Petuum Inc.

Carnegie Mellon University

Opro.ai Inc.

Eric Xing

Wei Dai

Pengtao Xie

Qirong Ho

Xiaodan Liang

Hao Zhang

Weiren Yu

Aurick Qiao

Joseph Doyle

Roey Flor

Zhiting Hu

Chenyu Wang

Bin Zhao

Adam Schwab

Dian Li

Hongbao Zhang

The current disclosure is directed towards system and method for controlling industrial process. In one example, a method comprising deploying a forecast model for controlling an industrial process with training configurations that can be used as a single point of truth for guiding training and retraining versions of the forecast model using a model training algorithm without human input. The retraining and redeployment of the forecast model may be triggered when the performance of the forecast model degrades.

Autonomous industrial process control system and method that provides autonomous retraining of forecast model

An unsupervised real to virtual domain unification model for highway driving, or DU-drive, employs a conditional generative adversarial network to transform driving images in a real domain to their canonical representations in the virtual domain, from which vehicle control commands are predicted. In the case where there are multiple real datasets, a real-to-virtual generator may be independently trained for each real domain and a global predictor could be trained with data from multiple real domains. Qualitative experiment results show this model can effectively transform real images to the virtual domain while only keeping the minimal sufficient information, and quantitative results verify that such canonical representation can eliminate domain shift and boost the performance of control command prediction task.

Unsupervised real-to-virtual domain unification for end-to-end highway driving

Methods and systems are presented for consuming different data sources, and deploying artificial intelligence and machine learning programs on different target devices or infrastructures. Many data types can be transformed into machine learning data shards (MLDS) while many machine learning programs written in various programming languages or frameworks are transformed to common operator representations. Operator representations are transformed into execution graphs (EG) for a chosen target device or infrastructure. The MLDS and EG are input to the targeted devices and infrastructures, which then execute the machine learning programs (now transformed to EGs) on the MLDS to produce trained models or predictions with trained models.

System and methods for distributed machine learning with multiple data sources, multiple programming languages or frameworks, and multiple devices or infrastructures

Accordingly, a data engineering system for machine learning at scale is disclosed. In one embodiment, the data engineering system includes an ingest processing module having a schema update submodule and a feature statistics update submodule, wherein the schema update submodule is configured to discover new features and add them to a schema, and wherein the feature statistics update submodule collects statistics for each feature to be used in an online transformation, a record store to store data from a data source, and a transformation module, to receive a low dimensional data instance from the record store and to receive the schema and feature statistics from the ingest processing module, and to transform the low dimensional data instance into a high dimensional representation. One embodiment provides a method for data engineering for machine learning at scale, the method including calling a built-in feature transformation or defining a new transformation, specifying a data source and compressing and storing the data, providing ingest-time processing by automatically analyzing necessary statistics for features, and then generating a schema for a dataset for subsequent data engineering. Other embodiments are disclosed herein.

System for automated data engineering for large scale machine learning

Organ segmentation in chest X-rays using convolutional neural networks is disclosed. One embodiment provides a method to train a convolutional segmentation network with chest X-ray images to generate pixel-level predictions of target classes. Another embodiment will also train a critic network with an input mask, wherein the input mask is one of a segmentation network mask and a ground truth annotation, and outputting a probability that the input mask is the ground truth annotation instead of the prediction by the segmentation network, and to provide the probability output by the critic network to the segmentation network to guide the segmentation network to generate masks more consistent with learned higher-order structures.

Structure correcting adversarial network for chest x-rays organ segmentation

A computer in a distributed peer-to-peer system is disclosed. The distributed system includes a plurality of computers configured to run a distributed machine learning (ML) program represented as an expression of a target loss function with a model parameter matrix. The computer includes: a parser module configured to convert a loss function in the distributed program into an expression graph and then one or more multiplication trees; a parameter replica module in communication with the parser module, the parameter replica module configured to maintain the model parameter matrix of the ML program; a compressor module in communication with the parameter replica module, the compressor module configured to extract sufficient factors from the expression graph for updating the model matrix; and a communication module in communication with the compressor module, the communication module configured to send the sufficient factors for updating model matrix to other machines in the distributed system.

