Abstract: Embodiments provide a computer implemented method of training an enhanced chatflow system, comprising: ingesting a corpus of information comprising at least one user input node corresponding to a user question and at least one expert-designed variation for each user input node; matching one or more user inputs to one or more corresponding dialog nodes using regular expressions and delimiters; ingesting one or more usage logs from a deployed dialog system, each usage log comprising at least one user input node; for each user input node: designating the node as a class; storing the node in a dialog node repository; designating each of the at least one variations as training examples for the designated class; converting the classes and the training examples into feature vector representations; training one or more classifiers and one or more classification objectives using the feature vector representations.

Abstract: The present disclosure provides systems and methods that use machine-learned models, such as deep neural networks, to predict and prevent adverse conditions at structural assets. One example method includes obtaining data descriptive of a plurality of images that depict at least a portion of a geographic area that contains a first structural asset. The plurality of images include at least a first image captured at a first time and a second image captured at a second time that is different than the first time. The method includes inputting data descriptive of at least the first image, the first time, the second image, and the second time into a condition prediction model. The method includes receiving, as an output of the condition prediction model, at least one prediction regarding the occurrence of an adverse condition at the first structural asset during one or more future time periods.

Abstract: Aspects disclosed in the detailed description include memory compression in a deep neural network (DNN). To support a DNN application, a fully connected weight matrix associated with a hidden layer(s) of the DNN is divided into a plurality of weight blocks to generate a weight block matrix with a first number of rows and a second number of columns. A selected number of weight blocks are randomly designated as active weight blocks in each of the first number of rows and updated exclusively during DNN training. The weight block matrix is compressed to generate a sparsified weight block matrix including exclusively active weight blocks. The second number of columns is compressed to reduce memory footprint and computation power, while the first number of rows is retained to maintain accuracy of the DNN, thus providing the DNN in an efficient hardware implementation without sacrificing accuracy of the DNN application.

Abstract: Systems and methods to predict availability of retail spaces in shopping malls. Robots are dispatched to temporary retail spaces to collect at least a portion of operation data stored in a database with leasing data of temporary retail spaces in a mall, tenant coordination data, and subscriber data. A web-based application predicts availability of the temporary retail spaces in the mall based on the leasing data, the tenant coordination data, and the operation data. The web-based application includes an interactive plan of the mall as a user interface for subscribers to access data related to the temporary retail spaces, and process applications for leasing the temporary retail spaces. A machine learning technique is applied to a dataset of the tenant to derive a predictive model for predicting tenant default in a period of time.

Abstract: Disclosed is a technique that can be performed by an electronic device. The technique can include generating raw data based on inputs to the electronic device, and sending the raw data or data items over a network to a server computer system. The sent raw data or the data items can include training data. The technique can further include receiving global model data from the server computer system over the network. The global model data may have been derived from the training data in accordance with a machine learning process. The technique can further include generating an updated local model by updating a local model associated with the electronic device based on the received global model data, and processing local data based on the updated local model to generate output data. The local data can include raw data or data items generated based on inputs to the electronic device.