Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium. In one aspect, a method includes the actions of receiving a request to perform computations for a neural network on a hardware circuit having a matrix computation unit, the request specifying a transpose operation to be performed on a first neural network matrix; and generating instructions that when executed by the hardware circuit cause the hardware circuit to transpose the first neural network matrix by performing first operations, wherein the first operations include repeatedly performing the following second operations: for a current subdivision of the first neural network matrix that divides the first neural network matrix into one or more current submatrices, updating the first neural network matrix by swapping an upper right quadrant and a lower left quadrant of each current submatrix, and subdividing each current submatrix into respective new submatrices to update the current subdivision.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium. In one aspect, a method includes the actions of receiving a request to perform computations for a neural network on a hardware circuit having a matrix computation unit, the request specifying a transpose operation to be performed on a first neural network matrix; and generating instructions that when executed by the hardware circuit cause the hardware circuit to transpose the first neural network matrix by performing first operations, wherein the first operations include repeatedly performing the following second operations: for a current subdivision of the first neural network matrix that divides the first neural network matrix into one or more current submatrices, updating the first neural network matrix by swapping an upper right quadrant and a lower left quadrant of each current submatrix, and subdividing each current submatrix into respective new submatrices to update the current subdivision.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for enabling a user to conduct a dialogue. Implementations of the system learn when to rely on supporting evidence, obtained from an external search system via a search system interface, and are also able to generate replies for the user that align with the preferences of a previously trained response selection neural network. Implementations of the system can also use a previously trained rule violation detection neural network to generate replies that take account of previously learnt rules.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for adjusting a target neural network using automatically generated test cases before deployment of the target neural network in a deployment environment. One of the methods may include generating a plurality of test inputs by using a test case generation neural network; processing the plurality of test inputs using a target neural network to generate one or more test outputs for each test input; and identifying, from the one or more test outputs generated by the target neural network for each test input, failing test inputs that result in generation of test outputs by the target neural network that fail one or more criteria.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium. In one aspect, a method includes the actions of receiving a request to perform computations for a neural network on a hardware circuit having a matrix computation unit, the request specifying a transpose operation to be performed on a first neural network matrix; and generating instructions that when executed by the hardware circuit cause the hardware circuit to transpose the first neural network matrix by performing first operations, wherein the first operations include repeatedly performing the following second operations: for a current subdivision of the first neural network matrix that divides the first neural network matrix into one or more current submatrices, updating the first neural network matrix by swapping an upper right quadrant and a lower left quadrant of each current submatrix, and subdividing each current submatrix into respective new submatrices to update the current subdivision.

Abstract: Embodiments disclosed herein provide systems and methods for preparing geometry for 3D printing. In one embodiment, a 3D printing preparation application receives 3D geometry and repairs non-manifold edges and non-manifold vertices, producing a topological manifold geometry. The 3D printing preparation application then welds coincident edges without coincident faces and fills holes in the geometry. The 3D printing preparation application may further perform resolution-aware thickening of the geometry by estimating distances to a medial axis based on distances to distance field shocks, and advecting the distance field using a velocity field. A similar approach may be used to perform resolution-aware separation enforcement. Alternatively, one component may be globally thickened and subtracted from another for separation enforcement. The 3D printing preparation application may also split large models and add connectors for connecting the split pieces after printing.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium. In one aspect, a method includes the actions of receiving a request to perform computations for a neural network on a hardware circuit having a matrix computation unit, the request specifying a transpose operation to be performed on a first neural network matrix; and generating instructions that when executed by the hardware circuit cause the hardware circuit to transpose the first neural network matrix by performing first operations, wherein the first operations include repeatedly performing the following second operations: for a current subdivision of the first neural network matrix that divides the first neural network matrix into one or more current submatrices, updating the first neural network matrix by swapping an upper right quadrant and a lower left quadrant of each current submatrix, and subdividing each current submatrix into respective new submatrices to update the current subdivision.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium. In one aspect, a method includes the actions of receiving a request to perform computations for a neural network on a hardware circuit having a matrix computation unit, the request specifying a transpose operation to be performed on a first neural network matrix; and generating instructions that when executed by the hardware circuit cause the hardware circuit to transpose the first neural network matrix by performing first operations, wherein the first operations include repeatedly performing the following second operations: for a current subdivision of the first neural network matrix that divides the first neural network matrix into one or more current submatrices, updating the first neural network matrix by swapping an upper right quadrant and a lower left quadrant of each current submatrix, and subdividing each current submatrix into respective new submatrices to update the current subdivision.

