Microsoft Technology Licensing, LLC

Cambia Health Solutions, Inc.

Nuance Communications, Inc.

Xiang Li

David McCarthy Peixotto

Ivan Nevraev

Michael Alan Dougherty

Marcelo Lopez Ruiz

Lin Chen

Weicheng Ma

Kai Cao

Bei Pan

Alejandro Acero

Dong Yu

Li Deng

David M Lubensky

Ea-Ee Jan

Osamuyimen Stewart

David Mccarthy Peixotto

Methods and systems are provided for receiving an input query at a Variational-sequence-to-sequence dialog generator (VSDG) of a chatbot, and calculating, via a variational autoencoder (VAE) combined with a generative adversarial network (GAN) of the VSDG, a response to the input query. The response may be in a dialog form. Further, in one or more examples, the GAN evaluates the response for updating the VSDG.

Systems and methods for continual updating of response generation by an artificial intelligence chatbot

Methods and systems are provided for a natural language processing system comprising a chatbot adapted for dialog generation. In one example, the system may include a combination of a variational autoencoder (VAE) and a generative adversarial network (GAN) for generating natural responses to input queries. The VAE may convert queries into vector embeddings that may then be used by the GAN to continuously update and improve responses provided by the chatbot.

A computer device is provided that includes a processor configured to receive a source code for a program including at least two code files, and process the source code for the program to generate a machine-level code file for each of the at least two code files of the source code. The processor is further configured to generate control flow graph data for each machine-level code file generated for the at least two code files of the source code, generate a machine-level intermediate representation for each machine-level code file using a machine-level code file and the generated control flow graph data for that machine-level code file, merge the machine-level intermediate representations into a merged machine-level intermediate representation, and perform machine-level optimizations on the merged machine-level intermediate representation and output an optimized merged machine-level intermediate representation.

Merged machine-level intermediate representation optimizations

The present disclosure relates to devices and methods for transforming program source code using a rematerialization operation. The devices and methods may identify at least one hot spot with high register pressure in a program source code for an application and identify a plurality of live variables within the at least one hot spot. The devices and methods may group the plurality of live variables by a basic block that has contained a define or single use of the plurality of live variables. The devices and methods may build a directed acyclic graph (DAG) for each basic block that has a grouped plurality of live variables. The devices and methods may save the DAG as a candidate instruction to move in the program source code and may generate transformed program source code for the application by moving the candidate instruction.

Compiler sub expression directed acyclic graph (DAG) remat for register pressure

Examples are disclosed herein that relate to performing rematerialization operation(s) on program source code prior to instruction scheduling. In one example, a method includes prior to performing instruction scheduling on program source code, for each basic block of the program source code, determining a register pressure at a boundary of the basic block, determining whether the register pressure at the boundary is greater than a target register pressure, based on the register pressure at the boundary being greater than the target register pressure, identifying one or more candidate instructions in the basic block suitable for rematerialization to reduce the register pressure at the boundary, and performing a rematerialization operation on at least one of the one or more candidate instructions to reduce the register pressure at the boundary to be less than the target register pressure.

Pre-instruction scheduling rematerialization for register pressure reduction

Examples described herein generally relate to generating, from a listing of source code, a plurality of basic blocks for compiling into intermediate language, determining, for a first basic block of the plurality of basic blocks, first heuristics related to applying a first plurality of optimizations to the first basic block, determining, for a second basic block of the plurality of basic blocks, second heuristics related to applying a second plurality of optimizations to the second basic block, and applying, based on the first heuristics and the second heuristics, one of the first plurality of optimizations to the first basic block to schedule first instructions for the first basic block and one of the second plurality of optimizations to the second basic block to schedule second instructions for the second basic block.

Techniques for scheduling instructions in compiling source code

Methods and devices for generating program code representations may include receiving program code or edited program code for an application executing on the computer device. The methods and devices may include receiving an identification of a selected pipeline from a plurality of pipelines that defines a plurality of passes of actions to execute on the program code or the edited program code to optimize the program code or the edited program code. The methods and devices may include running the selected pipeline and generate optimizer output with a program code representation of the program code.

Interactive code optimizer

Techniques are described for semantically comparing machine code traces generated by compilers that compile computer software code. For example, a trace of machine code generated by a compiler can be obtained. The trace can be transformed into a set of expressions in a uniform expression format (e.g., by performing translation of the trace instructions into corresponding expressions and/or by performing other transformations). The set of expressions in the uniform expression format can be compared to other sets of expressions in the uniform expression format (e.g., other sets of expressions created from traces of machine code from other compilers). Results of the comparison can comprise indications of whether the sets of expressions match.

Semantic comparison of computer compiler traces

A method of processing limited natural language data to automatically develop an optimal feature set, bypassing the standard Wizard of OZ (WOZ) approach. Natural language understanding models process existing data from other domains, such as the Internet, for domain-specific adaptation through the use of an optimal feature set. When the optimal feature set is passed on to any engine, the optimal feature set produces robust models that can be used for natural language call routing.

Systems and methods for building asset based natural language call routing application with limited resources

A method of producing at least one possible sequence of vocal tract resonance (VTR) for a fixed sequence of phonetic units, and producing the acoustic observation probability by integrating over such distributions is provided. The method includes identifying a sequence of target distributions for a VTR sequence corresponding to a phone sequence with a given segmentation. The sequence of target distributions is applied to a finite impulse response filter to produce distributions for possible VTR trajectories. Then these distributions are applied to a linearized nonlinear function to produce the acoustic observation probability for the given sequence of phonetic units. This acoustic observation probability is used for phonetic recognition.

Growing community of inventors

Bellevue, WA, United States of America

Xiang Li