Abstract: A system described herein may receive a set of outputs of a first model, which have been generated by the first model based on a set of inputs, and identify a set of historical values that correspond to the set of inputs and the set of outputs. The inputs and the historical values may be associated with the same time series. The system may train a second model based on the set of inputs to the first model, the set of outputs of the first model, and the set of historical values that correspond to the set of inputs and the set of outputs. The system may determine, based on training the second model, a set of weights associated with the set of historical values; and refine the first model based on the set of weights associated with the set of historical value.

Abstract: A device may receive historical time series data and output data associated with a forecasting model and may process the historical time series data and the output data, with a proxy regression model, to determine inference data. The device may create perturbed data from the historical time series data, the output data, and the inference data, and may process the perturbed data, with the proxy regression model, to generate labelled data and to identify top features of the labelled data. The device may process subsets of the top features of the labelled data, with the proxy regression model, to determine feature data identifying an importance of each of the subsets of the top features, and may evaluate the proxy regression model and the feature data to calculate validation data for validating the forecasting model. The device may validate the forecasting model with the validation data.

Abstract: One or more computing devices, systems, and/or methods for network equipment solution generation utilizing network signals are provided. Network signals collected by devices within a location are evaluated by a neural network to generate contextually aware pixel values forming a vision map of the location. The contextually aware pixel values and/or the vision map are evaluated to generate a coverage map corresponding to regions of low signal coverage and obstacles proximate the regions. A simulation of network equipment operating at the location is performed using the coverage map to generate a simulation result. The simulation result is used to generate an installation plan to install network equipment at the location.

Abstract: One or more computing devices, systems, and/or methods for privacy data augmentation are provided. An augmentation pipeline is selected to process data based upon a data type of the data. The augmentation pipeline processes the data to generate information that is input into a machine learning model. The machine learning model processes the information and privacy laws to determine a subset of the data to mask. In this way, the subset of the data is masked to create augmented data that complies with the privacy laws.

Abstract: A device may receive video data that includes a text transcript, audio sequences, and image frames, and may detect a network fluctuation. The device may process the text transcript to generate a new phrase, and may generate a response phoneme based on the new phrase. The device may generate a text embedding based on the response phoneme, and may process the audio sequences to generate a target voice sequence. The device may generate an audio embedding based on the target voice sequence, and may process the image frames to generate a target image sequence. The device may generate an image embedding based on the target image sequence, and may combine the embeddings to generate an embedding input vector. The device may generate a final voice response and a final video based on the embedding input vector, and may provide the video data, the final voice response, and the final video.

Abstract: Disclosed are embodiments for improving training data for machine learning (ML) models. In an embodiment, a method is disclosed where an augmentation engine receives a seed example, the seed example stored in a seed training data set; generates an encoded seed example of the seed example using an encoder; inputs the encoded seed example into a machine learning model and receives a candidate example generated by the machine learning model; determines that the candidate example is similar to the encoded seed example; and augments the seed training data set with the candidate example.

Abstract: Techniques for generating emotionally-aware digital content are disclosed. In one embodiment, a method is disclosed comprising obtaining audio input, obtaining a textual representation of the audio input; using the textual representation of the audio input to identify an emotion corresponding to the audio input; generating an emotionally-aware facial representation in accordance with the textual representation and the identified emotion; using the emotionally-aware facial representation to generate one or more images comprising at least one facial expression corresponding to the identified emotion; and providing digital content comprising the one or more images.

Abstract: One or more computing devices, systems, and/or methods for identifying anomalous behavior of users are provided. In an example, users of a telecommunication service provider may be segmented into a plurality of user segments based upon telecommunication service metrics associated with the users. A machine learning model may be trained using telecommunication service information associated with users of the first user segment to generate a trained machine learning model. Using the trained machine learning model, a forecast of telecommunication service usage associated with a first user segment of the plurality of user segments. A telecommunication service usage metric, associated with a user belonging to the first user segment, may be compared with a range indicated by the forecast. The user may be flagged as having anomalous behavior based upon a determination that one or more telecommunication usage metrics, associated with the user, are outside one or more ranges indicated by the forecast.

Abstract: A method may include receiving a number of images to train a first neural network, masking a portion of each of the images and inputting the masked images to the first neural network. The method may also include generating, by the first neural network, probable pixel values for pixels located in the masked portion of each of the plurality of images, forwarding the images including the probable pixel values to a second neural network and determining, by the second neural network, whether each of the probable pixel values is contextually suitable. The method may further include identifying pixels in each of the plurality of images that are not contextually suitable.

Abstract: A device may receive video data that includes a text transcript, audio sequences, and image frames, and may detect a network fluctuation. The device may process the text transcript to generate a new phrase, and may generate a response phoneme based on the new phrase. The device may generate a text embedding based on the response phoneme, and may process the audio sequences to generate a target voice sequence. The device may generate an audio embedding based on the target voice sequence, and may process the image frames to generate a target image sequence. The device may generate an image embedding based on the target image sequence, and may combine the embeddings to generate an embedding input vector. The device may generate a final voice response and a final video based on the embedding input vector, and may provide the video data, the final voice response, and the final video.

