搜索结果：Constraints

Deep Reinforcement Learning (RL) has demonstrated impressive results in solving complex robotic tasks such as quadruped locomotion. Yet, current solvers fail to produce efficient policies respecting hard constraints. In this work, we advocate for integrating constraints into robot learning and present Constraints as Terminations (CaT), a novel constrained RL algorithm. Departing from classical constrained RL formulations, we reformulate constraints through stochastic terminations during policy learning: any violation of a constraint triggers a probability of terminating potential future rewards the RL agent could attain. We propose an algorithmic approach to this formulation, by minimally modifying widely used off-the-shelf RL algorithms in robot learning (such as Proximal Policy Optimization). Our approach leads to excellent constraint adherence without introducing undue complexity and computational overhead, thus mitigating barriers to broader adoption. Through empirical evaluation on the real quadruped robot Solo crossing challenging obstacles, we demonstrate that CaT provides a compelling solution for incorporating constraints into RL frameworks. Videos and code are available a

This paper studies mean-field control with joint law dependence under dynamic expectation constraints and/or dynamic state-control-law constraints. We pioneer the establishment of the stochastic maximum principle (SMP) and the derivation of the backward SDE (BSDE) from the perspective of constrained optimization using the method of Lagrangian multipliers. We first propose to embed the constrained mean-field control (C-MFC) with joint-law dependence into some abstract optimization problems with constraints on Banach spaces, for which we develop the generalized Fritz-John (FJ) optimality conditions. We then prove the stochastic maximum principle (SMP) for C-MFC by transforming the FJ conditions into an equivalent stochastic first-order condition associated with a general type of constrained forward-backward SDEs (FBSDEs). Contrary to the existing literature, we treat the McKean-Vlasov SDE as an infinite-dimensional equality constraint such that the BSDE induced by the FJ first-order optimality conditions can be interpreted as the generalized Lagrange multiplier. We also employ the methodology to stochastic control and mean field game problems under dynamic constraints.

Global constraints proved themselves to be an efficient tool for modelling and solving large-scale real-life combinatorial problems. They encapsulate a set of binary constraints and using global reasoning about this set they filter the domains of involved variables better than arc consistency among the set of binary constraints. Moreover, global constraints exploit semantic information to achieve more efficient filtering than generalised consistency algorithms for n-ary constraints. Continued expansion of constraint programming (CP) to various application areas brings new challenges for design of global constraints. In particular, application of CP to advanced planning and scheduling (APS) requires dynamic additions of new variables and constraints during the process of constraint satisfaction and, thus, it would be helpful if the global constraints could adopt new variables. In the paper, we give a motivation for such dynamic global constraints and we describe a dynamic version of the well-known alldifferent constraint.

This work investigates the allowed parameter spaces of a simplified dark matter (DM) model characterized by a spin-0 mediator with masses in the low to intermediate range ($ < $ 10 GeV). We systematically divide the parameter space into various mass regions of the mediator and constrain the model parameters using a diverse set of observables, including flavour-changing charged and neutral current processes such as rare and semi-leptonic decays of pseudoscalar mesons (B and K), electroweak precision observables, alongside data from fixed-target experiments. Additionally, we explore the model's capability to explain recent Belle-II data on invisible B-meson decays. Our study includes a detailed examination of DM properties and the constraints from Big Bang nucleosynthesis. We present bounds on model parameters through individual and simultaneous analyses of the available inputs and highlight their implications for understanding DM phenomenology. Furthermore, we obtain bounds on the couplings of the possible gauge-invariant dimension-5 operators, leading to the possible interactions between the spin-0 mediator and the SM gauge bosons and fermions. This study comprehensively investi

Recent efforts to develop trustworthy AI systems have increased interest in learning problems with explicit requirements, or constraints. In deep learning, however, such problems are often handled through fixed weighted-sum penalization: the constraints are added to the task loss with fixed coefficients, and the resulting scalarized objective is minimized. This position paper argues that fixed penalization is often ill-suited for deep learning problems with non-negotiable requirements for several reasons. First, in non-convex settings, the penalized and constrained problems are generally not equivalent, so solving the former need not solve the latter. Second, fixed penalization weakens hard requirements into soft penalties to be traded off against task performance. Third, choosing penalty coefficients to indirectly solve the constrained problem often involves costly trial and error, because changing them alters the penalized objective itself, and hence can mean solving the wrong problem altogether. We therefore argue that, when a deep learning problem specifies non-negotiable requirements, the constrained formulation itself should be the starting point, not the surrogate problem de

The search for non-Gaussian signatures in the Cosmic Microwave Background (CMB) is crucial for understanding the physics of the early Universe. Given the possibility of non-Gaussian fluctuations in the CMB, a recent revision to the standard $Λ$-Cold Dark Matter ($Λ$CDM) model has been proposed, dubbed "Super-$Λ$CDM". This model introduces additional free parameters to account for the potential effects of a trispectrum in the primordial fluctuations. In this study, we explore the impact of the Super-$Λ$CDM model on current constraints on neutrino physics. In agreement with previous research, our analysis reveals that for most of the datasets, the Super-$Λ$CDM parameter $A_0$ significantly deviates from zero at over a $95\%$ confidence level. We then demonstrate that this signal might influence current constraints in the neutrino sector. Specifically, we find that the current constraints on neutrino masses may be relaxed by over a factor of two within the Super-$Λ$CDM framework, thanks to the correlation present with $A_0$. Consequently, locking $A_0=0$ might introduce a bias, leading to overly stringent constraints on the total neutrino mass.

