Abstract: A machine learning based model generates a feature representation of a text sequence, for example, a natural language sentence or phrase. The system trains the machine learning based model by receiving an input text sequence and perturbing the input text sequence by masking a subset of tokens. The machine learning based model is used to predict the masked tokens. A predicted text sequence is generated based on the predictions of the masked tokens. The system processes the predicted text sequence using the machine learning based model to determine whether a token was predicted or an original token. The parameters of the machine learning based model are adjusted to minimize an aggregate loss based on prediction of the correct word for a masked token and a classification of a word as original or replaced.

Abstract: Embodiments of the present disclosure include a method for token-position handling comprising: processing a first sequence of tokens to produce a second sequence of tokens, wherein the second sequence of tokens has a smaller number of tokens than the first sequence of tokens; masking at least some tokens in the second sequence to produce masked tokens; moving the masked tokens to the beginning of the second sequence to produce a third sequence; encoding tokens in the third sequence into a set of numeric vectors in a first array; and processing the first array in a transformer neural network to determine correlations among the third sequence, the processing the first array producing a second array.

Abstract: In one embodiment, a method includes receiving a user request from a client system associated with a first user, wherein the user request is associated with a semantic-intent, identifying one or more dialog-intents associated with the user request based on the semantic-intent and context information associated with the user request, wherein each dialog-intent is a sub-intent of the semantic-intent, determining one or more agents for executing one or more tasks associated with the one or more dialog-intents, and sending instructions for presenting information returned from the one or more agents responsive to executing the one or more tasks to the client system.

Abstract: A processor-implemented method includes determining a first quantization value by performing log quantization on a parameter from one of input activation values and weight values in a layer of a neural network, comparing a threshold value with an error between a first dequantization value obtained by dequantization of the first quantization value and the parameter, determining a second quantization value by performing log quantization on the error in response to the error being greater than the threshold value as a result of the comparing; and quantizing the parameter to a value in which the first quantization value and the second quantization value are grouped.

Abstract: A method for protecting visual private data by preventing data reconstruction from latent representations of deep networks is presented. The method includes obtaining latent features from an input image and learning, via an adversarial reconstruction learning framework, privacy-preserving feature representations to maintain utility performance and prevent the data reconstruction by simulating a black-box model inversion attack by training a decoder to reconstruct the input image from the latent features and training an encoder to maximize a reconstruction error to prevent the decoder from inverting the latent features while minimizing the task loss.

Abstract: The accuracy of machine translation is increased. A translated document with high translation accuracy is obtained. An original document is faithfully translated. An original document is translated with a neural network to generate a first translated document; a modification-target word or phrase is determined from words and phrases contained in the original document on the basis of an analysis result for the first translated document; the modification-target word or phrase is replaced with a high frequency word in learning data used for learning in the neural network to modify the original document; and the modified original document is translated with the neural network to generate a second translated document.

Abstract: A method, system, and computer program product is provided for tokenizing document citations. The method may include tokenizing each string of a plurality of strings into at least one citation token representing at least one citation parameter, resulting in a plurality of citation tokens, grouping the plurality of citation tokens into a plurality of token groups, each token group of the plurality of token groups including at least one citation token representing a separate citation in the at least one textual document, assigning metadata to each token group of the plurality of token groups, and generating a normalized citation for each token group of the plurality of token groups based on the metadata.

Abstract: Systems and methods include acquisition of a plurality of text segments, each of the text segments associated with a flag value indicating whether the text segment is associated with a correct replacement text or an incorrect replacement text, determination of one or more n-grams of each text segment of the plurality of text segments, generation, based on the one or more n-grams of each text segment and the flag value associated with each text segment, a model to determine a flag value based on one or more input n-grams, reception of an input text segment, determination of a second one or more n-grams of the input text segment, determination, using the model, of an output flag value based on the determined second one or more n-grams, and presentation of the input text segment and the output flag value on a display.

Abstract: An apparatus may include a processor that may be caused to access a script in a first format, the script comprising a rule element having a condition element that specifies conditional logic and an action element that specifies an action to be taken based on the conditional logic, build a condition object for the condition element, generate, based on the condition object, a condition sentence portion that represents the conditional logic in a natural language format, build an action object for the action element, generate, based on the action object, an action sentence portion that represents the action to be taken in the natural language format, and generate, in the natural language format, a natural language sentence for the rule element based on the condition sentence portion and the action sentence portion.

Abstract: A system and method for generating a response in a video grounded dialogue are provided. A video-grounded dialogue neural network language model receives video input and text input. The text input includes a dialogue history between the model and a human user and a current utterance by the user. Encoded video input is generated using video encoding layers. Encoded text input is generated using text encoding layers. The encoded video input and the encoded text input are concatenated in to a single input sequence. A generative pre-trained transformer model generates the response to the current utterance from the singe input sequence.

