Visual geometry transformers have become powerful architectures for multi-view 3D reconstruction, enabling joint prediction of multiple 3D attributes in a feed-forward manner. However, their computational cost grows quadratically with the input sequence length due to the global attention layers inside these models. This limits both their scalability and efficiency. In this work, we address this challenge with a simple yet general strategy: restricting the number of key/value tokens that each query interacts with during global attention. To achieve effective token selection, we introduce a two-stage framework. First, an inter-frame selection step operates at the frame level to identify frames that should be preserved. Second, an intra-frame selection step further discards more redundant tokens within the selected frames. Our analysis highlights the advantage of a diversity-based strategy for inter-frame selection, which ensures broad coverage of the scene. For intra-frame selection, we show that layer-aware sparsification is necessary, with the selection process guided by the entropy of the global attention pattern. Our approach offers a superior speed-accuracy trade-off compared to
Recommender systems are usually designed by engineers, researchers, designers, and other members of development teams. These systems are then evaluated based on goals set by the aforementioned teams and other business units of the platforms operating the recommender systems. This design approach emphasizes the designers' vision for how the system can best serve the interests of users, providers, businesses, and other stakeholders. Although designers may be well-informed about user needs through user experience and market research, they are still the arbiters of the system's design and evaluation, with other stakeholders' interests less emphasized in user-centered design and evaluation. When extended to recommender systems for social good, this approach results in systems that reflect the social objectives as envisioned by the designers and evaluated as the designers understand them. Instead, social goals and operationalizations should be developed through participatory and democratic processes that are accountable to their stakeholders. We argue that recommender systems aimed at improving social good should be designed *by* and *with*, not just *for*, the people who will experience
This work examines the role of recommender systems in promoting sustainability, social responsibility, and accountability, with a focus on alignment with the United Nations Sustainable Development Goals (SDGs). As recommender systems become increasingly integrated into daily interactions, they must go beyond personalization to support responsible consumption, reduce environmental impact, and foster social good. We explore strategies to mitigate the carbon footprint of recommendation models, ensure fairness, and implement accountability mechanisms. By adopting these approaches, recommender systems can contribute to sustainable and socially beneficial outcomes, aligning technological advancements with the SDGs focused on environmental sustainability and social well-being.
Suppose $G$ is a simple algebraic group defined over an algebraically closed field of good characteristic $p$. In 2018 Korhonen showed that if $H$ is a connected reductive subgroup of $G$ which contains a distinguished unipotent element $u$ of $G$ of order $p$, then $H$ is $G$-irreducible in the sense of Serre. We present a short and uniform proof of this result under an extra hypothesis using so-called good $A_1$ subgroups of $G$, introduced by Seitz. In the process we prove some new results about good $A_1$ subgroups of $G$ and their properties. We also formulate a counterpart of Korhonen's theorem for overgroups of $u$ which are finite groups of Lie type. Moreover, we generalize both results above by removing the restriction on the order of $u$ under a mild condition on $p$ depending on the rank of $G$, and we present an analogue of Korhonen's theorem for Lie algebras.
Numerous pre-training techniques for visual document understanding (VDU) have recently shown substantial improvements in performance across a wide range of document tasks. However, these pre-trained VDU models cannot guarantee continued success when the distribution of test data differs from the distribution of training data. In this paper, to investigate how robust existing pre-trained VDU models are to various distribution shifts, we first develop an out-of-distribution (OOD) benchmark termed Do-GOOD for the fine-Grained analysis on Document image-related tasks specifically. The Do-GOOD benchmark defines the underlying mechanisms that result in different distribution shifts and contains 9 OOD datasets covering 3 VDU related tasks, e.g., document information extraction, classification and question answering. We then evaluate the robustness and perform a fine-grained analysis of 5 latest VDU pre-trained models and 2 typical OOD generalization algorithms on these OOD datasets. Results from the experiments demonstrate that there is a significant performance gap between the in-distribution (ID) and OOD settings for document images, and that fine-grained analysis of distribution shifts
In this paper we study continuous-time stochastic control problems with both monotone and classical controls motivated by the so-called public good contribution problem. That is the problem of n economic agents aiming to maximize their expected utility allocating initial wealth over a given time period between private consumption and irreversible contributions to increase the level of some public good. We investigate the corresponding social planner problem and the case of strategic interaction between the agents, i.e. the public good contribution game. We show existence and uniqueness of the social planner's optimal policy, we characterize it by necessary and sufficient stochastic Kuhn-Tucker conditions and we provide its expression in terms of the unique optional solution of a stochastic backward equation. Similar stochastic first order conditions prove to be very useful for studying any Nash equilibria of the public good contribution game. In the symmetric case they allow us to prove (qualitative) uniqueness of the Nash equilibrium, which we again construct as the unique optional solution of a stochastic backward equation. We finally also provide a detailed analysis of the so-ca
We study the mechanism design problem of selling a public good to a group of agents by a principal in the correlated private value environment. We assume the principal only knows the expectations of the agents' values, but does not know the joint distribution of the values. The principal evaluates a mechanism by the worst-case expected revenue over joint distributions that are consistent with the known expectations. We characterize maxmin public good mechanisms among dominant-strategy incentive compatible and ex-post individually rational mechanisms for the two-agent case and for a special $N$-agent ($N>2$) case.