Many constraint satisfaction and optimisation problems can be solved effectively by encoding them as instances of the Boolean Satisfiability problem (SAT). However, even the simplest types of constraints have many encodings in the literature with widely varying performance, and the problem of selecting suitable encodings for a given problem instance is not trivial. We explore the problem of selecting encodings for pseudo-Boolean and linear constraints using a supervised machine learning approach. We show that it is possible to select encodings effectively using a standard set of features for constraint problems; however we obtain better performance with a new set of features specifically designed for the pseudo-Boolean and linear constraints. In fact, we achieve good results when selecting encodings for unseen problem classes. Our results compare favourably to AutoFolio when using the same feature set. We discuss the relative importance of instance features to the task of selecting the best encodings, and compare several variations of the machine learning method.

The model of Dynamic Meta-Constraints has special activity constraints which can activate other constraints. It also has meta-constraints which range over other constraints. An algorithm is presented in which constraints can be assigned one of five different satisfaction values, which leads to the assignment of domain values to the variables in the CSP. An outline of the model and the algorithm is presented, followed by some initial results for two problems: a simple classic CSP and the Car Configuration Problem. The algorithm is shown to perform few backtracks per solution, but to have overheads in the form of historical records required for the implementation of state.

Constraint propagation is one of the techniques central to the success of constraint programming. To reduce search, fast algorithms associated with each constraint prune the domains of variables. With global (or non-binary) constraints, the cost of such propagation may be much greater than the quadratic cost for binary constraints. We therefore study the computational complexity of reasoning with global constraints. We first characterise a number of important questions related to constraint propagation. We show that such questions are intractable in general, and identify dependencies between the tractability and intractability of the different questions. We then demonstrate how the tools of computational complexity can be used in the design and analysis of specific global constraints. In particular, we illustrate how computational complexity can be used to determine when a lesser level of local consistency should be enforced, when constraints can be safely generalized, when decomposing constraints will reduce the amount of pruning, and when combining constraints is tractable.

In the dark dimension scenario, which predicts an extra dimension of micron scale, dark gravitons (KK modes) are a natural dark matter candidate. In this paper, we study observable features of this model. In particular, their decay to standard matter fields can distort the CMB and impact other astrophysical signals. Using this we place bounds on the parameters of this model. In particular we find that the natural range of parameters in this scenario is consistent with these constraints and leads to the prediction that the mean mass of the dark matter today is close to a few hundred keV and the effective size of the extra dimension is around $1 - 30 \;μ\mathrm{m}$.

Primordial black holes (PBHs) can form in the early Universe via the collapse of large density perturbations. There are tight constraints on the abundance of PBHs formed due to their gravitational effects and the consequences of their evaporation. These abundance constraints can be used to constrain the primordial power spectrum, and hence models of inflation, on scales far smaller than those probed by cosmological observations. We compile, and where relevant update, the constraints on the abundance of PBHs before calculating the constraints on the curvature perturbation, taking into account the growth of density perturbations prior to horizon entry. We consider two simple parameterizations of the curvature perturbation spectrum on the scale of interest: constant and power-law. The constraints from PBHs on the amplitude of the power spectrum are typically in the range 10^{-2}-10^{-1} with some scale dependence.

In this work, we investigate the parameter space of the Georgi-Machacek (GM) model, where we consider many theoretical and experimental constraints such as the perturbativity, vacuum stability, unitarity, electroweak precision tests, the Higgs di-photon decay, the Higgs total decay width and the LHC measurements of the signal strengths of the SM-like Higgs boson $h$ in addition to the constraints from doubly charged Higgs bosons and Drell-Yan di-photon production and the indirect constraint from the $b\to s$ transition processes. We investigate also the possibility that the electroweak vacuum could be destabilized by unwanted wrong minima that may violate the CP and/or the electric charge symmetries. We found that about 40 \% of the parameter space that fulfills the above mentioned constraints; are excluded by these unwanted minima. In addition, we found that the negative searches for a heavy resonance could exclude a significant part of the viable parameter space; and future searches could exclude more regions in the parameter space.

We derive conditions for a nonholonomic system subject to nonlinear constraints (obeying Chetaev's rule) to preserve a smooth volume form. When applied to affine constraints, these conditions dictate that a basic invariant density exists if and only if a certain 1-form is exact and a certain function vanishes (this function automatically vanishes for linear constraints). Moreover, this result can be extended to geodesic flows for arbitrary metric connections and the sufficient condition manifests as integrability of the torsion. As a consequence, volume-preservation of a nonholonomic system is closely related to the torsion of the nonholonomic connection. Examples of nonlinear/affine/linear constraints are considered.

