Abstract: Various embodiments of the present disclosure provide a text interpretation technique. The text interpretation technique includes generating a plurality of text segment embeddings from a plurality of input text documents and identifying prompt embeddings associated with a predictive task. The technique includes generating task-specific similarity scores for the plurality of text segments based on a comparison between the plurality of text segment embeddings and the prompt embeddings, The technique includes identifying a set of task-specific text segments from the plurality of text segments based on the plurality of task-specific similarity scores and training a target machine learning model based on the set of task-specific text segments.

Abstract: Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for performing data denoising. Certain embodiments of the present invention utilize systems, methods, and computer program products that perform data denoising by utilizing at least one of encoder transformer machine learning models, decoder transformer machine learning models, contextual relevance determination non-linear machine learning models, contextual relevance decision-making machine learning models, denoising decision-making machine learning model, and denoising decision gates.

Abstract: Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for generating a representative embeddings for a plurality of temporal sequences by using a graph attention augmented temporal network based at least in part on dynamic co-occurrence graphs for preceding temporal sequences and initial embeddings, where the dynamic co-occurrence graphs are projections of a global guidance co-occurrence graph on features of the preceding temporal sequences, and the initial embeddings are generated by processing a latent representation of corresponding features that is generated by a sequential long short term memory model as well as a feature tree using a tree-based long short term memory model.

Abstract: Various embodiments of the present disclosure provide an automated language model-based training framework for machine learning classifiers. The techniques comprise receiving a data request that includes a search parameter to be applied to a plurality of data elements, generating a search embedding corresponding to the search parameter via an encoder language model, detecting a first subset of data elements among the plurality of data elements that comprise corresponding embeddings that align with the search embedding in accordance with a first threshold, detecting a second subset of data elements among the first subset of data elements that comprise corresponding classification scores that satisfy a second threshold, and transmitting one or more data packets that comprises one or more data elements of the second subset of data elements in response to the data request.

Abstract: Embodiments of the present disclosure provide for improved data processing using interconnected variational autoencoder models, which may be used for any of a myriad of purposes. Some embodiments specially train the interconnected variational autoencoder models by utilizing different training scenarios corresponding to presence and/or absence of particular data in a training data set. Particular encoder(s) and/or decoder(s) from the specially trained interconnected variational autoencoder models may then be utilized to improve accuracy of the desired data processing tasks, for example, to generate particular output data.

Abstract: Various embodiments of the present disclosure provide reinforcement learning for machine learning using dynamic confidence thresholds. In one example, an embodiment provides for generating a plurality of training datasets for a machine learning model by augmenting a labeled dataset for the machine learning model with a synthetic labeled dataset, generating a plurality of retrained model versions of the machine learning model based on the plurality of training datasets, generating a reward indicator for a retrained model version of the plurality of retrained versions of the machine learning model based on a comparison between a validation dataset for the machine learning model and a respective output dataset for the retrained model version, and modifying the defined confidence threshold based on the reward indicator for the retrained model version to generate a modified confidence threshold for the machine learning model.

Abstract: Various embodiments of the present disclosure disclose machine-learning based evaluation techniques for detecting feature bias. An evaluation framework is provided that utilizes new evaluation data structures for comprehensibly evaluating feature bias in machine learning models. The evaluation framework includes receiving evaluation dataset for a machine learning model that includes one or more different feature classes of an evaluation feature. The evaluation framework includes generating, using an evaluation function, at least two performance metrics for the machine learning model and generating a unitless dissimilarity metric for the evaluation feature based at least in part on the first performance metric, the second performance metric, and an average between the first and second performance metrics. In this way, the unitless dissimilarity metric is based on a variability associated with the evaluation feature.

Abstract: Various embodiments of the present disclosure disclose machine-learning based evaluation techniques for detecting feature bias. An evaluation framework is provided that utilizes new evaluation data structures for comprehensibly evaluating feature bias in machine learning models. The evaluation framework includes receiving evaluation dataset for a machine learning model that includes one or more different feature classes of an evaluation feature. The evaluation framework includes generating, using an evaluation function, at least two performance metrics for the machine learning model and generating a unitless dissimilarity metric for the evaluation feature based at least in part on the first performance metric, the second performance metric, and an average between the first and second performance metrics. In this way, the unitless dissimilarity metric is based on a variability associated with the evaluation feature.

Abstract: Various embodiments of the present disclosure provide methods, apparatus, systems, computing devices, computing entities, and/or the like for detecting and locating vulnerabilities in source code. The method comprises receiving one or more source code files, matching source code from the one or more source code files to one or more program slices by parsing the source code and mapping one or more portions of the source code to the one or more program slices, wherein each of the one or more program slices comprises one or more program statements associated with one or more vulnerabilities, and generating, using a predictive machine learning model, a vulnerability prediction for each of the one or more source code files, the vulnerability prediction comprising one or more locations of vulnerable code in the source code based on the matching and a vulnerability class associated with each location of vulnerable code.

