Abstract: Systems and methods are disclosed to implement a neural network training system to train a multitask neural network (MNN) to generate a low-dimensional entity representation based on a sequence of events associated with the entity. In embodiments, an encoder is combined with a group of decoders to form a MNN to perform different machine learning tasks on entities. During training, the encoder takes a sequence of events in and generates a low-dimensional representation of the entity. The decoders then take the representation and perform different tasks to predict various attributes of the entity. As the MNN is trained to perform the different tasks, the encoder is also trained to generate entity representations that capture different attribute signals of the entities. The trained encoder may then be used to generate semantically meaningful entity representations for use with other machine learning systems.

Abstract: At least some embodiments are directed to an insights inference system that produces multiple insights associated with an entity. The insights inference system generates a decision tree machine learning model, assigning a first insight to a parent node of a decision tree machine learning model and assigning at least one second insight to child nodes of the decision tree machine learning model. Each child node is associated with a sequence number and a rank number. The sequence number and the rank number are indicative of a significance associated with the at least one second insight. The insight inference system responds to queries by traversing the decision tree machine learning model to compute at least one response insight based on the sequence number and the rank number associated with each child node and outputs the at least one response insight to a client terminal.

Abstract: Techniques are described for generating and applying mini-machine learning variants of machine learning algorithms to save computational resources in tuning and selection of machine learning algorithms. In an embodiment, at least one of the hyper-parameter values for a reference variant is modified to a new hyper-parameter value thereby generating a new variant of machine learning algorithm from the reference variant of machine learning algorithm. A performance score is determined for the new variant of machine learning algorithm using a training dataset, the performance score representing the accuracy of the new machine learning model for the training dataset. By performing training of the new variant of machine learning algorithm with the training data set, a cost metric of the new variant of machine learning algorithm is measured by measuring usage the used computing resources for the training.

Abstract: A method comprising, by a processing unit and a memory: obtaining a training set of data; dividing sets of data into a plurality of groups, wherein all sets of data for which feature values meet at least one similarity criterion, are in the same group, storing in a reduced training set of data, for each group, at least one aggregated set of data, wherein, for a plurality of the groups, a number of aggregated sets of data is less than a number of the sets of data of the group, wherein the reduced training set of data is suitable to be used in a classification algorithm for determining a relationship between the at least one label and the features of the electronic items, thereby reducing computation complexity when processing the reduced training set of data, compared to processing the training set of data.

Abstract: A post-processing method, system, and computer program product for post-hoc improvement of instance-level and group-level prediction metrics, including training a bias detector that learns to detect a sample that has an individual bias greater than a predetermined individual bias threshold value with constraints on a group bias, applying the bias detector on a run-time sample to select a biased sample in the run-time sample having a bias greater than the predetermined individual bias threshold bias value, and suggesting a de-biased prediction for the biased sample.

Abstract: Embodiments of the invention include a computer-implemented method for detecting anomalies in non-stationary data in a network of computing entities. The method collects non-stationary data in the network and classifies the non-stationary data according to a non-Markovian, stateful classification, based on an inference model. Anomalies can then be detected, based on the classified data. The non-Markovian, stateful process allows anomaly detection even when no a priori knowledge of anomaly signatures or malicious entities exists. Anomalies can be detected in real time (e.g., at speeds of 10-100 Gbps) and the network data variability can be addressed by implementing a detection pipeline to adapt to changes in traffic behavior through online learning and retain memory of past behaviors. A two-stage scheme can be relied upon, which involves a supervised model coupled with an unsupervised model.

Abstract: A method and apparatus for recognizing a low-quality article based on artificial intelligence, a device and a medium. The method comprises: obtaining a user feedback behavior feature of a to-be-recognized article in a news-recommending system; according to the user feedback behavior feature of the to-be-recognized article and a predetermined low-quality article recognition model, recognizing whether the to-be-recognized article is a low-quality article.

Abstract: A technique is disclosed for generating class level rules that globally explain the behavior of a machine learning model, such as a model that has been used to solve a classification problem. Each class level rule represents a logical conditional statement that, when the statement holds true for one or more instances of a particular class, predicts that the respective instances are members of the particular class. Collectively, these rules represent the pattern followed by the machine learning model. The techniques are model agnostic, and explain model behavior in a relatively easy to understand manner by outputting a set of logical rules that can be readily parsed. Although the techniques can be applied to any number of applications, in some embodiments, the techniques are suitable for interpreting models that perform the task of classification. Other machine learning model applications can equally benefit.

Abstract: A learning device is configured to perform learning of a decision tree by gradient boosting. The learning device includes a plurality of learning units and a plurality of model memories. The plurality of learning units are configured to perform learning of the decision tree using learning data divided to be stored in a plurality of data memories. The plurality of model memories are each configured to store data of the decision tree learned by corresponding one of the plurality of learning units.