Abstract: Embodiments of the invention provide systems and methods for nesting objects in 2D sheets and 3D volumes. In one embodiment, a nesting application simplifies the shapes of parts and performs a rigid body simulation of the parts dropping into a 2D sheet or 3D volume. In the rigid body simulation, parts begin from random initial positions on one or more sides and drop under the force of gravity into the 2D sheet or 3D volume until coming into contact with another part, a boundary, or the origin of the gravity. The parts may be dropped according to a particular order, such as alternating large and small parts. Further, the simulation may be translation- and/or position-only, meaning the parts do not rotate and/or do not have momentum, respectively. Tighter packing may be achieved by incorporating user inputs and simulating jittering of the parts using random forces.

Abstract: Embodiments disclosed herein provide systems and methods for preparing geometry for 3D printing. In one embodiment, a 3D printing preparation application receives 3D geometry and repairs non-manifold edges and non-manifold vertices, producing a topological manifold geometry. The 3D printing preparation application then welds coincident edges without coincident faces and fills holes in the geometry. The 3D printing preparation application may further perform resolution-aware thickening of the geometry by estimating distances to a medial axis based on distances to distance field shocks, and advecting the distance field using a velocity field. A similar approach may be used to perform resolution-aware separation enforcement. Alternatively, one component may be globally thickened and subtracted from another for separation enforcement. The 3D printing preparation application may also split large models and add connectors for connecting the split pieces after printing.

Abstract: Embodiments disclosed herein provide systems and methods for preparing geometry for 3D printing. In one embodiment, a 3D printing preparation application receives 3D geometry and repairs non-manifold edges and non-manifold vertices, producing a topological manifold geometry. The 3D printing preparation application then welds coincident edges without coincident faces and fills holes in the geometry. The 3D printing preparation application may further perform resolution-aware thickening of the geometry by estimating distances to a medial axis based on distances to distance field shocks, and advecting the distance field using a velocity field. A similar approach may be used to perform resolution-aware separation enforcement. Alternatively, one component may be globally thickened and subtracted from another for separation enforcement. The 3D printing preparation application may also split large models and add connectors for connecting the split pieces after printing.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium. In one aspect, a method includes the actions of receiving a request to perform computations for a neural network on a hardware circuit having a matrix computation unit, the request specifying a transpose operation to be performed on a first neural network matrix; and generating instructions that when executed by the hardware circuit cause the hardware circuit to transpose the first neural network matrix by performing first operations, wherein the first operations include repeatedly performing the following second operations: for a current subdivision of the first neural network matrix that divides the first neural network matrix into one or more current submatrices, updating the first neural network matrix by swapping an upper right quadrant and a lower left quadrant of each current submatrix, and subdividing each current submatrix into respective new submatrices to update the current subdivision.

Abstract: Embodiments of the invention provide systems and methods for nesting objects in 2D sheets and 3D volumes. In one embodiment, a nesting application simplifies the shapes of parts and performs a rigid body simulation of the parts dropping into a 2D sheet or 3D volume. In the rigid body simulation, parts begin from random initial positions on one or more sides and drop under the force of gravity into the 2D sheet or 3D volume until coming into contact with another part, a boundary, or the origin of the gravity. The parts may be dropped according to a particular order, such as alternating large and small parts. Further, the simulation may be translation- and/or position-only, meaning the parts do not rotate and/or do not have momentum, respectively. Tighter packing may be achieved by incorporating user inputs and simulating jittering of the parts using random forces.

Abstract: Embodiments of the invention provide systems and methods for nesting objects in 2D sheets and 3D volumes. In one embodiment, a nesting application simplifies the shapes of parts and performs a rigid body simulation of the parts dropping into a 2D sheet or 3D volume. In the rigid body simulation, parts begin from random initial positions on one or more sides and drop under the force of gravity into the 2D sheet or 3D volume until coming into contact with another part, a boundary, or the origin of the gravity. The parts may be dropped according to a particular order, such as alternating large and small parts. Further, the simulation may be translation- and/or position-only, meaning the parts do not rotate and/or do not have momentum, respectively. Tighter packing may be achieved by incorporating user inputs and simulating jittering of the parts using random forces.

