Mobile Agents can autonomously execute user instructions, which requires hybrid-capabilities reasoning, including screen summary, subtask planning, action decision and action function. However, existing agents struggle to achieve both decoupled enhancement and balanced integration of these capabilities. To address these challenges, we propose Channel-of-Mobile-Experts (CoME), a novel agent architecture consisting of four distinct experts, each aligned with a specific reasoning stage, CoME activates the corresponding expert to generate output tokens in each reasoning stage via output-oriented activation. To empower CoME with hybrid-capabilities reasoning, we introduce a progressive training strategy: Expert-FT enables decoupling and enhancement of different experts' capability; Router-FT aligns expert activation with the different reasoning stage; CoT-FT facilitates seamless collaboration and balanced optimization across multiple capabilities. To mitigate error propagation in hybrid-capabilities reasoning, we propose InfoGain-Driven DPO (Info-DPO), which uses information gain to evaluate the contribution of each intermediate step, thereby guiding CoME toward more informative reasoni
Large language models (LLMs) often retain outdated or incorrect information from pre-training, which undermines their reliability. While model editing methods have been developed to address such errors without full re-training, they frequently suffer from knowledge conflicts, where outdated information interferes with new knowledge. In this work, we propose Conflict-free Model Editing (CoME), a novel framework that enhances the accuracy of knowledge updates in LLMs by selectively removing outdated knowledge. CoME leverages unlearning to mitigate knowledge interference, allowing new information to be integrated without compromising relevant linguistic features. Through experiments on GPT-J and LLaMA-3 using Counterfact and ZsRE datasets, we demonstrate that CoME improves both editing accuracy and model reliability when applied to existing editing methods. Our results highlight that the targeted removal of outdated knowledge is crucial for enhancing model editing effectiveness and maintaining the model's generative performance.
World models are critical for autonomous driving to simulate environmental dynamics and generate synthetic data. Existing methods struggle to disentangle ego-vehicle motion (perspective shifts) from scene evolvement (agent interactions), leading to suboptimal predictions. Instead, we propose to separate environmental changes from ego-motion by leveraging the scene-centric coordinate systems. In this paper, we introduce COME: a framework that integrates scene-centric forecasting Control into the Occupancy world ModEl. Specifically, COME first generates ego-irrelevant, spatially consistent future features through a scene-centric prediction branch, which are then converted into scene condition using a tailored ControlNet. These condition features are subsequently injected into the occupancy world model, enabling more accurate and controllable future occupancy predictions. Experimental results on the nuScenes-Occ3D dataset show that COME achieves consistent and significant improvements over state-of-the-art (SOTA) methods across diverse configurations, including different input sources (ground-truth, camera-based, fusion-based occupancy) and prediction horizons (3s and 8s). For example
Machine learning models must continuously self-adjust themselves for novel data distribution in the open world. As the predominant principle, entropy minimization (EM) has been proven to be a simple yet effective cornerstone in existing test-time adaption (TTA) methods. While unfortunately its fatal limitation (i.e., overconfidence) tends to result in model collapse. For this issue, we propose to Conservatively Minimize the Entropy (COME), which is a simple drop-in replacement of traditional EM to elegantly address the limitation. In essence, COME explicitly models the uncertainty by characterizing a Dirichlet prior distribution over model predictions during TTA. By doing so, COME naturally regularizes the model to favor conservative confidence on unreliable samples. Theoretically, we provide a preliminary analysis to reveal the ability of COME in enhancing the optimization stability by introducing a data-adaptive lower bound on the entropy. Empirically, our method achieves state-of-the-art performance on commonly used benchmarks, showing significant improvements in terms of classification accuracy and uncertainty estimation under various settings including standard, life-long and
For each smooth curve over a finite field, after puncturing it at finitely many points, we construct local systems on it of geometric origin which do not come from a family of abelian varieties. We do so by proving a criterion which must be satisfied by local systems which do come from abelian varieties, inspired by an analogous Hodge theoretic criterion in characteristic zero.