Abstract: Techniques for determining anomalous user behavior in connection with an online application are disclosed. In one embodiment, a method is disclosed comprising obtaining user behavior data in connection with a user of an application, generating feature data using the obtained user behavior data, obtaining one or more user behavior anomaly predictions from one or more anomaly prediction models trained to output a user behavior anomaly prediction in response to the feature data. Each user behavior anomaly prediction indicates a probability that the user behavior is anomalous. A user behavior anomaly determination is made using the user behavior anomaly prediction(s).

Abstract: A device may receive channel data associated with digital channels utilized by a customer, and may identify, in the channel data, potential issues and sequential events for the potential issues. The device may generate a causal relationship graph based on the sequential events and the potential issues, and may process the causal relationship graph, with a machine learning model, to identify issues of the customer and events associated with the issues. The device may apply causal inference to identify causal relationships between the issues and the events, and may calculate an inference score for the issues and the events based on the causal relationships. The device may modify, based on the inference score, a customer journey defined by the events to generate a modified customer journey, and may perform one or more actions based on the modified customer journey.

Abstract: Disclosed are embodiments for improving training data for machine learning (ML) models. In an embodiment, a method is disclosed where an augmentation engine receives a seed example, the seed example stored in a seed training data set; generates an encoded seed example of the seed example using an encoder; inputs the encoded seed example into a machine learning model and receives a candidate example generated by the machine learning model; determines that the candidate example is similar to the encoded seed example; and augments the seed training data set with the candidate example.

Abstract: A device may receive and convert audio data to text data in real-time, and may detect a network fluctuation that causes missing voice packets. The device may process partial text and context of the text data, with a model, to generate a new phrase, and may generate a response phoneme for the new phrase. The device may utilize a text embedding model to generate a text embedding for the response phoneme, and may process the audio data, with the model, to generate a target voice sequence. The device may utilize an audio embedding model to generate an audio embedding for the target voice sequence, and may combine the text embedding and the audio embedding to generate an embedding input vector. The device may process the embedding input vector, with an audio synthesis model, to generate a final voice response, and may provide the audio data and the final voice response.

Abstract: Disclosed are systems and methods for a computerized framework enacted by service contact centers that provides a proactive and adaptive response system that accurately identifies security and/or legal concerns of service requests, and enables artificial intelligence/machine learning (AI/ML)-based mechanisms for dynamically addressing the underlying technical and/or service related concerns of such service requests. The disclosed framework can computationally determine how effective service agents have been, and are currently being in curating solutions/responses to each customer service call, which can enable modified functionality for the customer as well as curated services based on how sufficiently handled the service call was responded to by the agent. The disclosed systems and methods provide a generative service call experience that can improve agent performance while reducing the strain on user experience, both during and/or after service calls.

Abstract: Techniques for identifying user reaction in connection with an online application are disclosed. In one embodiment, a method is disclosed comprising obtaining activity data in connection with a user of an application, generating feature data using the obtained activity data, obtaining a user reaction prediction from a user reaction prediction model trained to output the user reaction prediction in response to the feature data. The user reaction prediction indicates a probability of the user reaction in connection with the application. A determination is made, using the user reaction prediction, whether or not to take a remedial action in connection with the user and the application.

Abstract: The present teaching relates to compressing a model for an application to generate a compressed model. The model has multiple layers, each of which has multiple nodes. Operating the model utilizing an application-dependent dataset, redundant nodes/layers in the model are identified via a loss-based assessment. The loss-based assessment using aggregated output vectors computed based on output vectors produced by the nodes/layers of the model in response to the data samples of the application-dependent dataset. Removing the redundant nodes/layers yields the compressed model.

Abstract: A video summary device may generate a textual summary of a transcription of a virtual event. The video summary device may generate a phonemic transcription of the textual summary and generate a text embedding based on the phonemic transcription. The video summary device may generate an audio embedding based on a target voice. The video summary device may generate an audio output of the phonemic transcription uttered by the target voice. The audio output may be generated based on the text embedding and the audio embedding. The video summary device may generate an image embedding based on video data of a target user. The image embedding may include information regarding images of facial movements of the target user. The video summary device may generate a video output of different facial movements of the target user uttering the phonemic transcription, based on the text embedding and the image embedding.

Abstract: An improved search engine is disclosed. The search engine receives search queries from client devices and inputs these queries into a first neural network (an action understanding model) that includes an action embedding layer. The action embedding layer can be a word embedding layer constructed using action terms. The action understanding model outputs a filter match associated with a type of filter and, in some scenarios, an action-condition pair. The action-condition pair includes an action associated with the type of filter and a condition comprising an adaptive value associated with the action. Based on the filter and, if present, action-condition pair(s), the embodiments generate a structured query and issue the structured query to a data repository (e.g., database). The search engine then returns a search results page responsive to the search query that includes the results returned by the data repository in response to the structured query.

Abstract: A device may receive taxonomy data and may preprocess the taxonomy data with preprocessing techniques to generate preprocessed data. The device may process the taxonomy data, with a machine learning interpolative-based feedback model, to generate intents, features of each of the intents, and a taxonomy collection, and may process the taxonomy data, with a machine learning-based feedback model, to generate concepts or entities associated with the intents. The device may combine the intents, the features, the taxonomy collection, and the concepts or the entities to generate an association collection, and may train a machine learning model with the association collection and the preprocessed data to generate a trained machine learning model. The device may process text data, the taxonomy collection, and the association collection, with the trained machine learning model, to determine a crux of the text data, and may perform actions based on the crux of the text data.