Abstract: Embodiments disclosed herein provide techniques for decomposing 3D geometry into developable surface patches and cut patterns. In one embodiment, a decomposition application receives a triangulated 3D surface as input and determines approximately developable surface patches from the 3D surface using a variant of k-means clustering. Such approximately developable surface patches may have undesirable jagged boundaries, which the decomposition application may eliminate by generating a data structure separate from the mesh that contains patch boundaries and optimizing the patch boundaries or, alternatively, remeshing the mesh such that patch boundaries fall on mesh edges. The decomposition application may then flatten the patches into truly developable surfaces by re-triangulating the patches as ruled surfaces. The decomposition application may further flatten the ruled surfaces into 2D shapes and lay those shapes out on virtual sheets of material.

Abstract: Embodiments of the invention provide systems and methods for nesting objects in 2D sheets and 3D volumes. In one embodiment, a nesting application simplifies the shapes of parts and performs a rigid body simulation of the parts dropping into a 2D sheet or 3D volume. In the rigid body simulation, parts begin from random initial positions on one or more sides and drop under the force of gravity into the 2D sheet or 3D volume until coming into contact with another part, a boundary, or the origin of the gravity. The parts may be dropped according to a particular order, such as alternating large and small parts. Further, the simulation may be translation- and/or position-only, meaning the parts do not rotate and/or do not have momentum, respectively. Tighter packing may be achieved by incorporating user inputs and simulating jittering of the parts using random forces.

Abstract: Embodiments of the invention provide systems and methods for nesting objects in 2D sheets and 3D volumes. In one embodiment, a nesting application simplifies the shapes of parts and performs a rigid body simulation of the parts dropping into a 2D sheet or 3D volume. In the rigid body simulation, parts begin from random initial positions on one or more sides and drop under the force of gravity into the 2D sheet or 3D volume until coming into contact with another part, a boundary, or the origin of the gravity. The parts may be dropped according to a particular order, such as alternating large and small parts. Further, the simulation may be translation- and/or position-only, meaning the parts do not rotate and/or do not have momentum, respectively. Tighter packing may be achieved by incorporating user inputs and simulating jittering of the parts using random forces.

Abstract: Embodiments of the invention provide systems and methods for nesting objects in 2D sheets and 3D volumes. In one embodiment, a nesting application simplifies the shapes of parts and performs a rigid body simulation of the parts dropping into a 2D sheet or 3D volume. In the rigid body simulation, parts begin from random initial positions on one or more sides and drop under the force of gravity into the 2D sheet or 3D volume until coming into contact with another part, a boundary, or the origin of the gravity. The parts may be dropped according to a particular order, such as alternating large and small parts. Further, the simulation may be translation- and/or position-only, meaning the parts do not rotate and/or do not have momentum, respectively. Tighter packing may be achieved by incorporating user inputs and simulating jittering of the parts using random forces.

Abstract: Embodiments disclosed herein provide techniques for decomposing 3D geometry into developable surface patches and cut patterns. In one embodiment, a decomposition application receives a triangulated 3D surface as input and determines approximately developable surface patches from the 3D surface using a variant of k-means clustering. Such approximately developable surface patches may have undesirable jagged boundaries, which the decomposition application may eliminate by generating a data structure separate from the mesh that contains patch boundaries and optimizing the patch boundaries or, alternatively, remeshing the mesh such that patch boundaries fall on mesh edges. The decomposition application may then flatten the patches into truly developable surfaces by re-triangulating the patches as ruled surfaces. The decomposition application may further flatten the ruled surfaces into 2D shapes and lay those shapes out on virtual sheets of material.

Abstract: Embodiments disclosed herein provide systems and methods for preparing geometry for 3D printing. In one embodiment, a 3D printing preparation application receives 3D geometry and repairs non-manifold edges and non-manifold vertices, producing a topological manifold geometry. The 3D printing preparation application then welds coincident edges without coincident faces and fills holes in the geometry. The 3D printing preparation application may further perform resolution-aware thickening of the geometry by estimating distances to a medial axis based on distances to distance field shocks, and advecting the distance field using a velocity field. A similar approach may be used to perform resolution-aware separation enforcement. Alternatively, one component may be globally thickened and subtracted from another for separation enforcement. The 3D printing preparation application may also split large models and add connectors for connecting the split pieces after printing.