The works of Poincare, Birkhoff, Witt and Cartier, Milnor, Moore on the connected cocommutative Hopf algebras translated in the language of operads means that the triple of operads (Com, As, Lie) endowed with the Hopf compatiblity relation is good. In this paper, we focus on left dipterous (resp. right dipterous) algebras which are associative algebras with an extra left (resp. right) module on themselves and look for good triples were $As$ is replaced by the dipterous operad Dipt. Since the work of Loday and Ronco, the triple of operads (As, Dipt, B_\infty) endowed with the semi-Hopf compatibility relations is known to be good. In this paper, we prove that the triple of operads (As, Dipt, Grove) endowed with the so-called (nonunital) semi-infinitesimal compatibility relations is good. For that, explicit constructions of the free dipterous algebra and the free grove-algebra over a K-vector space V are given. These constructions turn out to be related to rooted planar trees and the little an large Schroeder numbers. Many examples of dipterous algebras are given, notably the free L-dipterous algebras. As a corollary of our results, we also recover that the triple of operads (2As, Dip
This paper targets to search so-called \emph{good} generators by doing a brief survey over the generators developed in the history of pseudo-random number generators (PRNGs), verify their claims and rank them based on strong empirical tests in same platforms. To do this, the genre of PRNGs developed so far are explored and classified into three groups -- linear congruential generator based, linear feedback shift register based and cellular automata based. From each group, the well-known widely used generators which claimed themselves to be `\emph{good}' are chosen. Overall $30$ PRNGs are selected in this way on which two types of empirical testing are done -- blind statistical tests with Diehard battery of tests, battery \emph{rabbit} of TestU01 library and NIST statistical test-suite as well as graphical tests (lattice test and space-time diagram test). Finally, the selected PRNGs are divided into $24$ groups and are ranked according to their overall performance in all empirical tests.
We modify the transchromatic character maps to land in a faithfully flat extension of Morava E-theory. Our construction makes use of the interaction between topological and algebraic localization and completion. As an application we prove that centralizers of tuples of commuting prime-power order elements in good groups are good and we compute a new example.
The AI for social good movement has now reached a state in which a large number of one-off demonstrations have illustrated that partnerships of AI practitioners and social change organizations are possible and can address problems faced in sustainable development. In this paper, we discuss how moving from demonstrations to true impact on humanity will require a different course of action, namely open platforms containing foundational AI capabilities to support common needs of multiple organizations working in similar topical areas. We lend credence to this proposal by describing three example patterns of social good problems and their AI-based solutions: natural language processing for making sense of international development reports, causal inference for providing guidance to vulnerable individuals, and discrimination-aware classification for supporting unbiased allocation decisions. We argue that the development of such platforms will be possible through convenings of social change organizations, AI companies, and grantmaking foundations.
Knowledge distillation (KD) is a general neural network training approach that uses a teacher model to guide the student model. Existing works mainly study KD from the network output side (e.g., trying to design a better KD loss function), while few have attempted to understand it from the input side. Especially, its interplay with data augmentation (DA) has not been well understood. In this paper, we ask: Why do some DA schemes (e.g., CutMix) inherently perform much better than others in KD? What makes a "good" DA in KD? Our investigation from a statistical perspective suggests that a good DA scheme should reduce the covariance of the teacher-student cross-entropy. A practical metric, the stddev of teacher's mean probability (T. stddev), is further presented and well justified empirically. Besides the theoretical understanding, we also introduce a new entropy-based data-mixing DA scheme, CutMixPick, to further enhance CutMix. Extensive empirical studies support our claims and demonstrate how we can harvest considerable performance gains simply by using a better DA scheme in knowledge distillation.
Data for good implies unfettered access to data. But data owners must be conservative about how, when, and why they share data or risk violating the trust of the people they aim to help, losing their funding, or breaking the law. Data sharing agreements can help prevent privacy violations, but require a level of specificity that is premature during preliminary discussions, and can take over a year to establish. We consider the generation and use of synthetic data to facilitate ad hoc collaborations involving sensitive data. A good synthetic dataset has two properties: it is representative of the original data, and it provides strong guarantees about privacy. In this paper, we discuss important use cases for synthetic data that challenge the state of the art in privacy-preserving data generation, and describe DataSynthesizer, a dataset generation tool that takes a sensitive dataset as input and generates a structurally and statistically similar synthetic dataset, with strong privacy guarantees, as output. The data owners need not release their data, while potential collaborators can begin developing models and methods with some confidence that their results will work similarly on th
We study classical and quantum LDPC codes of constant rate obtained by the lifted product construction over non-abelian groups. We show that the obtained families of quantum LDPC codes are asymptotically good, which proves the qLDPC conjecture. Moreover, we show that the produced classical LDPC codes are also asymptotically good and locally testable with constant query and soundness parameters, which proves a well-known conjecture in the field of locally testable codes.
