive rewards. We introduce realistic constraints under which higher reward is attainable only by violating the verification protocol, and find that
agents increasingly deviate from the protocol over repeated interactions. Collusion emerges in 94% of trajectories across 10 models, and more capable models within the same family reach it earlier. Controlled peer interventions show that collusion is shaped by peer behavior, while ablations reveal additional effects of the verification feedback
agents receive, their interaction history, and the reward structure. In particular, restricting the amount and scope of interaction history available to
agents reduces collusion. Overall, our findings show that long-horizon interaction can reshape how
agents coordinate in ways that create safety risks.
Read the full abstract ▾
Systems and Control
Efficient physiological control of an integrated system architecture for continuous-flow ventricular assist devices: in-silico study
Bruno J. Santos and Idagene A. Cestari
This study presents the development and in silico evaluation of an Integrated System Architecture (ISA) for the physiological control of continuous-flow ventricular assist devices (VADs). The system employs an automatic controller based on pressure measurements at the VAD inflow cannula to estimate heart rate and ventricular filling pressure, enabling dynamic speed adjustments that optimize VAD-patient interaction. The evaluation encompassed 27 simulation scenarios assessing four aspects: operating speed regulation, responsiveness to variable demand, adverse event mitigation, and physiological impact. The controller continuously adapted to preload, afterload, and heart rate, accommodating flow demands with speed adjustments ranging from -10% (-600 rpm) to +43% (+2,600 rpm) relative to the baseline (6,000 rpm). Furthermore, dynamic regulation eliminated ventricular suction and backflow events, which were previously observed in 12 of the 27 uncontrolled scenarios. Hemodynamic and metabolic outcomes demonstrated improved left ventricular ejection fraction (+6.23% to +62.31%), reduced ejection work (-1,658.64 to -206.37 mmHg*mL), decreased pressure-volume loop area (-9,485.17 to -1,056.05 mmHg*mL), and reduced myocardial oxygen consumption (-20.5 to -1.14 mL/min). Concurrently, total oxygen delivery increased (+56.9 to +626.23 mL/min) along with cardiac power (+100 to +3,070 mW). These findings demonstrate that the ISA successfully maintains physiological regulation and mitigates adverse events, establishing a solid foundation for future in vitro validation.
Read the full abstract ▾
Computation and Language
Jev for Scientific Decisions: Evaluating Semantic Choices and Their Consequences
Boyuan Deng, Shuyi Fan, Hongyang Zhang et al. (+1 more)
Scientific workflows often require choosing among known relations before a deterministic calculation can proceed. Whether observations share a culture, treatment or reference standard can change the scientific meaning of the resulting count or comparison. Once the evidence is prepared and the calculation is written as code, this choice reduces to one semantic judgment among listed candidates. We evaluate Jev on this step, using a harness that follows its documented guidance. The study compares twelve model configurations on twenty Choices in ten scientific cases built from published sources, each repeated five times. We measure semantic selections, downstream outputs and final claim labels separately. Jev tied with five other configurations at complete semantic correctness and achieved the lowest observed median latency and known mean cost among successful responses. On one question, whether samples shared a culture, three comparison models made seven wrong selections that changed downstream counts while preserving the correct final label. These results support using Jev to select relations when the evidence and candidates are prepared in advance, and show that evaluating it requires checking the relations and quantities that a workflow will reuse.
Read the full abstract ▾
All authors Boyuan Deng, Shuyi Fan, Hongyang Zhang, Xinhong Xie
Multiagent Systems
Perception-Aware Communication Middleware for Distributed Visual Perception in UAV Swarms
Manveen Kaur, Kevin Loi, Ifunanya Okafor et al. (+2 more)
Unmanned Aerial Vehicle (
UAV) swarms increasingly support safety-critical applications that rely on distributed visual perception. Meeting the low-latency requirements of these applications can require perception models to execute within the swarm on inference-capable
UAVs, creating a need for efficient
UAV-to-
UAV transport of high-bandwidth perception data. However, the Quality-of-Service (QoS) requirements of perception differ from conventional packet-level QoS; successful delivery of individual packets does not ensure that a complete, timely, and usable image is available for inference. We present a novel perception-aware communication middleware that treats complete perception-data samples as the communication objects for which QoS must be satisfied. The middleware extends a lightweight UDP broker-based publish-subscribe architecture with perception-specific services, including image fragmentation and reconstruction, concurrent packet transmission, priority-aware scheduling, and image quality assessment. The middleware is evaluated on a heterogeneous hardware testbed emulating a
UAV swarm using YOLOv8n object detection. Experimental results demonstrate low end-to-end application latency, substantially higher throughput than a lightweight UDP broker, effective prioritization of perception traffic under increasing background load, and mitigation of object-detection degradation through middleware-level image quality assessment. This work provides an initial framework for integrating AI-specific data handling into communication middleware to support emerging distributed AI applications in multi-
agent mobile cyber-physical systems.
Read the full abstract ▾
All authors Manveen Kaur, Kevin Loi, Ifunanya Okafor, Daniel Ng, Joseph Lucey-Renteria