Abstract: A system according determines a machine learning based model for forecasting time series data for a given use case. The system determines a model metric for a specific use case of time series data. The system accesses a pool of machine learning based models including a plurality of machine learning based models machine learning based models based on different machine learning techniques. For each of the plurality of machine learning based models the system performs forecasting using the machine learning based model and determines the value of the model metric for the machine learning based model. The system selects a machine learning based model based on comparison of values of the model metric for machine learning based models. The system uses the selected machine learning based model for forecasting values for the time series data for the application.

Abstract: An information processing apparatus designates one or more constraints for constraining a configuration of a neural network, which include information for specifying the configuration of the neural network and the maximum number of computations in the neural network, and executes a computation of a neural network configured based on the designated constraints. When a neural network is trained, the information processing apparatus trains respective neural networks under the one or more designated constraints, and selects, for estimation, a predetermined learned model out of the learned models trained under the one or more designated constraints.

Abstract: A method performed by a node for predicting a behavior of users of a communications network is described. The node manages an artificial neural network. The node merges a first pre-existing predictive model of the behavior in a first group of users with a second model of the behavior in a second group of users. The merging comprises establishing connections between the first model and the second model. Each of the connections has a respective weight. The respective weights of the connections are learned by respective connections of neurons in the artificial neural network based on data from a third group of users. The node also obtains a third model for predicting the behavior in the third group of users, based on the merged models and the data from the third group of users.

Abstract: Methods and systems for managing data collection are disclosed. To manage data collection, a system may include a data aggregator and a data collector. The data aggregator may utilize complex inference models to predict the future operation of the data collector, while the data collector may host simpler inference models. The data collector may access inferences from the complex models by obtaining a difference between complex and simple inferences from the data aggregator and locally reconstructing the complex differences. To reduce data transmission, the data collector may transmit a data difference (e.g., a reduced-size representation of a measurement) to the data aggregator using the reconstructed complex inferences. The data aggregator may reconstruct data from the data collectors using the data difference from the data collector and inferences from the complex inference model.

Abstract: The embodiments describe a technique for customizing activation functions automatically, resulting in reliable improvements in performance of deep learning networks. Evolutionary search is used to discover the general form of the function, and gradient descent to optimize its parameters for different parts of the network and over the learning process. The new approach discovers new parametric activation functions which improve performance over previous activation functions by utilizing a flexible search space that can represent activation functions in an arbitrary computation graph. In this manner, the activation functions are customized to both time and space for a given neural network architecture.

Abstract: The present disclosure relates to a method, a device, and a program product for training a model. The method includes: receiving at least one unlabeled sample and at least one labeled sample for training a pre-training model, the pre-training model being used to extract features of the samples; creating an undirected graph associated with the pre-training model using the at least one unlabeled sample and a set of training samples associated with the pre-training model; dividing the undirected graph to form a plurality of sub-graphs based on corresponding features of the unlabeled sample and the set of training samples, the plurality of sub-graphs corresponding to a plurality of classifications of the samples, respectively; and training, based on the plurality of sub-graphs and the at least one labeled sample, the pre-training model to generate a training model. A corresponding device and a corresponding computer program product are provided.

Abstract: Embodiments described a field extraction system that does not require field-level annotations for training. Specifically, the training process is bootstrapped by mining pseudo-labels from unlabeled forms using simple rules. Then, a transformer-based structure is used to model interactions between text tokens in the input form and predict a field tag for each token accordingly. The pseudo-labels are used to supervise the transformer training. As the pseudo-labels are noisy, a refinement module that contains a sequence of branches is used to refine the pseudo-labels. Each of the refinement branches conducts field tagging and generates refined labels. At each stage, a branch is optimized by the labels ensembled from all previous branches to reduce label noise.

Abstract: A system for updating metadata associated with historic data. The system includes an electronic computing device. The electronic computing device is configured to receive historic data and metadata associated with the historic data and, using a low capacity machine learning model, analyze the metadata associated with the historic data to determine a probability that the historic data includes the object of interest. The electronic computing device is also configured to compare the probability to a predetermined threshold. The electronic computing device is further configured to use a high capacity machine learning model to analyze the historic data to determine whether the historic data includes the object of interest and update the metadata of the historic data based on whether the historic data includes the object of interest to generate updated metadata, when the probability is greater than or equal to the predetermined threshold.

Abstract: A method for scoring training data samples according to an ability to preserve latent decision boundaries for previously observed classes while promoting learning from an input batch of new images from an online data stream, comprising: receiving the input batch of the new images from the online data stream, performing a memory retrieval process that retrieves data to be learned along with a new set of data from the memory to retain the previously learned knowledge, and performing a memory update process that selects and exchanges a small set of data to be saved in the memory in the memory update process. In addition, the method performs data valuation based on KNN-SV for both the memory retrieval and memory update processes to perform strategic and intuitive data selection based on the properties of KNN-SV.