Abstract: A broadband resonator for a fuel cell compressor is disclosed. The resonator having a resonator insert having a tubular pipe surrounded by a plurality of disc-shaped walls separating the resonator insert receiving chamber into one or more individual resonator chambers. The tubular pipe is positioned eccentrically within a resonator insert receiving chamber. An entrained water removal system is formed in the resonator, preventing entrained water from accumulating in the resonator chambers which would result in an undesirable detuning of the designed amplitude frequency spectrum response.

Abstract: A broadband resonator for a fuel cell compressor is disclosed. The resonator having a resonator insert having a tubular pipe surrounded by a plurality of disc-shaped walls separating the resonator insert receiving chamber into one or more individual resonator chambers. The tubular pipe is positioned eccentrically within a resonator insert receiving chamber. An entrained water removal system is formed in the resonator, preventing entrained water from accumulating in the resonator chambers which would result in an undesirable detuning of the designed amplitude frequency spectrum response.

Abstract: Embodiments described herein provide a reinforcement learning based framework engaging pretrained language models (LMs) for program synthesis tasks. Specifically, the framework adopts a training strategy that optimizes pretrained LMs for program synthesis tasks in an actor-critic approach.

Abstract: Embodiments described herein provide a reinforcement learning based framework engaging pretrained language models (LMs) for program synthesis tasks. Specifically, the framework adopts a training strategy that optimizes pretrained LMs for program synthesis tasks in an actor-critic approach.

Abstract: A method for dialog state tracking includes decoding, by a fertility decoder, encoded dialog information associated with a dialog to generate fertilities for generating dialog states of the dialog. Each dialog state includes one or more domains. Each domain includes one or more slots. Each slot includes one or more slot tokens. The method further includes generating an input sequence to a state decoder based on the fertilities. A total number of each slot token in the input sequence is based on a corresponding fertility. The method further includes encoding, by a state encoder, the input sequence to the state decoder, and decoding, by the state decoder, the encoded input sequence to generate a complete sequence of the dialog states.

Abstract: A system and method for generating a response in a video grounded dialogue are provided. A video-grounded dialogue neural network language model receives video input and text input. The text input includes a dialogue history between the model and a human user and a current utterance by the user. Encoded video input is generated using video encoding layers. Encoded text input is generated using text encoding layers. The encoded video input and the encoded text input are concatenated in to a single input sequence. A generative pre-trained transformer model generates the response to the current utterance from the singe input sequence.

Abstract: A piezoelectric bimorph cantilever beam system includes a shim having a first main surface, a second main surface opposite the first main surface, a proximal end connected to an anchor, and a distal end opposite the proximal end. The system further includes a first piezoelectric layer laminated on the first main surface of the shim and a second piezoelectric layer laminated on the second main surface of the shim. A first beam stiffener is provided over the first main surface of the shim adjacent to the anchor with the first beam stiffener at least partially covering the first piezoelectric layer. A second beam stiffener is provided over the second main surface of the shim adjacent to the anchor with the second beam stiffener at least partially covering the second piezoelectric layer.

Abstract: Systems and methods are provided for performing a video-grounded dialogue task by a neural network model using bi-directional spatial-temporal reasoning. According to some embodiments, the systems and methods implement a dual network architecture or framework. This framework includes one network or reasoning module that learns dependencies between text and video in the direction of spatial?temporal, and another network or reasoning module that learns in the direction of temporal?spatial. The output of the multimodal reasoning modules may be combined to learn dependencies between language features in dialogues. The result joint representation is used as a contextual feature to the decoding components which allow the model to semantically generate meaningful responses to the users. In some embodiments, pointer networks are extended to the video-grounded dialogue task to allow the model to point to specific tokens from multiple source sequences to generate responses.

Abstract: A system and method for generating a response in a video grounded dialogue are provided. A video-grounded dialogue neural network language model receives video input and text input. The text input includes a dialogue history between the model and a human user and a current utterance by the user. Encoded video input is generated using video encoding layers. Encoded text input is generated using text encoding layers. The encoded video input and the encoded text input are concatenated in to a single input sequence. A generative pre-trained transformer model generates the response to the current utterance from the singe input sequence.

Abstract: Systems and methods are provided for performing a video-grounded dialogue task by a neural network model using bi-directional spatial-temporal reasoning. According to some embodiments, the systems and methods implement a dual network architecture or framework. This framework includes one network or reasoning module that learns dependencies between text and video in the direction of spatial?temporal, and another network or reasoning module that learns in the direction of temporal?spatial. The output of the multimodal reasoning modules may be combined to learn dependencies between language features in dialogues. The result joint representation is used as a contextual feature to the decoding components which allow the model to semantically generate meaningful responses to the users. In some embodiments, pointer networks are extended to the video-grounded dialogue task to allow the model to point to specific tokens from multiple source sequences to generate responses.

