Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: An aircraft has a fuselage, a wing assembly coupleable to the fuselage, and an empennage including a pair of tail booms configured to be removably coupled to the wing assembly. The wing assembly includes a pair of boom interfaces located on laterally opposite sides of the fuselage. Each tail boom has a boom forward end configured to be mechanically attached to one of the boom interfaces using an externally-accessible mechanical fastener.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. One of the methods includes, at each of a plurality of generation time steps: generating a combined sequence for the generation time step that includes the input sequence followed by the output tokens that have already been generated as of the generation time step; processing the combined sequence using a self-attention decoder neural network to generate a time step output that defines a score distribution over a set of possible output tokens; and selecting, using the time step output, an output token from the set of possible output tokens as the next output token in the output sequence.

Abstract: An aircraft has a fuselage, a wing assembly coupleable to the fuselage, and an empennage including a pair of tail booms configured to be removably coupled to the wing assembly. The wing assembly includes a pair of boom interfaces located on laterally opposite sides of the fuselage. Each tail boom has a boom forward end configured to be mechanically attached to one of the boom interfaces using an externally-accessible mechanical fastener.

Abstract: Aerial launch and/or recovery for unmanned aircraft, and associated systems and methods. A representative method for operating an unmanned aerial vehicle (UAV) system includes directing a first, multi-rotor carrier aircraft to carry a second, carried aircraft aloft, and release the second aircraft for flight, while powering the first aircraft with an on-board battery. The method can further include directing the first aircraft to position a capture line in a flight path of the second aircraft to capture the second aircraft.

Abstract: A computerized financial product sales system is disclosed. The system includes a customer profile calibration module, a product selection module, and a visualization module. The customer profile calibration module may be configured to allow selection of one or more customer parameters. The product selection module may be configured to allow selection of a financial product. The visualization module may be configured to graphically display an impact of the selected financial product on one or more of the selected customer parameters based on actual historical data over a specified time period.

Abstract: The present invention relates generally to a collapsible shipping container utilizing standard box making equipment manufactured from material such as paper, corrugated board, and corrugated plastic, and in particular, to a package that employs a tensioned film system, that holds a product hammock style, between two layers of film and/or between a film and a wall of the package. To use the box, the formed box which has been stored in a flat configuration, is unfolded to a substantially rectangular or square configuration by folding at least one flap panel about its fold line to a position superadjacent a central panel and conforming the container to a hexagonal cross-section. In this configuration the box retains its hexagonal cross-section and can be self supporting in a vertical position, in order to facilitate the filing operation.

Abstract: A retention package is formed from a tray member containing a retention member which includes a stiff board and a flexible, stretchable film attached to end panels of the stiff board. The end panels are folded under thus pulling the film taut for retaining an object between the film and the stiff board. The stiff board has two vertical walls that abut the interior of two walls of the tray member. An end cap is placed over the tray member thereby forming an enclosed retention package. The end cap is provided with kick plate means for holding the two vertical walls against the interior of two walls of the tray member.

Abstract: The present invention relates generally to a collapsible shipping container utilizing standard box making equipment manufactured from material such as paper, corrugated board, and corrugated plastic, and in particular, to a package that employs a tensioned film system, that holds a product hammock style, between two layers of film and/or between a film and a wall of the package. To use the box, the formed box which has been stored in a flat configuration, is unfolded to a substantially rectangular or square configuration by folding at least one flap panel about its fold line to a position superadjacent a central panel and conforming the container to a hexagonal cross-section. In this configuration the box retains its hexagonal cross-section and can be self-supporting in a vertical position, in order to facilitate the filing operation. The container has retaining means for securing a pair of plastic films at their ends to opposing sides of the container.

Abstract: The present invention provides a cargo shipping crate having a base, four sides and a top cap, a first panel and a second panel. The first panel and the second panel form four sides of the crate. Each panel has a first, a second and a third subpanel. The first subpanel of the first panel is positioned to overlap the third subpanel of the second panel, and the second subpanel of the first and the second panel have substantially identical dimensions. The first subpanel of the first and the second panel have substantially identical dimensions, and the third subpanel of the first and the second panel have substantially identical dimensions. The top cap has four flaps, each of the four flaps are rotatable to a position superjacent a subpanel.

Abstract: The disclosure indicates that the compression strength of a corrugated shipping crate can be increased through the use of horizontal panels adhered to the inside surface of the vertically fluted outer walls of the crate. The panels can comprise a layer of horizontally fluted single layer of double wall or triple wall corrugated, and a layer of vertically oriented double or triple wall corrugated. The crates' outer walls can be vertically fluted 1100 weight corrugated or less. The double wall vertically fluted corrugated, can be corrugated board have a weight in the range from 500 to 1100 pounds.

Abstract: The present invention provides a cargo shipping crate having a base, four sides and a top cap, a first panel and a second panel. The first panel and the second panel form four sides of the crate. Each panel has a first, a second and a third subpanel. The first subpanel of the first panel is positioned to overlap the third subpanel of the second panel, and the second subpanel of the first and the second panel have substantially identical dimensions. The first subpanel of the first and the second panel have substantially identical dimensions, and the third subpanel of the first and the second panel have substantially identical dimensions. The top cap has four flaps, each of the four flaps are rotatable to a position superjacent a subpanel.