Efficient peer-to-peer architecture for distributed machine learning

A computer in a distributed computing system is disclosed. The computer includes: a graphics processing unit (GPU) memory; a central processing unit (CPU) memory comprising a Key-Value Store (KVS) module; an execution engine module configured to run a deep learning (DL) program to create a plurality of operator graph layers in the graphics processing unit memory; a client library module configured to create a GPU-CPU synchronization (GCS) module for each of the plurality of operator graph layers; a coordination service module configured to compute network cost of a first and a second communication scheme and select, based on the network cost, one of the first and second communication scheme for transmitting data associated with one of the plurality of operator graph layers from a corresponding GCS module.

System with hybrid communication strategy for large-scale distributed deep learning

A system for predicting medications to prescribe to a patient includes a text encoding module and a medication prediction module. The text encoding module is configured to obtain a clinical-information vector from clinical information of the patient. The medication prediction module configured to apply a machine-learned medication-prediction algorithm to the clinical-information vector to select a subset of medications to prescribe to the patient. The machine-learned medication-prediction algorithm is designed with a diversity-promoting regularization model, and is configured to simultaneously consider correlations among different medications and dependencies between patient information and medications when selecting a subset of medications to prescribe to the patient.

Systems and methods for predicting medications to prescribe to a patient based on machine learning

A system for assigning concepts to a medical image includes a visual feature module and a tagging module. The visual feature module is configured to obtain an image feature vector from the medical image. The tagging module is configured to apply a machine-learned algorithm to the image feature vector to assign a set of concepts to the image. The system may also include a text report generator that is configured to generate a written report describing the medical image based on the set of concepts assigned to the medical image.

Systems and methods for automatically tagging concepts to, and generating text reports for, medical images based on machine learning

A system for manipulating images according to styles chosen by a user includes a feed-forward image manipulation model for everyday use and an optimization image manipulation model for more professional use. The optimization image manipulation model optimizes directly over output image pixels to minimize both the content loss and style loss. Users can select their own content and style images, and can choose between using the feed-forward image manipulation model and optimization image manipulation model.

Real-time intelligent image manipulation system

A system including a master machine and a plurality of worker machines is disclosed. The master machine includes, for example, an API server configured to receive a job description; a resource allocation module configured to determine a number of virtual machines required to perform a job based on the job description; a container scheduling module configured to create a container containing the number of virtual machines required to perform the job, wherein at least two of the virtual machines in the container resides on different worker machines, and wherein each of the virtual machines is configured to run a same application to perform the job.

Operating system for distributed enterprise artificial intelligence programs on data centers and the clouds

A constituent-centric neural architecture for reading comprehension is disclosed. One embodiment provides a method that performs reading comprehension comprising encoding individual constituents from a text passage using a chain of trees long short-term encoding, encodes question related to the text passage using a tree long short-term memory encoding, generates a question-aware representation for each constituent in the passage using a tree-guided attention mechanism, generates a plurality of candidate answers from the question-aware representation using hierarchical relations among constituents, and predicts an answer to the question in relation to the text passage using a feed-forward network. Other embodiments are disclosed herein.

Constituent centric architecture for reading comprehension

Structure correcting adversarial network for chest X-rays organ segmentation

A computer system implemented a method for elastic resource management for executing a machine learning (ML) program. The system is configured to create a set of logical executors, assign them across a set of networked physical computation units of a distributed computing system, partition and distribute input data and Work Tasks across the set of logical executors, assign them across the set of networked physical computation units, where the Work Tasks are partitioned into short units of computation (micro-tasks), each calculates a partial update to the ML program's model parameters and each last for less than one second; create a set of logical servers (LSes); partition and distribute globally shared model parameters of the ML program across the set of logical servers; execute partitioned Work Tasks according to a bounded asynchronous parallel standard, where a current Work Task is allowed to execute with stale model parameters without having all the current calculation updates from Work Tasks it depend on, provided the staleness of the model parameters is within a predefined limit.

Growing community of inventors

Pittsburgh, PA, United States of America

Eric Xing