Abstract: An upper-level integrated processing device recognizes an inquiry, generates a search request in a primitive form to be output to each of the individual AI search devices in response to the inquiry, receives an individual reply and the probability thereof corresponding to the search request in the primitive form from a plurality of individual AI search devices, normalizes the probabilities of the individual replies from the individual AI search devices to acquire normalized probabilities, and generates an inquiry reply corresponding to the individual replies and the normalized probabilities thereof, and the lower-level individual AI search device searches Individual AI database using artificial intelligence in response to reception of the search request in the primitive form, acquires an individual reply and the probability thereof, and outputs the same in the primitive form to the integrated processing device. The upper-level devices output reply using a plurality of AI search devices in combination.

Abstract: A computer-implemented method, a computer system and a computer program product create rules for a rule-based natural language interface for databases (NLIDB). The method may include receiving a natural language query from a user. The method may also include generating a first explanation for the natural language query using a deep learning model and a second explanation for the natural language query using the rule-based NLIDB and validating whether the first and second explanations correctly represent the natural language query. The method may further include identifying the database value in the first explanation in response to the first explanation correctly representing the natural language query and the second explanation not correctly representing the natural language query. Lastly, the method may include creating a rule in a table for the rule-based natural language interface for databases that associates the database value with the original word of the natural language query.

Abstract: Existing research on cross-lingual retrieval cannot take good advantage of large-scale pretrained language models, such as multilingual BERT and XLM. The absence of cross-lingual passage-level relevance data for finetuning and the lack of query-document style pretraining are some of the key factors of this issue. Accordingly, embodiments of two novel retrieval-oriented pretraining tasks are presented herein to further pretrain cross-lingual language models for downstream retrieval tasks, such as cross-lingual ad-hoc retrieval (CUR) and cross-lingual question answering (CLQA). In one or more embodiments, distant supervision data was constructed from multilingual texts using section alignment to support retrieval-oriented language model pretraining. In one or more embodiments, directly finetuning language models on part of an evaluation collection was performed by making Transformers capable of accepting longer sequences.

Abstract: A flexible listings search system can receive and return results for flexible listing searches. For example, the system can perform micro-flexible searches (e.g., plus or minus a few days) or super flexible searches (e.g., a time span in one or more months), using listing arrays that can be rapidly accessed to efficiently identify and return results. The search system can further perform flexible destination searches for different categories of accommodations for display in a viewport (e.g., map bounding box). The system can further perform fuzzy searches to identify and return broader results for flexible queries.

Abstract: A method for compressing molecular tagged sequence data includes: grouping sequence reads associated with a molecular tag sequence to form a family of sequence reads, corresponding vectors of flow space signal measurements and corresponding sequence alignments, calculating an arithmetic mean of the corresponding vectors of flow space signal measurements to form a vector of consensus flow space signal measurements, calculating a standard deviation of the corresponding vectors of flow space signal measurements to form a vector of standard deviations, determining a consensus base sequence based on the vector of consensus flow space signal measurements, determining a consensus sequence alignment and generating a compressed data structure comprising consensus compressed data, the consensus compressed data including for each family, the consensus base sequence, the consensus sequence alignment, the vector of consensus flow space signal measurements, the vector of standard deviations and the number of members.

Abstract: A file versioning management method and a file system executing the method are provided. The method includes: writing a file into a first file path, wherein the first file path includes a first folder, generating a second file path according to the first file path, wherein the second file path includes a second folder at least one level under the first folder; and creating a link pointing from the file of the first file path to a next level of the second folder so as to generate a version information file at the next level of the second folder and make the link point to the version information file.

Abstract: From a set of natural language text documents, a concept tree is constructed. For a node in the concept tree a polarity of the subset represented by the node is scored. A second set of natural language text documents is added to the subset, the adding resulting in a modified subset of natural language text documents having a polarity score within a predefined neutral polarity score range. From the modified subset, a bin of sentences is selected according to a sentence selection parameter, a sentence in the bin of sentences being extracted from a selected document in the modified subset. A sentence having a factuality score below a threshold factuality score is removed from the bin of sentences. From the filtered bin of sentences a new natural language text document corresponding to the filtered bin of sentences is generated using a transformer deep learning narration generation model.

Abstract: In an embodiment, a computer stores source files and source clone files in a source filesystem. The source clone files are shallow copies of the source files, which initially share the same data blocks as the source files. A source file or a source clone file may eventually be modified, where some of the shared blocks are replaced by modified blocks. In a first phase, all data blocks of all source files that are not clones are replicated to a standby filesystem. A second phase compares each clone file on the source filesystem with its base file to detect a set of differed blocks. The second phase copies the differed blocks into the standby filesystem and, in the standby filesystem, applies the differed blocks to a corresponding clone file. Efficiency of keeping the standby filesystem synchronized with the source filesystem is improved by transferring, from the source filesystem to the standby filesystem, only one copy of only data blocks that are modified.

Abstract: Systems and methods for indexing data include formulating an objective function to index a dataset, a portion of the dataset including supervision information. A data property component of the objective function is determined, which utilizes a property of the dataset to group data of the dataset. A supervised component of the objective function is determined, which utilizes the supervision information to group data of the dataset. The objective function is optimized using a processor based upon the data property component and the supervised component to partition a node into a plurality of child nodes.