Pretty good state transfer in networks of qubits occurs when a continuous-time quantum walk allows the transmission of a qubit state from one node of the network to another, with fidelity arbitrarily close to 1. We prove that in a Heisenberg chain with n qubits there is pretty good state transfer between the nodes at the j-th and (n-j+1)-th position if n is a power of 2. Moreover, this condition is also necessary for j=1. We obtain this result by applying a theorem due to Kronecker about Diophantine approximations, together with techniques from algebraic graph theory.
We prove that the "good" Boussinesq model with the periodic boundary condition is locally well-posed in the space $H^{s}\times H^{s-2}$ for $s > -3/8$. In the proof, we employ the normal form approach, which allows us to explicitly extract the rougher part of the solution. This also leads to the conclusion that the remainder is in a smoother space $C([0,T], H^{s+a}), where $0 <= a < \min (2s+1, 1/2)$. If we have a mean-zero initial data, this implies a smoothing effect of this order for the non-linearity. This is new even in the previously considered cases $s > -1/4$.
Soft robots have struggled to support large forces and moments while also supporting their own weight against gravity. This limits their ability to reach certain configurations necessary for tasks such as inspection and pushing objects up. We have overcome this limitation by creating an electrically driven metamaterial soft arm using handed shearing auxetics (HSA) and bendable extendable torque resistant (BETR) shafts. These use the large force and torque capacity of HSAs and the nestable torque transmission of BETRs to create a strong soft arm. We found that the HSA arm was able to push 2.3 kg vertically and lift more than 600 g when positioned horizontally, supporting 0.33 Nm of torque at the base. The arm is able to move between waypoints while carrying the large payload and demonstrates consistent movement with path variance below 5 mm. The HSA arm's ability to perform active grasping with HSA grippers was also demonstrated, requiring 20 N of pull force to dislodge the object. Finally, we test the arm in a pipe inspection task. The arm is able to locate all the defects while sliding against the inner surface of the pipe, demonstrating its compliance.
We have performed an extensive statistical investigation of how interplanetary fast forward shocks affect certain turbulence parameters, namely, the cross-helicity, $σ_c$, residual energy, $σ_r$, and magnetic helicity, $σ_m$. A total of 371 shocks detected by Wind at 1 au and seven shocks by Solar Orbiter at 0.3-0.5 au have been analysed. We explore how the aforementioned turbulence parameters and their variation across the shock depend on shock characteristics including the gas compression ratio, upstream plasma beta, velocity jump and shock angle. In the shock vicinity, fluctuations tend on average to show antisunward imbalance (measured as $σ_c>0$ when rectified to the Parker spiral direction), a dominance of magnetic energy ($σ_r<0$) and zero $σ_m$, all being typical solar wind properties . Antisunward imbalance and equipartition ($σ_r \sim0$) in the upstream is increasingly prevalent with increasing shock velocity jump and decreasing upstream beta and shock angle. Shocks with large velocity jumps and gas compression ratios have considerably more balanced ($σ_c\sim0$) and more magnetically dominated fluctuations downstream than upstream. From upstream to downstream, we al
Intermittency has been studied extensively in the fast and slow solar winds but to a far lesser extent in interplanetary coronal mass ejections (ICMEs). While ICMEs are often characterized by their relatively smooth, large-scale magnetic flux rope structures, a spectrum of fluctuations is nonetheless present at smaller scales. We have examined kurtosis and its scaling exponents at magnetohydrodynamic inertial scales in 49 ICMEs observed between 0.25 and 1 au by Parker Solar Probe and Solar Orbiter, and compared the results to those obtained for the ICME sheath regions and ambient solar wind intervals. Kurtosis behaves similarly in all intervals studied and presents a universal behavior typical of intermittent time series. The ICMEs displayed a radially invariant level of intermittency, suggesting that they are relatively static, well-developed turbulent environments. In the sheath regions, the level of intermittency increased with distance, indicating that the turbulence is not yet fully developed at small heliocentric distances. In addition to intermittent fluctuations related to turbulence, the sheath regions may possess a population of non-turbulent structures that increase the
We report the first nucleon gluon parton distribution function (PDF) using Large-Momentum Effective Theory (LaMET). We focus on the gluon operator which was demonstrated to have the best signal-to-noise in the previous attempt [1] in computing gluon PDFs using LaMET. We compute the corresponding Wilson coefficients needed for the hybrid-renormalized matrix elements and the matching kernel to convert the quasi-PDF to the lightcone one at the one-loop level. We demonstrate that with the proper Wilson coefficients in place, the counterterms for the renormalization are independent of the hadron and mass within statistical error. Using the resulting renormalization, we then compute the nucleon PDF using a HISQ ensemble generated by the MILC collaboration with $N_f=2+1+1$, $a \approx 0.12$ fm, with valence pion masses of 310 and 690 MeV and two gauge link smearing techniques. Despite the physics effects of the heavier than physical pion masses and gauge link smearing, this calculation provides excellent proof of principle and compares reasonably with selected global fit results.