Machine translation of low-resource languages(LRLs)has long been hindered by limited corpora and linguistic complexity.This review summarizes key developments,from traditional methods to recent progress with large lan...Machine translation of low-resource languages(LRLs)has long been hindered by limited corpora and linguistic complexity.This review summarizes key developments,from traditional methods to recent progress with large language models(LLMs),while highlighting ongoing challenges such as data bottlenecks,biases,fairness,and computational costs.Finally,it discusses future directions,including efficient parameter fine-tuning,multimodal translation,and community-driven corpus construction,providing insights for advancing LRL translation research.展开更多
We show first that an orbit, which is naturally characterized by its eccentricity and semi-latus rectum, can equally be characterized by other sets of parameters, and proceed to determine mass-independent characteriza...We show first that an orbit, which is naturally characterized by its eccentricity and semi-latus rectum, can equally be characterized by other sets of parameters, and proceed to determine mass-independent characterizations. The latter is employed to obtain the laws of equivalent orbits, which by definition have the same eccentricity and orbit’s parameter [1]. These laws relate the values of the same physical observables on two equivalent orbits to the corresponding total mass;they include the laws of velocity, angular velocity, radial velocity, areal velocity, acceleration, period, energy and angular momentum. Regardless of the share of the two bodies of a fixed total mass, the same relative orbit occurs for the same initial conditions. Moreover, the same orbit can be traced by different total masses but with different relative velocities. The concept of a gravitational field generated by a set of masses is shown to be meaningful only when the center of mass is not changed by the test mass. The associated concept of the “nothing”, which is an infinitesimal mass that allows for the property just mentioned to be fulfilled, is introduced and its orbits are determined. The perturbation of the nothing orbits due to its replacement by a finite mass is determined. It is proved that such a replacement can have a qualitative effect resulting in a “phase transition” of an orbit from unbound to bound, and that the nothing’s circular orbits cannot be occupied by any material body. The Galileo law of free fall, on which the equivalence principle hinges and which is exact only for “nothing-like” falling objects, is revised to determine the duration of free fall of a body of an arbitrary mass. The wholeness of Newton’s laws and the associated concept of force as an interaction are highlighted, and some contradictions between the Newtonian laws of equivalent Kepler’s orbits and the general relativistic predictions are discussed. It is demonstrated that Newton’s law of gravitation is not an approximation of Einstein field Equations even in the case of a static weak field. However, both theories have a common limit corresponding to the case in which the alien concept of a field can be incorporated in the Newtonian theory. We also show that the relative velocity’s hodograph [2-4], the alternative Laplace-Runge-Lenz (LRL) vector derived by Hamilton [4-6], as well as an infinite set of LRL vectors, result all from one vector. The hodograph is a proper circular arc for hyperbolic motion, a circle less a point for parabolic motion, and a full circle for bound motion.展开更多
在最小二乘方法(RLS,recursive least square)的基础上,提出利用格型递归最小二乘(LRLS,lattice recursiveleast square)算法对AR模型参数进行自适应估计。该算法为模块式的多极格型结构,降低了一般RLS算法的计算复杂度。利用实测的动...在最小二乘方法(RLS,recursive least square)的基础上,提出利用格型递归最小二乘(LRLS,lattice recursiveleast square)算法对AR模型参数进行自适应估计。该算法为模块式的多极格型结构,降低了一般RLS算法的计算复杂度。利用实测的动态数据结合AIC准则建立自适应AR预报模型,并将该模型应用于船舶运动预报中,仿真结果表明,相对于最小二乘算法,基于LRLS算法的AR预报模型可有效提高船舶运动预报精度。展开更多
基金supported by China Undergraduate Innovation Training Program[Grant No.202410699184]Humanities and Social Sciences Research Project funded by the Ministry of Education of China[Grant No.23YJAZH139].
文摘Machine translation of low-resource languages(LRLs)has long been hindered by limited corpora and linguistic complexity.This review summarizes key developments,from traditional methods to recent progress with large language models(LLMs),while highlighting ongoing challenges such as data bottlenecks,biases,fairness,and computational costs.Finally,it discusses future directions,including efficient parameter fine-tuning,multimodal translation,and community-driven corpus construction,providing insights for advancing LRL translation research.
文摘We show first that an orbit, which is naturally characterized by its eccentricity and semi-latus rectum, can equally be characterized by other sets of parameters, and proceed to determine mass-independent characterizations. The latter is employed to obtain the laws of equivalent orbits, which by definition have the same eccentricity and orbit’s parameter [1]. These laws relate the values of the same physical observables on two equivalent orbits to the corresponding total mass;they include the laws of velocity, angular velocity, radial velocity, areal velocity, acceleration, period, energy and angular momentum. Regardless of the share of the two bodies of a fixed total mass, the same relative orbit occurs for the same initial conditions. Moreover, the same orbit can be traced by different total masses but with different relative velocities. The concept of a gravitational field generated by a set of masses is shown to be meaningful only when the center of mass is not changed by the test mass. The associated concept of the “nothing”, which is an infinitesimal mass that allows for the property just mentioned to be fulfilled, is introduced and its orbits are determined. The perturbation of the nothing orbits due to its replacement by a finite mass is determined. It is proved that such a replacement can have a qualitative effect resulting in a “phase transition” of an orbit from unbound to bound, and that the nothing’s circular orbits cannot be occupied by any material body. The Galileo law of free fall, on which the equivalence principle hinges and which is exact only for “nothing-like” falling objects, is revised to determine the duration of free fall of a body of an arbitrary mass. The wholeness of Newton’s laws and the associated concept of force as an interaction are highlighted, and some contradictions between the Newtonian laws of equivalent Kepler’s orbits and the general relativistic predictions are discussed. It is demonstrated that Newton’s law of gravitation is not an approximation of Einstein field Equations even in the case of a static weak field. However, both theories have a common limit corresponding to the case in which the alien concept of a field can be incorporated in the Newtonian theory. We also show that the relative velocity’s hodograph [2-4], the alternative Laplace-Runge-Lenz (LRL) vector derived by Hamilton [4-6], as well as an infinite set of LRL vectors, result all from one vector. The hodograph is a proper circular arc for hyperbolic motion, a circle less a point for parabolic motion, and a full circle for bound motion.
文摘在最小二乘方法(RLS,recursive least square)的基础上,提出利用格型递归最小二乘(LRLS,lattice recursiveleast square)算法对AR模型参数进行自适应估计。该算法为模块式的多极格型结构,降低了一般RLS算法的计算复杂度。利用实测的动态数据结合AIC准则建立自适应AR预报模型,并将该模型应用于船舶运动预报中,仿真结果表明,相对于最小二乘算法,基于LRLS算法的AR预报模型可有效提高船舶运动预报精度。