Abstract: The present disclosure relates to systems and methods for an intelligent assistant (e.g., a chatbot) that can be used to enable a user to generate a machine learning system. Techniques can be used to automatically generate a machine learning system to assist a user. In some cases, the user may not be a software developer and may have little or no experience in either machine learning techniques or software programming. In some embodiments, a user can interact with an intelligent assistant. The interaction can be aural, textual, or through a graphical user interface. The chatbot can translate natural language inputs into a structural representation of a machine learning solution using an ontology. In this way, a user can work with artificial intelligence without being a data scientist to develop, train, refine, and compile machine learning models as stand-alone executable code.

Abstract: In an approach to improve file recall between storage devices by optimizing file recall for multiple users. Embodiments of the present invention obtain one or more similarities among one or more users based on one or more usage points and divide the one or more users into a plurality of clusters based on a content of the one or more usage points. Further, embodiments select a cluster from the plurality of clusters based on cluster importance, and select the content used by more than a predetermined percentage of the weighted users pertaining to the selected cluster as an automated recall subject. Additionally, embodiments recal one or more selected content from a secondary storage device to a primary storage device.

Abstract: Techniques for efficiently managing a file clone from a filesystem which supports efficient volume snapshots are provided. In some embodiments, a system may receive an instruction to remove the file clone from the filesystem. The file clone may be a point-in-time copy of metadata of an original file. The system may further—for a file map entry in a filesystem tree associated with the file clone, the file map entry indicating a data block—decrement a reference count in a reference count entry associated with the file map entry. The reference count entry may be stored in the filesystem tree according to a key and the key may comprise an identification of the original file. The system may further reclaim the data block in a storage system when the reference count is zero.

Abstract: An active layer of the index includes a first shard group, and shards in the first shard group are configured to store indexes of a part of data objects in a streaming storage system. In response to determining that the state of the first shard group meets a predetermined expansion condition, a second shard group is created in the index, and shards in the second shard group are configured to store indexes of data objects that will enter the storage system. The storage system is managed based on the shards in an active layer (where the second shard group is located) and frozen layers (where the second shard group) in the index. The number of shards in the storage system can be dynamically set to process streaming data at a relatively high speed, and it is suitable for processing streaming data that continuously enters the storage system.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, that creates a machine learning model with k-means clustering. In some implementations, an instruction to create a model is obtained. A data set including geographic data and non-geographic data is received. The data set includes multiple data entries. Geographic centroids are determined from the geographic data. The data set is analyzed to obtain statistics of the data set. Transformed data is generated from the data set, the statistics, and the geographic centroids. A model is generated with the transformed data, the model indicating multiple data groupings.

Abstract: Systems and methods are disclosed for processing data associated with isolated execution environments. A chunk of data associated with an isolated execution environment can include log data and non-log data. At least a portion of the log data can include log data generated by the isolated execution environment. The system can parse the chunk of data to identify the log data and the non-log data and extract at least a portion of the log data from the chunk of data. The extracted data can be further processed to generate one or more events.

Abstract: A system, method, and computer program product for treatment planning are disclosed. The system includes at least one processing component, at least one memory component, a training module, a retrieval module, and a plan generator. The training module generates hash codes by hashing features from data sources with data source-specific hash functions, and generates superclass hash codes by hashing the generated hash codes with at least one superclass hash function. The retrieval module extracts features from case data, and locates features from the data sources that are similar to the extracted features. The plan generator calculates outcome probabilities for the case data based on known outcomes associated with the located features.

Abstract: A machine-learning model may be previously trained with a supervised learning algorithm to identify whether a pair of labels provided as input are similar. A locality sensitive hashing forest (LSH) may be generated for the set of candidate labels. When a user later identifies an input label (e.g., by search query, by interface selection, etc.) the input label may be used to query the LSH forest to identify a subset of the candidate labels. This subset may be used to generate respective pairs comprising the input label, one of the subset candidate labels, and a corresponding feature set generated for the pair. This data may be provided to the model to identify a degree to which the pair of labels are similar. The user may be provided one or more recommendations including similar terms identified from the model's output.

Abstract: A method for personalizing reviews for a user on a social network, comprising the steps of: receiving from at least two or more users a rating and a recommendation for a reviewable item, wherein the rating is numerical value in a rating scale, and wherein a recommendation is a binary endorsement; receiving a request from a user to view the ratings and recommendations for one or more reviewable items; determining a combined rating and an combined recommendation for all users of the social network at one or more social degrees for the one or more reviewable items; and displaying, by a user interface, the combined rating and combined recommendation for all users of the social network at one or more social degrees for the one or more reviewable items.

Abstract: A computer-implemented method is provided. The method comprises: receiving, from a text field of a user interface, input text strings containing medical data; identifying, based on language semantics and grammatical structure, a keyword and one or more data values from the input text strings; providing the keyword as an input to a query of a medical data category database to obtain one or more categories associated with the keyword, the one or more categories including a first category; retrieving, from a template database, a first template associated with the first category, the first template including one or more fixed blocks and one or more variable blocks; inserting a data value corresponding to the keyword into a variable block of the first template to generate a replacement text string; and displaying the replacement text string in place of the input text strings in the text field.

Abstract: A method and apparatus for generating personalized suggestions for natural language search queries, where the method includes receiving a natural language query input from a user, obtaining set of suggestions for the natural language query, identifying a set of concepts in the set of suggestions, applying co-occurrence model to the set of concepts to obtain user specific concept probabilities and vocabulary probabilities, computing suggestion scores for at least one suggestion in the set of suggestions based on the user specific concept probabilities and vocabulary probabilities, and ranking the set of suggestions based on suggestion scores.