Abstract: A method for dialog state tracking includes decoding, by a fertility decoder, encoded dialog information associated with a dialog to generate fertilities for generating dialog states of the dialog. Each dialog state includes one or more domains. Each domain includes one or more slots. Each slot includes one or more slot tokens. The method further includes generating an input sequence to a state decoder based on the fertilities. A total number of each slot token in the input sequence is based on a corresponding fertility. The method further includes encoding, by a state encoder, the input sequence to the state decoder, and decoding, by the state decoder, the encoded input sequence to generate a complete sequence of the dialog states.

Abstract: A computer system includes a dispatch routing network to dispatch a plurality of instructions, and a processor in signal communication with the dispatch routing network. The processor determines a move instruction from the plurality of instructions to move data produced by an older second instruction, and copies a splice target file (STF) tag from a source register of the move instruction to a destination register of the move instruction without physically copying data in a slice target register and without assigning a new STF tag destination to the move instruction.

Abstract: Systems, methods, and computer-readable media are described for performing speculative execution of both paths/branches of a weakly predicted branch instruction. A branch instruction may be fetched from an instruction queue and determined to be a weakly predicted branch instruction, in which case, both paths of the branch instruction may be dispatched and speculatively executed. When the actual path taken becomes known, instructions corresponding to the path not taken may be flushed. Instructions from both paths of a weakly predicted branch instruction that are speculatively executed may be dispatch and executed in an interleaved manner.

Abstract: Systems, methods, and computer-readable media are described for performing instruction execution using an instruction completion table (ICT) that is configured to accommodate shared ICT entries. A shared ICT entry maps to multiple instructions such as, for example, two instructions. Each shared ICT entry may be referenced by an even instruction tag (ITAG) and an odd ITAG that correspond to respective instructions that have been grouped together in the shared ICT entry. The instructions corresponding to a given shared ICT entry can be executed and finished independently of one another. A shared ICT entry is completed when each execution of each instruction corresponding to the shared ICT entry has finished and when all prior ICT entries have completed. Also described herein are system, methods, and computer-readable media for flushing instructions in shared ICT entries in response to execution of a branch instruction.

Abstract: A piezoelectric bimorph cantilever beam system includes a shim having a first main surface, a second main surface opposite the first main surface, a proximal end connected to an anchor, and a distal end opposite the proximal end. The system further includes a first piezoelectric layer laminated on the first main surface of the shim and a second piezoelectric layer laminated on the second main surface of the shim. A first beam stiffener is provided over the first main surface of the shim adjacent to the anchor with the first beam stiffener at least partially covering the first piezoelectric layer. A second beam stiffener is provided over the second main surface of the shim adjacent to the anchor with the second beam stiffener at least partially covering the second piezoelectric layer.

Abstract: A computer system includes a dispatch routing network to dispatch a plurality of instructions, and a processor in signal communication with the dispatch routing network. The processor determines a move instruction from the plurality of instructions to move data produced by an older second instruction, and copies a splice target file (STF) tag from a source register of the move instruction to a destination register of the move instruction without physically copying data in a slice target register and without assigning a new STF tag destination to the move instruction.

Abstract: Examples are described for mapping primitives in circular images, generated from texture coordinate that compensate for rotation of cameras, onto respective mesh models to generate images representative of a 360-degree video or image. Portions of the rectangular images are blended to generate a stitched rectangular image, and image content for display is generated based on the stitched rectangular image.

Abstract: Systems, methods, and computer-readable media are described for performing speculative execution of both paths/branches of a weakly predicted branch instruction. A branch instruction may be fetched from an instruction queue and determined to be a weakly predicted branch instruction, in which case, both paths of the branch instruction may be dispatched and speculatively executed. When the actual path taken becomes known, instructions corresponding to the path not taken may be flushed. Instructions from both paths of a weakly predicted branch instruction that are speculatively executed may be dispatch and executed in an interleaved manner.

Abstract: Systems, methods, and computer-readable media are described for performing instruction execution using an instruction completion table (ICT) that is configured to accommodate shared ICT entries. A shared ICT entry maps to multiple instructions such as, for example, two instructions. Each shared ICT entry may be referenced by an even instruction tag (ITAG) and an odd ITAG that correspond to respective instructions that have been grouped together in the shared ICT entry. The instructions corresponding to a given shared ICT entry can be executed and finished independently of one another. A shared ICT entry is completed when each execution of each instruction corresponding to the shared ICT entry has finished and when all prior ICT entries have completed. Also described herein are system, methods, and computer-readable media for flushing instructions in shared ICT entries in response to execution of a branch instruction.

Abstract: Techniques are described related to generating image content. A graphics processing unit (GPU) is configured to receive a first set of images generated from a first camera device in a first location, the first camera device having a first orientation, render for display the first set of images oriented to an orientation reference, receive a second, different set of images generated from a second, different camera device in a second, different location, the second camera device having a second orientation, the second orientation being different than the first orientation, and render for display the second set of images oriented to the orientation reference.