Abstract: A data source configured to provide a representation of an environment of one or more agents is identified. Using a data set obtained from the data source, a neural network-based reinforcement learning model with one or more attention layers is trained. Importance indicators generated by the attention layers are used to identify actions to be initiated by an agent. A trained version of the model is stored.

Abstract: Various embodiments of a hierarchical system or method of identifying sensitive content in data is described. In some embodiments, sensitive data classifiers local to a data storage system can analyze a plurality of data items and classify at least some data items as potentially containing sensitive data. The sensitive data classifiers can provide the classified data items to a separate sensitive data discovery component. The sensitive data discovery component can, in some embodiments, obtain the classified data items, perform a sensitive data location analysis on the classified data items to identify a location of sensitive data within some of the classified data items, and generate location information for the sensitive data within the data items containing sensitive data. The sensitive data discovery component can provide to a destination this information, in some embodiments, where the destination might redact, tokenize, highlight, or perform other actions on the located sensitive data.

Abstract: A data source configured to provide a representation of an environment of one or more agents is identified. Using a data set obtained from the data source, a neural network-based reinforcement learning model with one or more attention layers is trained. Importance indicators generated by the attention layers are used to identify actions to be initiated by an agent. A trained version of the model is stored.

Abstract: Techniques for multi-label document classification are described. Clustering is used to cluster labels in a set. A machine learning model including a multi-label classifier for each cluster is created, the multi-label classifier for a given cluster to classify a document with one or more of the labels in the cluster.

Abstract: Techniques for creating a text classifier machine learning (ML) model are described. According to some embodiments, a language processing service finetunes a language ML model on unlabeled documents of a user, and then trains that finetuned language ML model on labeled documents of the user to be a text classifier that is customized for that user’s domain, e.g., the user’s documents. Additionally, the finetuned language ML model may be trained on labeled documents of the user, for prediction objectives for unlabeled data, before being trained as the text classifier.

Abstract: Techniques for implementing a chatbot that utilizes context embeddings are described. An exemplary method includes determining a next turn by: applying a language model to the utterance to determine a probability of a sequence of words, generating a context embedding for the utterance based at least on one or more of: a dialog act as defined by a chatbot definition of the chatbot, a topic vector identifying a domain of the chatbot, a previous chatbot response, and one or more slot options; performing neural language model rescoring using the determined probability of a sequence of words as a word embedding and the generated context embedding to predict an hypothesis; determining at least a name of a slot and type to be fulfilled based at least in part on the hypothesis and the chatbot definition; and determining a next turn based at least in part on the chatbot definition, any previous state, and the name of the slot and type to be fulfilled.

Abstract: Techniques for data lake-based text generation and data augmentation for machine learning training are described. A user-provided dataset including documents and corresponding label information can be automatically supplemented by creating additional high-quality document samples, with labels, via a large repository of documents in a data lake. Documents from the data lake may be identified as being semantically similar to the user-provided documents but different enough to allow a resulting model to learn from the variation in these documents. New documents can be generated from user-provided document samples or data lake sample documents by identifying and replacing slots within the samples and rewriting adjunct tokens.

Abstract: Techniques for phased deployment of machine learning models are described. Customers can call a training API to initiate model training, but then must wait while the training completes before the model can be used to perform inference. Depending on the type of model, machine learning algorithm being used for training, size of the training dataset, etc. this training process may take hours or days to complete. This leads to significant downtime where inference requests cannot be served. Embodiments improve upon existing systems by providing phased deployment of custom models. For example, a simple, less accurate model, can be provided synchronously in response to a request for a custom model. At the same time, one or more machine learning models can be trained asynchronously in the background. When the machine learning model is ready for use, the customers' traffic and jobs can be transferred over to the better model.

Abstract: Techniques for predicting punctuation and casing using multimodal fusion are described.

Abstract: A transcription service may receive a request from a developer to build a custom speech-to-text model for a specific domain of speech. The custom speech-to-text model for the specific domain may replace a general speech-to-text model or add to a set of one or more speech-to-text models available for transcribing speech. The transcription service may receive a training data and instructions representing tasks. The transcription service may determine respective schedules for executing the instructions based at least in part on dependencies between the tasks. The transcription service may execute the instructions according to the respective schedules to train a speech-to-text model for a specific domain using the training data set. The transcription service may deploy the trained speech-to-text model as part of a network-accessible service for an end user to convert audio in the specific domain into texts.

Abstract: Techniques for extending sensitive data tagging without reannotating training data are described. A method for extending sensitive data tagging without reannotating training data may include hosting a plurality of models at a model endpoint in a machine learning service, each model trained to identify a different sensitive data type in a transcript of content, adding a new model to the model endpoint, the new model trained to identify a new sensitive data entity in the transcript of content, identifying sensitive entities in the transcript by each of the plurality of models and the new model, merging inference responses generated by each of the plurality of models and the new model using at least one inference policy, and returning a merged inference response identifying a plurality of sensitive entities in the transcript.

Abstract: Techniques are described for a de-obfuscation framework that utilizes image recognition of text. A word input by a user is received by the de-obfuscation service. Visual feature data associated with an image corresponding to each character of the word is generated. Word embeddings are generated using the visual feature data and each character of the word using a character encoder layer. Feature vectors are generated from the word embedding by combining the generated word embeddings and a provided word embedding using a second neural network. The generated feature vector is classified. Potential text obfuscation is detected from the classified generated feature vector using a lexicon to determine de-obfuscated text closet to the user text.

Abstract: Techniques for analyzing a message are described. For example, in some examples a message analyzer is to receive a message including message content and an indication of at least one message recipient, determine a sentiment of the message to generate at least one first sentiment value, determine at least on topic of the message content, and determine that the at least one first sentiment value is less than a second sentiment value associated with the message recipient and the at least one topic of the message content.

Abstract: A machine-learning system analyzes electronic books to determine a “start-of-reading location” (SRL) in each book. Based on this location, when an electronic book is opened on a reading device for the first time, the book can be opened to where a reader is likely to want to start reading, automatically skipping past introductory pages. Books are divided into logical blocks (e.g., title page, forward, chapters, etc.), and a title portion and a body-text portion is identified in each block. A title classifier attempts to determine whether or not a block should be marked as the SRL. If the score from the title classifier is indefinite, a body-text classifier is used.

Abstract: A machine learning engine may correlate contextual information associated with a misspelling in a publication with a likelihood that the misspelling is intentional in nature. Training data may be generated by analyzing one or more past publication to identify misspellings and labeling the misspellings as intentional. A contextual indicators application may analyze the context in which intentional misspellings have been previously included within publication to identify indicators of future misspellings being intentional. A machine learning engine may use the training data and indicators to generate an intentional linguistic deviation (ILD) prediction model to determine whether a new misspelling is an intentional misspelling. The machine learning engine may also determine weights for individual indicators that may calibrate the influence of the respective individual indicators.