Abstract: The present disclosure is generally directed to methods, apparatus, and computer-readable media (transitory and non-transitory) for learning to automatically navigate interactive web documents and/or websites. More particularly, various approaches are presented for training various deep Q network (DQN) agents to perform various tasks associated with reinforcement learning, including hierarchical reinforcement learning, in challenging web navigation environments with sparse rewards and large state and action spaces. These agents include a web navigation agent that can use learned value function(s) to automatically navigate through interactive web documents, as well as a training agent, referred to herein as a “meta-trainer,” that can be trained to generate synthetic training examples. Some approaches described herein may be implemented when expert demonstrations are available. Other approaches described herein may be implemented when expert demonstrations are not available.

Abstract: The present disclosure is generally directed to methods, apparatus, and computer-readable media (transitory and non-transitory) for learning to automatically navigate interactive web documents and/or websites. More particularly, various approaches are presented for training various deep Q network (DQN) agents to perform various tasks associated with reinforcement learning, including hierarchical reinforcement learning, in challenging web navigation environments with sparse rewards and large state and action spaces. These agents include a web navigation agent that can use learned value function(s) to automatically navigate through interactive web documents, as well as a training agent, referred to herein as a “meta-trainer,” that can be trained to generate synthetic training examples. Some approaches described herein may be implemented when expert demonstrations are available. Other approaches described herein may be implemented when expert demonstrations are not available.

Abstract: The present disclosure is generally directed to methods, apparatus, and computer-readable media (transitory and non-transitory) for learning to automatically navigate interactive web documents and/or websites. More particularly, various approaches are presented for training various deep Q network (DQN) agents to perform various tasks associated with reinforcement learning, including hierarchical reinforcement learning, in challenging web navigation environments with sparse rewards and large state and action spaces. These agents include a web navigation agent that can use learned value function(s) to automatically navigate through interactive web documents, as well as a training agent, referred to herein as a “meta-trainer,” that can be trained to generate synthetic training examples. Some approaches described herein may be implemented when expert demonstrations are available. Other approaches described herein may be implemented when expert demonstrations are not available.

Abstract: The present disclosure is generally directed to methods, apparatus, and computer-readable media (transitory and non-transitory) for learning to automatically navigate interactive web documents and/or websites. More particularly, various approaches are presented for training various deep Q network (DQN) agents to perform various tasks associated with reinforcement learning, including hierarchical reinforcement learning, in challenging web navigation environments with sparse rewards and large state and action spaces. These agents include a web navigation agent that can use learned value function(s) to automatically navigate through interactive web documents, as well as a training agent, referred to herein as a “meta-trainer,” that can be trained to generate synthetic training examples. Some approaches described herein may be implemented when expert demonstrations are available. Other approaches described herein may be implemented when expert demonstrations are not available.

Abstract: The present disclosure is generally directed to methods, apparatus, and computer-readable media (transitory and non-transitory) for learning to automatically navigate interactive web documents and/or websites. More particularly, various approaches are presented for training various deep Q network (DQN) agents to perform various tasks associated with reinforcement learning, including hierarchical reinforcement learning, in challenging web navigation environments with sparse rewards and large state and action spaces. These agents include a web navigation agent that can use learned value function(s) to automatically navigate through interactive web documents, as well as a training agent, referred to herein as a “meta-trainer,” that can be trained to generate synthetic training examples. Some approaches described herein may be implemented when expert demonstrations are available. Other approaches described herein may be implemented when expert demonstrations are not available.

Abstract: The turbomachine comprises a casing (26), there being arranged in the casing an impeller (23) arranged on a driven shaft. The turbomachine has an inflow region (21) and an outflow region (22) and, in operation, is flowed through by a working medium. The working medium flows into the inflow region, along a front side (23a) formed on the impeller and subsequently out of the outflow region, there being a pressure drop at the front side between the inflow region and the outflow region. A pressure divider (9) is arranged on the rear side of the impeller, opposite the front side.

Abstract: The invention relates to a device and a method for separating dirt particles from the working medium of a turbine (10). The turbine (10) comprises at least one rotor (11) which is arranged in a housing (17). A swirl generator (20) is provided that sets the working medium and the dirt particles in a spiral-shaped rotational movement along a principal axis (22) by means of the geometry of the swirl generator (20) and thereby separates the dirt particles from the working medium. The swirl generator (20) is designed in such a way that the working medium experiences a reversal of the speed component parallel to the principal axis (22) within the swirl generator (20).

Abstract: The turbomachine comprises a casing (26), there being arranged in the casing an impeller (23) arranged on a driven shaft. The turbomachine has an inflow region (21) and an outflow region (22) and, in operation, is flowed through by a working medium. The working medium flows into the inflow region, along a front side (23a) formed on the impeller and subsequently out of the outflow region, there being a pressure drop at the front side between the inflow region and the outflow region. A pressure divider (9) is arranged on the rear side of the impeller, opposite the front side.

Abstract: The invention relates to a device and a method for separating dirt particles from the working medium of a turbine (10). The turbine (10) comprises at least one rotor (11) which is arranged in a housing (17). A swirl generator (20) is provided that sets the working medium and the dirt particles in a spiral-shaped rotational movement along a principal axis (22) by means of the geometry of the swirl generator (20) and thereby separates the dirt particles from the working medium. The swirl generator (20) is designed in such a way that the working medium experiences a reversal of the speed component parallel to the principal axis (22) within the swirl generator (20).

Abstract: The invention relates to a method of conditioning a condensate generated in the compression section of a gas purification unit. The invention also relates to system for this method.

Abstract: The invention relates to a steam turbine 5, in particular for utilizing the waste heat of an internal combustion engine, having at least one turbine housing, having a guide wheel 10 that has at least one nozzle, and having at least one rotor 6, wherein the nozzle is in the form of a duct 11 formed into the guide wheel 10. According to the invention, a steam turbine 5 is provided, the nozzles of which can be manufactured in a simple manner. This is achieved by virtue of the fact that the duct 11 has a constant width B and has a depth T that varies along the duct 11. In this way, the duct 11 can be produced by means of a single tool in a single working operation.

Abstract: The invention relates to a method of conditioning a condensate generated in the compression section of a gas purification unit. The invention also relates to system for this method.