Abstract: Systems and methods are provided for implementing a machine learning approach to modeling entity behavior. Fixed information and periodically updated information may be utilized to predict the behavior of an entity. By incorporating periodically updated information, the system is able to maintain an up-to-date prediction of each entity's behavior, while also accounting for entity action with respect to ongoing obligations. The system may generate behavior scores for the set of entities. In some embodiments, the behavior scores that are generated may indicate the transactional risk associated with each entity. Using the behavior scores generated, a user may be able to assess the credit riskiness of individual entities and instruct one or more individuals assigned to the entities to take one or more actions based on the credit riskiness of the individual entities.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a policy neural network used to select actions to be performed by a reinforcement learning agent interacting with an environment. In one aspect, a method includes obtaining path data defining a path through the environment traversed by the agent. A consistency error is determined for the path from a combined reward, first and last soft-max state values, and a path likelihood. A value update for the current values of the policy neural network parameters is determined from at least the consistency error. The value update is used to adjust the current values of the policy neural network parameters.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a policy neural network used to select actions to be performed by a reinforcement learning agent interacting with an environment. In one aspect, a method includes obtaining path data defining a path through the environment traversed by the agent. A consistency error is determined for the path from a combined reward, first and last soft-max state values, and a path likelihood. A value update for the current values of the policy neural network parameters is determined from at least the consistency error. The value update is used to adjust the current values of the policy neural network parameters.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a policy neural network used to select actions to be performed by a reinforcement learning agent interacting with an environment. In one aspect, a method includes obtaining path data defining a path through the environment traversed by the agent. A consistency error is determined for the path from a combined reward, first and last soft-max state values, and a path likelihood. A value update for the current values of the policy neural network parameters is determined from at least the consistency error. The value update is used to adjust the current values of the policy neural network parameters.

Abstract: The invention relates to an arrangement of one or more cryogenic tank containers for storing liquefied natural gas (LNG) or compressed natural gas (CNG) on an open deck of a tugboat. The one or more cryogenic tank containers are removably mounted on and at an aft deck area of the tugboat and are provided with means for supplying LNG or CNG to the engine of the tugboat.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a policy neural network used to select actions to be performed by a reinforcement learning agent interacting with an environment. In one aspect, a method includes obtaining path data defining a path through the environment traversed by the agent. A consistency error is determined for the path from a combined reward, first and last soft-max state values, and a path likelihood. A value update for the current values of the policy neural network parameters is determined from at least the consistency error. The value update is used to adjust the current values of the policy neural network parameters.

Abstract: The invention relates to an arrangement of one or more cryogenic tank containers for storing liquefied natural gas (LNG) or compressed natural gas (CNG) on an open deck of a tugboat. The one or more cryogenic tank containers are removably mounted on and at an aft deck area of the tugboat and are provided with means for supplying LNG or CNG to the engine of the tugboat.

Abstract: A method includes providing a layer having a plurality of spaced-apart lines of a first material extending along a first direction and forming a line of a second material on opposing surfaces of each line of the first material, the first and second materials being different and adjacent lines of the second material being discontinuous. After forming the lines of the second material, forming pairs of spaced-apart lines of a third material between adjacent pairs of the lines of the second material, wherein each line of the third material is spaced apart from the closest line of the second material and the first and third materials are different.

Abstract: In general, in a first aspect, the invent features a method, including providing a layer having a plurality of spaced-apart lines of a first material extending along a first direction and forming a line of a second material on opposing surfaces of each line of the first material, the first and second materials being different and adjacent lines of the second material being discontinuous. After forming the lines of the second material, forming pairs of spaced-apart lines of a third material between adjacent pairs of the lines of the second material, wherein each line of the third material is spaced apart from the closest line of the second material and the first and third materials are different.