Abstract: Systems, methods, and computer-readable media are disclosed for visual labeling of training data items for training a machine learning model. Training data items may be generated for training the machine learning model. Visual labels, such as QR codes, may be created for the training data items. The creation of the training data item and the visual label may be automated. The visual labels and the training data items may be combined to obtain a labeled training data item. The labeled training data item may comprise a separator to distinguish the training data item from the visual label. The labeled training data item may be used for training and validation of the machine learning model. The machine learning model may analyze the training data item, attempt to identify the training data item, and compare the identification against the embedded label.

Abstract: Text-to-speech translation is used to generate a transcript for an audio file. Text segments are associated with time segments in the transcript. A trained machine learning model determines, based on the text in the transcript, one or more topics for the audio file. The transcript is modified to include the determined one or more topics. A user interface may be presented that allows a user to search for portions of an audio file that relate to a particular topic. In response to the selected or entered topic, the user interface presents segments having a matching topic. The user may use voice or other user interface commands to modify the annotation of the audio file. User commands may also be used to extract data from the transcript and copy the data to a clipboard or to another application.

Abstract: A bipartite graph is created that represents related metadata and datasets. The graph comprises disjoint and independent sets D (datasets) and M (metadata entries). For each dataset in the collection that has corresponding metadata, a relation is created in the graph between the node in D for the dataset and the node in M for the metadata. A vector embedding is generated for each dataset and metadata entry. New relations are added to the graph to relate datasets having similar embeddings. The graph is enriched by the new relations, enabling new kinds of search queries as well as recommendations and suggestions.

Abstract: Datasets are available from different dataset servers and often lack well-defined metadata. Thus, comparing datasets is difficult. Additionally, there might be different versions of the same dataset which makes the search even more difficult. Using systems and methods described herein, quality scores, dataset versioning, topic identification, and semantic relatedness metadata is stored about datasets stored on dataset servers. A user interface is provided to allow a user to search for datasets by specifying search criteria (e.g., a topic and a minimum quality score) and to be informed of responsive datasets. The user interface may further inform the user of the quality scores of the responsive datasets, the versions of the responsive datasets, or other metadata. From the search results, the user may select and download one or more of the responsive datasets.

Abstract: Example methods and systems are directed to automatic data card generation for datasets. A data card is a summary that describes quantitative aspects of a dataset, qualitative aspects of a dataset, or both. The data samples and documentation of a dataset are analyzed automatically to determine a number of samples, a primary data type, a license, or any suitable combination thereof. Data formats for data and documentation of the dataset may be automatically recognized. Language of text data may be automatically recognized. The most frequent language for the text data may be identified as the primary language of the dataset. A data card may be created for the dataset. The data card may indicate the number of samples, the data formats used in the data set, the language of text data in the dataset, or any suitable combination thereof.

Abstract: An input image is divided into segments. The segments may be reassembled to reform the input image. The order of the segments may be stored in an encrypted database for which approved applications have the decryption key but users do not. This allows the approved applications to determine the order and reform the input image without allowing users to do the same. To further increase the difficulty of reforming the input image, the segments may be transformed. Example transformations include rotation and mirroring. The encrypted database may store an indication of the transformation applied to each segment. The effort of reforming the input image without access to the database is increased substantially. The reformed input image may be stored in transient memory only, without being stored to long-term storage. Thus, the reformed image cannot be accessed from a file system by unauthorized users.

Abstract: To perform 3-dimensional interpolation, a 3-dimensional model of an input text character is generated. For example, a 2-dimensional character may be given depth using an extrusion transformation. The 3-dimensional model of the input text character is compared to 3-dimensional models of candidate characters and the results of the 3-dimensional comparisons are used to select the optical character recognition (OCR) output for the input text character. The 3-dimensional comparison may be performed directly on the 3-dimensional models. Alternatively, a set of 2-dimensional images may be generated for each 3-dimensional model and 2-dimensional comparisons performed. By use of the additional information gathered from the comparisons of the 3-dimensional models, the correct OCR output character can be identified with greater confidence.

Abstract: Example methods and systems are directed to intelligent quick response (QR) code compression. Different versions of QR codes comprise different numbers of modules and represent different amounts of text. To ensure that a QR code can be read correctly, the minimum printed size of the QR code varies with the version. As described herein, intelligent QR code compression involves converting a QR code to text, compressing the text, and generating a smaller QR code that represents the compressed text. The resulting QR code may be printed at a smaller size or stored using less memory than the original QR code. Text processing may include sentence splitting, sentence ranking, and key phrase detection. The compressed text comprises one or more detected key phrases. The amount of compression may be configurable, such that greater compression results in less original information being included in the resulting QR code.

Abstract: Example methods and systems are directed to generating knowledge graph entities from text. Natural language text is received as input and processed using named entity recognition (NER), part of speech (POS) recognition, and business object recognition (BOR). The outputs of the NER, POS, and BOR processes are combined to generate knowledge entity triples comprising two entities and a relationship between them. Keywords are extracted from the text using NER to generate a set of entities. A node in a knowledge graph is created for at least some of the entities. A POS tagger identifies verbs in the text, generating a set of verbs. Relational verbs (e.g., “talk to” or “communicated with”) are detected and used to create edges in the knowledge graph. The knowledge graph may be converted back to natural language text using a trained machine learning model.

Abstract: A machine learning model is trained to translate source code from one or more programming languages into a common programming language. The machine learning model translates source code from the other languages into the common programming language. A language embedder generates a vector for each function in the source code, all of which is now in the common programming language. A user provides a text search query which is converted by a language embedder to a vector. Based on the vector of the text search query and the vectors for the source code, search results are generated and presented in a user interface. Additional machine learning models may be trained and used to measure function complexity, test coverage, documentation quantity and complexity, or any suitable combination thereof. These measures may be used to determine which search results to present, an order in which to present search results, or both.