Abstract: A system for and method of characterizing a target application acoustic domain analyzes one or more speech data samples from the target application acoustic domain to determine one or more target acoustic characteristics, including a CODEC type and bit-rate associated with the speech data samples. The determined target acoustic characteristics may also include other aspects of the target speech data samples such as sampling frequency, active bandwidth, noise level, reverberation level, clipping level, and speaking rate. The determined target acoustic characteristics are stored in a memory as a target acoustic data profile. The data profile may be used to select and/or modify one or more out of domain speech samples based on the one or more target acoustic characteristics.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for incorporating a computationally efficient expressive output layer in a neural network. The output layer is configured to map a received hidden state to a probability distribution over a vocabulary of possible outputs by generating, from the hidden state, a respective context embedding for each of a plurality of gates; for each of the possible outputs in the vocabulary, computing a gated logit for the possible output by applying an output embedding for the possible output to the weighed sum; and generating the probability distribution over the vocabulary of possible outputs by applying a softmax to the gated logits for the possible outputs in the vocabulary.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training an action selection neural network that is used to select actions to be performed by a reinforcement learning agent interacting with an environment. In particular, the actions are selected from a continuous action space and the system trains the action selection neural network jointly with a distribution Q network that is used to update the parameters of the action selection neural network.

Abstract: Various examples described herein are directed to systems and methods that learn to evaluate data items. A computing device receives data items and evaluates the data items. The evaluation performed by the computing device includes comparing a first data item with a second data item and determining a difference between the first data item and the second data item based on the comparison. The computing device also prepares a recommendation based on the difference between the first data item and the second data item and forwards the recommendation to a subject matter expert. The computing device receives an updated recommendation from the subject matter expert that is based on the recommendation. Using the updated recommendation, the computing device refines the comparing and determining operations where the computing device further learns to evaluate the data items by receiving the updated recommendation and refining the comparing and determining operations.

Abstract: Embodiments of the present disclosure provide a method and apparatus for performing word segmentation on a text, a device and a medium, which relate to the field of data processing technology and particularly to a smart search technology. The method may include: dividing a to-be-segmented text into at least two layers of character fragment combinations, any layer of character fragments being child character fragments of a previous layer of character fragments and/or parent character fragments of a next layer of character fragments; and segmenting the to-be-segmented text according to a target word granularity based on the at least two layers of character fragment combinations.

Abstract: A system and method for automatically and algorithmically resolving service tickets through the utilization of historical solution data obtained from multiple sources. The system and method are optionally capable of providing and executing a BOT capable of implementing one or solutions based on the automatic and algorithmic recommendation of a solution to resolve the service ticket(s).

Abstract: Techniques for text evaluation are provided. A curated dataset comprising a plurality of textual documents is received. A tree of classifiers is trained, based on the curated dataset, to identify use cases. A feature graph model is generated, based on the curated dataset, to determine textual similarity. A new document is received, and a plurality of requirements is extracted from the new document. For each requirement, one or more vector scores are generated by evaluating the requirement using the tree of classifiers, one or more feature scores are generated by evaluating the requirement using the feature graph model, and one or more use cases are identified for the new textual document based on the one or more vector scores and the one or more feature scores. An implementation is generated for the new document based on the one or more use cases.

Abstract: Methods, systems, and computer-readable storage media for receiving, by an automated review system, a legal document as a computer-readable file, and determining, by the automated review system, that the legal document is of a first type, and in response: converting the legal document to a set of images, extracting text data from one or more images in the set of images, the text data including sub-sets of text data, each sub-set of text data representing text in a respective clause of a set of clauses of the legal document, for each sub-set of text data receiving a prediction from a machine learning (ML) model in a set of ML models, the ML model being specific to a clause in the set of clauses, and outputting a set of predictions and respective prediction values for display in a user interface (UI).

Abstract: A distributed Industrial Virtual Assistant (IVA) platform and method that that allows operational staff at industrial facilities to have natural language conversations with an IVA to obtain information about, and to control operations of, industrial facilities. For each facility, the IVA platform contains a facilities asset database, one or more domain-specific knowledge databases, a natural language processing engine, and machine learning algorithms for both knowledge domains and context which update the domain-specific knowledge databases based on continuing interactions with operational staff over time. The IVA platform is capable of generating user-specific and device-specific IVA instances that can be queried by voice or text independently of the platform but can connect to the platform to update the databases based on new knowledge, exceptions, and incidents.

Abstract: Systems, devices, and methods are provided for using an automated assessment and evaluation of machine translations. A system may receive a request associated with translating first content from a first language to a second language. The system may translate the first content from the first language to the second language. The system may determine an attribute associated with the first content. The system may determine a translation score associated with second content translated from the first language to the second language and associated with the attribute, the translation score indicative of a machine translation accuracy. The system may determine, and based on the translation score, a translation protocol, and may execute the translation protocol.