Abstract: Methods, systems, and non-transitory computer readable storage media are disclosed for analyzing feature impact of a machine-learning model using prototypes across analytical spaces. For example, the disclosed system can identify features of data points used to generate outputs via a machine-learning model and then map the features to a feature space and the outputs to a label space. The disclosed system can then utilize an iterative process to determine prototypes from the data points based on distances between the data points in the feature space and the label space. Furthermore, the disclosed system can then use the prototypes to determine the impact of the features within the machine-learning model based on locally sensitive directions; region variability; or mean, range, and variance of features of the prototypes.

Abstract: Example aspects of the present disclosure are directed to systems and methods that enable improved adversarial training of machine-learned models. An adversarial training system can generate improved adversarial training examples by optimizing or otherwise tuning one or hyperparameters that guide the process of generating of the adversarial examples. The adversarial training system can determine, solicit, or otherwise obtain a realism score for an adversarial example generated by the system. The realism score can indicate whether the adversarial example appears realistic. The adversarial training system can adjust or otherwise tune the hyperparameters to produce improved adversarial examples (e.g., adversarial examples that are still high-quality and effective while also appearing more realistic). Through creation and use of such improved adversarial examples, a machine-learned model can be trained to be more robust against (e.g.

Abstract: Herein are techniques for exploring hyperparameters of a machine learning model (MLM) and to train a regressor to predict a time needed to train the MLM based on a hyperparameter configuration and a dataset. In an embodiment that is deployed in production inferencing mode, for each landmark configuration, each containing values for hyperparameters of a MLM, a computer configures the MLM based on the landmark configuration and measures time spent training the MLM on a dataset. An already trained regressor predicts time needed to train the MLM based on a proposed configuration of the MLM, dataset meta-feature values, and training durations and hyperparameter values of landmark configurations of the MLM. When instead in training mode, a regressor in training ingests a training corpus of MLM performance history to learn, by reinforcement, to predict a training time for the MLM for new datasets and/or new hyperparameter configurations.

Abstract: A computer system includes a processor and a memory connected to the processor, and manages pieces of reward function information for defining rewards for states and actions of the control targets for each of the control targets. The pieces of reward function information includes first reward function information for defining the reward of a first control target and second reward function information for defining the reward of a second control target. When updating the first reward function information, the processor compares the rewards of the first reward function information and the second reward function information with each other, specifies a reward, which is reflected in the first reward function information from rewards set in the second reward function information, updates the first reward function information on the basis of the specified reward, and decides an optimal action of the first control target by using the first reward function information.

Abstract: Presence of malicious code can be identified in one or more data samples. A feature set extracted from a sample is vectorized to generate a sparse vector. A reduced dimension vector representing the sparse vector can be generated. A binary representation vector of reduced dimension vector can be created by converting each value of a plurality of values in the reduced dimension vector to a binary representation. The binary representation vector can be added as a new element in a dictionary structure if the binary representation is not equal to an existing element in the dictionary structure. A training set for use in training a machine learning model can be created to include one vector whose binary representation corresponds to each of a plurality of elements in the dictionary structure.

Abstract: A non-transitory computer-readable medium including instructions, which when executed by one or more processors of a computing system, causes the computing system to: access a machine learning model m, an input data point P to m, P including one or more features, and a prediction m(P) of m for P; create a set of perturbed input data points Pk from P by selecting a new value for at least one feature of P for each perturbed input data point; obtain a prediction m(Pk) for each of the perturbed input data points; analyze the predictions m(Pk) for the perturbed input data points to determine which features are most influential to the prediction; and output the analysis results to a user.

Abstract: Some embodiments provide a method for compiling a neural network (NN) program for an NN inference circuit (NNIC) that includes multiple partial dot product computation circuits (PDPCCs) for computing dot products between weight values and input values. The method receives an NN definition with multiple nodes. The method assigns a group of filters to specific PDPCCs. Each filter is assigned to a different set of the PDPCCs. When a filter does not have enough weight values equal to zero for a first set of PDPCCs to which the filter is assigned to compute dot products for nodes that use the filter, the method divides the filter between the first set and a second set of PDPCCs. The method generates program instructions for instructing the NNIC to execute the NN by using the first and second PDPCCs to compute dot products for the nodes that use the filter.

Abstract: Provided are a neural network method and an apparatus, the method including obtaining a set of floating point data processed in a layer included in a neural network, determining a weighted entropy based on data values included in the set of floating point data, adjusting quantization levels assigned to the data values based on the weighted entropy, and quantizing the data values included in the set of floating point data in accordance with the adjusted quantization levels.