Abstract: In an example embodiment, machine learning is used to intelligently compress documents to reduce the overall footprint of storing large amounts of files for an organization. Specifically, a document is split into parts, with each part representing a grouping of text or an image. Optical character recognition is performed to identify the text in images. Machine learning techniques are then applied to a part of a document in order to determine how relevant the document is for the organization. The parts that are deemed to be not relevant may then be reduced in size, either by omitting them completely or by summarizing them. This allows for the compression to be tailored specifically to the organization, resulting in the ability to compress or eliminate parts of documents that other organizations might have found relevant (and thus would not have been compressed or eliminated through traditional means).

Abstract: Systems, methods, and computer-readable media are disclosed for list attribute normalization and standardization for creation of a controlled vocabulary. A vocabulary set comprising a plurality of vocabulary term may be received. For each vocabulary term, semantic duplicates may be identified. The semantic duplicates may be identified by analyzing semantics, syntactics, or phonetics of the vocabulary terms. Semantic chains may be formed from each vocabulary term and the corresponding semantic duplicates. The terms in each semantic chain may be ranked to determine a most probable vocabulary term. The most probable vocabulary term may then replace the semantic chain. The most probable vocabulary term across all semantic chains from the vocabulary set may form the controlled vocabulary.

Abstract: Systems, methods, and computer-readable media are disclosed for visual labeling of training data items for training a machine learning model. Training data items may be generated for training the machine learning model. Visual labels, such as QR codes, may be created for the training data items. The creation of the training data item and the visual label may be automated. The visual labels and the training data items may be combined to obtain a labeled training data item. The labeled training data item may comprise a separator to distinguish the training data item from the visual label. The labeled training data item may be used for training and validation of the machine learning model. The machine learning model may analyze the training data item, attempt to identify the training data item, and compare the identification against the embedded label.

Abstract: Source code is analyzed to identify components. The components are each assigned a complexity score. Documentation for the source code is identified, related to the components, and given a score based on the quantity of the documentation for the component and the complexity score for the component. To determine semantic meaning of the documentation, vector embeddings for the documentation languages may be generated and aligned. Alignment causes the different machine learning models to generate similar vectors for semantically similar words in the different languages. Since the vectors of the words of the other languages are similar to the vectors of the words in a primary language with similar meanings, the vector representation of the documentation in the other languages will match the vector representation of the source code when the documentation is substantially on the same topic.

Abstract: Datasets are available from different dataset servers and often lack well-defined metadata. Thus, comparing datasets is difficult. Additionally, there might be different versions of the same dataset which makes the search even more difficult. Using systems and methods described herein, quality scores, dataset versioning, topic identification, and semantic relatedness metadata is stored about datasets stored on dataset servers. A user interface is provided to allow a user to search for datasets by specifying search criteria (e.g., a topic and a minimum quality score) and to be informed of responsive datasets. The user interface may further inform the user of the quality scores of the responsive datasets, the versions of the responsive datasets, or other metadata. From the search results, the user may select and download one or more of the responsive datasets.

Abstract: Text-to-speech translation is used to generate a transcript for an audio file. Text segments are associated with time segments in the transcript. A trained machine learning model determines, based on the text in the transcript, one or more topics for the audio file. The transcript is modified to include the determined one or more topics. A user interface may be presented that allows a user to search for portions of an audio file that relate to a particular topic. In response to the selected or entered topic, the user interface presents segments having a matching topic. The user may use voice or other user interface commands to modify the annotation of the audio file. User commands may also be used to extract data from the transcript and copy the data to a clipboard or to another application.

Abstract: Files are automatically named based on their contents and metadata. Contents include words in a text file, text recognized using optical character recognition (OCR) in an image file, and objects recognized using object recognition in an image file. Metadata includes creation date, modification date, user owning the file, file type, and file extension. Multiple files may be processed. A file sorter may determine an order in which to process the multiple files. For example, smaller files may be processed first. In addition to using the words discussed above to name the file, the file may be tagged based on the contents of the file. A search function for files may search both names and tags to identify responsive files. Two or more files may be linked based on their contents or metadata.

Abstract: A machine learning model is trained to translate source code from one or more programming languages into a common programming language. The machine learning model translates source code from the other languages into the common programming language. A language embedder generates a vector for each function in the source code, all of which is now in the common programming language. A user provides a text search query which is converted by a language embedder to a vector. Based on the vector of the text search query and the vectors for the source code, search results are generated and presented in a user interface. Additional machine learning models may be trained and used to measure function complexity, test coverage, documentation quantity and complexity, or any suitable combination thereof. These measures may be used to determine which search results to present, an order in which to present search results, or both.

Abstract: Source code is analyzed to identify components. The components are each assigned a complexity score. Documentation for the source code is identified, related to the components, and given a score based on the quantity of the documentation for the component and the complexity score for the component. To determine semantic meaning of the documentation, vector embeddings for the documentation languages may be generated and aligned. Alignment causes the different machine learning models to generate similar vectors for semantically similar words in the different languages. Since the vectors of the words of the other languages are similar to the vectors of the words in a primary language with similar meanings, the vector representation of the documentation in the other languages will match the vector representation of the source code when the documentation is substantially on the same topic.

Abstract: Provided herein are a system, apparatus, device, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for identifying attributes in unstructured data files using a machine-learning model. In an embodiment, a server can receive a request to identify an attribute associated with a set of unstructured data files. The server can extract a first and second subset of features from each unstructured data file of the set of unstructured data files. The server can identify the attribute in the set of unstructured data files request based on each of the first and second subset of features using the machine-learning model.