We combine the Ly-alpha forest power spectrum (LYA) from the Sloan Digital Sky Survey (SDSS) and high resolution spectra with cosmic microwave background (CMB) including 3-year WMAP, and supernovae (SN) and galaxy clustering constraints to derive new constraints on cosmological parameters. The existing LYA power spectrum analysis is supplemented by constraints on the mean flux decrement derived using a principle component analysis for quasar continua, which improves the LYA constraints on the linear power. We find some tension between the WMAP3 and LYA power spectrum amplitudes, at the ~2 sigma level, which is partially alleviated by the inclusion of other observations: we find σ_8=0.85\pm 0.02 compared to sigma_8=0.80 \pm 0.03 without LYA. For the slope we find ns=0.965\pm0.012. We find no evidence for the running of the spectral index in the combined analysis, dn/dln k=-(1.5\pm 1.2) x 10^{-2}, in agreement with inflation. The limits on the sum of neutrino masses are significantly improved: $\sum m_ν<0.17\eV$ at 95% (<0.32eV at 99.9%). This result, when combined with atmospheric and solar neutrino mixing constraints, requires that the neutrino masses cannot be degenerate, m_

We address the issue of incorporating a particular yet expressive form of integrity constraints (namely, denial constraints) into probabilistic databases. To this aim, we move away from the common way of giving semantics to probabilistic databases, which relies on considering a unique interpretation of the data, and address two fundamental problems: consistency checking and query evaluation. The former consists in verifying whether there is an interpretation which conforms to both the marginal probabilities of the tuples and the integrity constraints. The latter is the problem of answering queries under a "cautious" paradigm, taking into account all interpretations of the data in accordance with the constraints. In this setting, we investigate the complexity of the above-mentioned problems, and identify several tractable cases of practical relevance.

This paper studies the dynamic programming principle using the measurable selection method for stochastic control of continuous processes. The novelty of this work is to incorporate intermediate expectation constraints on the canonical space at each time t. Motivated by some financial applications, we show that several types of dynamic trading constraints can be reformulated into expectation constraints on paths of controlled state processes. Our results can therefore be employed to recover the dynamic programming principle for these optimal investment problems under dynamic constraints, possibly path-dependent, in a non-Markovian framework.

This paper focuses on martingale optimal transport problems when the martingales are assumed to have bounded quadratic variation. First, we give a result that characterizes the existence of a probability measure satisfying some convex transport constraints in addition to having given initial and terminal marginals. Several applications are provided: martingale measures with volatility uncertainty, optimal transport with capacity constraints, and Skorokhod embedding with bounded times. Next, we extend this result to multi-marginal constraints. Finally, we consider an optimal transport problem with constraints and obtain its Kantorovich duality. A corollary of this result is a monotonicity principle which gives a geometric way of identifying the optimizer.

Galaxy cluster surveys will be a powerful probe of dark energy. At the same time, cluster abundance is sensitive to any non-Gaussianity of the primordial density field. It is therefore possible that non-Gaussian initial conditions might be misinterpreted as a sign of dark energy or at least degrade the expected constraints on dark energy parameters. To address this issue, we perform a likelihood analysis of an ideal cluster survey similar in size and depth to the upcoming South Pole Telescope/Dark Energy Survey (SPT-DES). We analyze a model in which the strength of the non-Gaussianity is parameterized by the constant fNL; this model has been used extensively to derive Cosmic Microwave Background (CMB) anisotropy constraints on non-Gaussianity, allowing us to make contact with those works. We find that the constraining power of the cluster survey on dark energy observables is not significantly diminished by non-Gaussianity provided that cluster redshift information is included in the analysis. We also find that even an ideal cluster survey is unlikely to improve significantly current and future CMB constraints on non-Gaussianity. However, when all systematics are under control, it c

In this paper, the tracking performance limitation of networked control systems (NCSs) is studied. The NCSs is considered as continuous-time linear multi-input multi-output (MIMO) systems with random reference noises. The controlled plants include unstable poles and non-minimum phase (NMP) zeros. The output feedback path is affected by multiple communication constraints. We focus on some basic communication constraints, including additive white noise (AWN), quantization noise, bandwidth, as well as encoder-decoder. The system performance is evaluated with the tracking error energy, and used a two-degree of freedom (2DOF) controller. The explicit representation of the tracking performance is given in this paper. The results indicate the tracking performance limitations rely to internal characteristics of the plant (unstable poles and NMP zeros), reference noises (the reference noise power distribution (RNPD) and its directions) and the characteristics of communication constraints. Moreover, the tracking performance limitations are also affected by the angles between the each transform NMP zero direction and RNPD direction, and these angles between each transform unstable poles direc