Abstract: Method and apparatus for generating natural language observations using sequence-to-sequence models are provided. The method includes receiving a first electronic document and dynamically generating, without requiring user intervention, a first natural language observation corresponding to a first portion of the first electronic document by processing the first portion of the first electronic document using a first sequence-to-sequence model. A second natural language observation is received for the first portion of the first electronic document, and the generated first natural language observation and the authoritative natural language observation are provided. The method further includes receiving an evaluation of the first natural language observation and the second natural language observation, and refining the first sequence-to-sequence model based on the evaluation of the first natural language observation and second natural language observation.

Abstract: A method, computer system, and a computer program product for designing one or more folded structural proteins from at least one raw amino acid sequence is provided. The present invention may include computing one or more character embeddings based on the at least one raw amino acid sequence by utilizing a multi-scale neighborhood-based neural network (MNNN) model. The present invention may then include refining the computed one or more character embeddings with at least one set of sequence neighborhood information. The present invention may further include predicting one or more dihedral angles based on the refined one or more character embeddings.

Abstract: Mechanisms are provided for executing a trained deep learning (DL) model. The mechanisms receive, from a trained autoencoder executing on a client computing device, one or more intermediate representation (IR) data structures corresponding to training input data input to the trained autoencoder. The mechanisms train the DL model to generate a correct output based on the IR data structures from the trained autoencoder, to thereby generate a trained DL model. The mechanisms receive, from the trained autoencoder executing on the client computing device, a new IR data structure corresponding to new input data input to the trained autoencoder. The mechanisms input the new IR data structure to the trained DL model executing on the deep learning service computing system, to generate output results for the new IR data structure. The mechanisms generate an output response based on the output results, which is transmitted to the client computing device.

Abstract: A method for authenticating a user is presented. Responsive to a request for access to a computer resource, a computer system prompts the user making the request to access the computer resource to perform a new motion in an environment in which the user is monitored by a sensor system. Detected biometric data in the new motion performed by the user is identified by the computer system. A determination is made as to whether the user performing the new motion is an authenticated user based on comparing the detected biometric data with stored biometric data for a prior motion performed by the authenticated user. The computer system provides access to the computer resource when the user is identified as the authenticated user.

Abstract: Systems, methods, and computer-readable media are described for determining a score for a target student answer using unlabeled data. The target answer is provided by a student to a question for which there is no ground-truth answer data. A set of student answers serves as a set of pseudo-reference answers and a classifier is used to score each answer based on each other answer. In this manner, each student answer serves as a pseudo-reference answer for each other student answer. A clustering approach can also be employed to cluster a set of student answers into clusters. The centroids of the clusters can then serve as the set of pseudo-reference answers. Clustering improves the robustness and efficiency of the score determined for a target student answer.

Abstract: Techniques that facilitate layered stochastics anonymization of data are provided. In one example, a system includes a machine learning component and an evaluation component. The machine learning component performs a machine learning process for first data associated with one or more features to generate second data indicative of one or more example datasets within a degree of similarity to the first data. The first data and the second data comprise a corresponding data format. The evaluation component evaluates the second data for a particular feature from the one or more features and generates third data indicative of a confidence score for the second data.

Abstract: Method and apparatus for generating natural language observations using sequence-to-sequence models are provided. The method includes receiving a first electronic document and dynamically generating, without requiring user intervention, a first natural language observation corresponding to a first portion of the first electronic document by processing the first portion of the first electronic document using a first sequence-to-sequence model. A second natural language observation is received for the first portion of the first electronic document, and the generated first natural language observation and the authoritative natural language observation are provided. The method further includes receiving an evaluation of the first natural language observation and the second natural language observation, and refining the first sequence-to-sequence model based on the evaluation of the first natural language observation and second natural language observation.

Abstract: A computer-implemented method executed on a processor for lexicon extraction from non-parallel data is provided. The computer-implemented method includes representing each word of a plurality of words by a vector of documents in which the word appears, modeling each word as a topic distribution by using the vector of documents, receiving a first word in a source language, and finding a second word in a target language as a translation of the first word based on similarity of topic distributions of the first word and the second word.

Abstract: A computer-implemented method executed on a processor for lexicon extraction from non-parallel data is provided. The computer-implemented method includes representing each word of a plurality of words by a vector of documents in which the word appears, modeling each word as a topic distribution by using the vector of documents, receiving a first word in a source language, and finding a second word in a target language as a translation of the first word based on similarity of topic distributions of the first word and the second word.

Abstract: A computer-implemented method executed on a processor for lexicon extraction from non-parallel data is provided. The computer-implemented method includes representing each word of a plurality of words by a vector of documents in which the word appears, modeling each word as a topic distribution by using the vector of documents, receiving a first word in a source language, and finding a second word in a target language as a translation of the first word based on similarity of topic distributions of the first word and the second word.

Abstract: A computer-implemented method executed on a processor for lexicon extraction from non-parallel data is provided. The computer-implemented method includes representing each word of a plurality of words by a vector of documents in which the word appears, modeling each word as a topic distribution by using the vector of documents, receiving a first word in a source language, and finding a second word in a target language as a translation of the first word based on similarity of topic distributions of the first word and the second word.