The Born Rule plays a critical role in quantum mechanics (QM) since it supplies the link between the mathematical formalism and experimental results in terms of probabilities. The Born Rule does not occur in ordinary probability theory. Where then does it come from? This has been a topic of considerable controversy in the literature. We take the approach of asking what is the simplest extension of ordinary probability theory where the Born rule appears. This is answered by showing that the Born Rule appears by adding the notion of superposition events (in addition to the ordinary discrete events) to finite probability theory. Hence the rule does not need any physics-based derivation. It is simply a feature of the mathematics of superposition when only superposition events are added to ordinary probability theory.
Specially adapted speech recognition models are necessary to handle stuttered speech. For these to be used in a targeted manner, stuttered speech must be reliably detected. Recent works have treated stuttering as a multi-class classification problem or viewed detecting each dysfluency type as an isolated task; that does not capture the nature of stuttering, where one dysfluency seldom comes alone, i.e., co-occurs with others. This work explores an approach based on a modified wav2vec 2.0 system for end-to-end stuttering detection and classification as a multi-label problem. The method is evaluated on combinations of three datasets containing English and German stuttered speech, yielding state-of-the-art results for stuttering detection on the SEP-28k-Extended dataset. Experimental results provide evidence for the transferability of features and the generalizability of the method across datasets and languages.
We study actions of higher rank lattices $Γ<G$ on hyperbolic spaces, and we show that all such actions satisfying mild properties come from the rank-one factors of $G$. In particular, all non-elementary actions on an unbounded hyperbolic space are of this type.
China’s Tianwen-1 orbiter captured a faint but remarkable view of 3I/ATLAS as the interstellar comet raced past Mars。 The images reveal its tail and glowing coma, offering scientists a glimpse of material born beyond our solar system
We pick up the thread of $B\to Kπ$ direct CPV (DCPV) difference, linking to the 4th generation prediction (and current quest) of $t$-dependent CPV (TCPV) in $B_s \to J/ψ\,φ$. We then soar up to the heavens with possible source of CPV for the Universe (CPV-4-U); for the "cauldron" that stirs strong phase transitions, we raise the possible link between electroweak symmetry breaking and large Yukawa couplings (4-EWSB). With the prognosis for 2011, 2012 and beyond, we stress these as preview for the possible 4th generation to come: CPV-4-U and the \emph{raison d'être} for our Universe (and ourselves); 4-EWSB and the \emph{raison d'être} for the LHC itself.
The current standard theory of the origin of the Moon is that the Earth was hit by a giant impactor the size of Mars causing ejection of iron poor impactor mantle debris that coalesced to form the Moon. But where did this Mars-sized impactor come from? Isotopic evidence suggests that it came from 1AU radius in the solar nebula and computer simulations are consistent with it approaching Earth on a zero-energy parabolic trajectory. But how could such a large object form in the disk of planetesimals at 1AU without colliding with the Earth early-on before having a chance to grow large or before its or the Earth's iron core had formed? We propose that the giant impactor could have formed in a stable orbit among debris at the Earth's Lagrange point $L_4$ (or $L_5$). We show such a configuration is stable, even for a Mars-sized impactor. It could grow gradually by accretion at $L_4$ (or $L_5$), but eventually gravitational interactions with other growing planetesimals could kick it out into a chaotic creeping orbit which we show would likely cause it to hit the Earth on a zero-energy parabolic trajectory. This paper argues that this scenario is possible and should be further studied.
We establish some connections between nonresonant $A$-hypergeometric systems and de Rham-type complexes. This allows us to determine which of these $A$-hypergeometric systems "come from geometry."
Neural Style Transfer is a striking, recently-developed technique that uses neural networks to artistically redraw an image in the style of a source style image. This paper explores the use of this technique in a production setting, applying Neural Style Transfer to redraw key scenes in 'Come Swim' in the style of the impressionistic painting that inspired the film. We document how the technique can be driven within the framework of an iterative creative process to achieve a desired look, and propose a mapping of the broad parameter space to a key set of creative controls. We hope that this mapping can provide insights into priorities for future research.
We study the homogeneous Landau equation with Maxwell molecules and prove that the entropy production is non-increasing provided the directional temperatures are well-distributed and the solution admits a moment of order $\ell$, for some $\ell$ arbitrarily close to $2$. It implies that for an initial condition with finite moment of order $\ell$, the entropy production is guaranteed to be non-increasing after a certain time, that we explicitly compute. This is the first partial answer to a conjecture made by Henry P. McKean in 1966 on the sign of the time-derivatives of the entropy. Without moment assumptions, we obtain a possibly sharp short-time regularization rate for the entropy production, and exponential decay for large times.
We build solutions to Kac's particle system and show that their empirical measures converge to the solution of the space-homogeneous Boltzmann equation in the regime of very soft potentials. This proves propagation of chaos for the last class of kernels for which it was still open. The proof relies on new estimates on the dissipation of the Fisher information along the Boltzmann equation, which allow us to control the strong singularities of the system. These estimates are obtained thanks to a new inequality related to the fractional heat flow on the sphere, that might be of independent interest.
The $Λ$-Fleming-Viot process is a probability measure-valued process that is dual to a $Λ$-coalescent that allows multiple collisions. In this paper, we consider a class of $Λ$-Fleming-Viot processes with Brownian spatial motion and with associated $Λ$-coalescents that come down from infinity. Notably, these processes have the compact support property: the support of the process becomes finite as soon as $t>0$, even though the initial measure has unbounded support. We obtain asymptotic results characterizing the rates at which the initial supports become finite. The rates of coming down are expressed in terms of the asymptotic inverse function of the tail distribution of the initial measure and the speed function of coming down from infinity for the corresponding $Λ$-coalescent.
We derive a weak-strong uniqueness and stability principle for the Landau equation in the soft potentials case (including Coulomb interactions). The distance between two solutions is measured by their relative entropy, which to our knowledge was never used before in stability estimates. The logarithm of the strong solution is required to have polynomial growth while the weak solution can be any H-solution with sufficiently many moments at initial time. Since we require a substantial amount of regularity on the strong solution, we also provide an example of sufficient conditions on the initial data that ensure this regularity in the Coulomb (and very soft potentials) case.
Neurofeedback training (NFT) aims to teach self-regulation of brain activity through real-time feedback, but suffers from highly variable outcomes and poorly understood mechanisms, hampering its validation. To address these issues, we propose a formal computational model of the NFT closed loop. Using Active Inference, a Bayesian framework modelling perception, action, and learning, we simulate agents interacting with an NFT environment. This enables us to test the impact of design choices (e.g., feedback quality, biomarker validity) and subject factors (e.g., prior beliefs) on training. Simulations show that training effectiveness is sensitive to feedback noise or bias, and to prior beliefs (highlighting the importance of guiding instructions), but also reveal that perfect feedback is insufficient to guarantee high performance. This approach provides a tool for assessing and predicting NFT variability, interpret empirical data, and potentially develop personalized training protocols.
Hodge theory associates to a smooth projective variety over $\mathbb{C}$ a piece of linear algebra information, called a $\mathbb{Q}$-Hodge structure. Conversely, it is a natural question which abstract $\mathbb{Q}$-Hodge structures arise from the cohomology of a smooth projective complex variety, or more generally, from a pure motive over $\mathbb{C}$. By a classical argument involving Griffiths transversality and a Baire category argument, it is well known that there are many Hodge structures which do not come from geometry in this sense. However, the argument is not constructive, and does not seem to give a criterion to decide whether a given Hodge structure comes from geometry. We formulate an intrinsic condition on a $\mathbb{Q}$-Hodge structure that we expect to be satisfied for all Hodge structures coming from geometry. We prove that this expectation follows from the conjunction of two fundamental conjectures in Hodge theory and transcendence theory: the conjecture that Hodge cycles are motivated and André's generalized Grothendieck period conjecture. By doing so, we exhibit explicit examples of $\mathbb{Q}$-Hodge structures which should not come from geometry.
We consider a drift-diffusion process of $N$ stochastic particles and show that its empirical measure converges, as $N\rightarrow \infty$, to the solution of the Landau equation. We work in the regime of very soft and Coulomb potentials using a tightness/uniqueness method. To claim uniqueness, we need high integrability estimates that we obtain by crucially exploiting the dissipation of the Fisher information at the level of the particle system. To be able to exploit these estimates as $N\rightarrow \infty$, we prove the affinity in infinite dimension of the entropy production and Fisher information dissipation (and other first and second-order versions of the Fisher information through a general theorem), results which were up to now only known for the entropy and the usual Fisher information.