Abstract: Embodiments utilize interpolation to smooth animation frames by generating synthetic Shapley values utilized to generate the animation. Additionally, some embodiments provide for generation of an embedding space for improved understandability of the model training based on corresponding Shapley values. The embedding space is mapped to using activations derived from a specially configured LSTM for each model, where an embedded representation of the LSTM activations is generated by a specially configured autoencoder such that the embedded representation may be mapped to the embedding space. The embedding space and animations are renderable to an interface for user investigation and may be further customizable to narrow investigation on particular data thereof.

Abstract: Various embodiments of the present disclosure provide machine learning training techniques for implementing a multi-modal interpretation process to generate holistic outputs for an event. The techniques may include generating, using first layers of a multi-modal machine learning model, text-based intermediate representations for an entity based on textual input data. The techniques include generating, using second layers of the multi-modal machine learning model, image-based intermediate representations for the entity based on the text-based intermediate representations and input images for the entity. The techniques include generating, using one or more third layers of the multi-modal machine learning model, an entity representation summary based on the image-based intermediate representations and an image narrative summary for the input images. The techniques include initiating the performance of a prediction-based action based on the entity representation summary.

Abstract: Various embodiments of the present invention disclose techniques for traversal path optimization given a traversal network comprising a group of nodes comprising a plurality of navigation orchestration nodes and using a traversal path optimization machine learning model. In some embodiments, a path feature set is determined for each candidate traversal path of a plurality of candidate traversal paths. A traversal path optimization machine learning model is configured to generate path scores for each candidate traversal path based at least in part on the path feature set.

Abstract: Techniques for data clustering and outcome generation may comprise determining a state sequence associated with a sample of one or more samples that satisfies an occurrence threshold and generating one or more timing features for the state sequence. The techniques may further comprise clustering, by executing a machine-learned model, the sample corresponding to at least the state sequence into a cluster of a set of clusters based on the state sequence and the one or more timing features. The techniques may further comprise generating a data object indicating one or more outcomes associated with the cluster that includes the sample. These data objects indicate outcomes that are specific enough to provide meaningful insights contained in associations between the input data and the associated clusters and eliminate/reduce the misleading information commonly output by typical techniques.

Abstract: Techniques for automatically detecting inactive addresses of providers using unsupervised learning approaches are provided. The techniques include determining an activity trend of an address. Responsive to determining that the address is associated with increasing activity, determining an active metric of the address, which indicates a likelihood of the address being active. Responsive to determining that the address is associated with decreasing activity, determining an inactive metric of the address, which indicates a likelihood of the address being inactive. The techniques further include determining whether the address is active or inactive based on the active metric and/or inactive metric. In some embodiments, the active metric or inactive metric is a weighted sum of z scores determined based on Gaussian distributions generated with respect to various benchmarks or provider features.

Abstract: Various embodiments of the present disclosure provide methods, apparatus, systems, computing devices, computing entities, and/or the like for providing suggestion keywords based on historical search data of a user by: generating one or more keyword feature vectors associated with a plurality of keywords from a list of suggestion keywords, generating one or more personalized feature vectors associated with the user based on search session data, generating a plurality of predictions of the user selecting the plurality of keywords based on the one or more keyword feature vectors and the one or more personalized feature vectors, assigning a plurality of rankings to the plurality of keywords based on the plurality of prediction probabilities, and generating one or more typeahead suggestion keywords based on the plurality of rankings.

Abstract: A method includes receiving, by one or more processors, a dataset including transition data and factor data. The method includes generating a feature for a machine learning model based on the transition data, generating, via input of at least the feature into the machine learning model, one or more data objects indicative of a transition prediction for a transition from the first stage to the second stage, the machine learning model having been trained: with data sources including training factor data having information other than a chemical constituent of blood, and to output information associated with a transition prediction. The method further includes initiating performance of one or more remedial or analytical actions in response to generating the one or more data objects indicative of the transition prediction.

Abstract: Disclosed are systems and methods for detecting a table from a document, classifying each cell of the table as one of: a value cell, a row header cell, or a column header cell, and performing a bounding box elongation operation to match each cell that is classified as a value cell to a first corresponding cell that is identified as a row header cell and a second corresponding cell that is identified as a column header cell. For each cell classified as a value cell, a data tuple is generated comprising a row header element, a column header element, and a value element, wherein the row header element corresponds to a first value in the first corresponding cell, the column header element corresponds to a second value in the second corresponding cell, and the value element corresponds to a third value in the value cell.

Abstract: A method includes performing by a host system processor: providing a database including a plurality of records, the database having at least one attribute associated therewith; determining a plurality of maximum block sizes as a plurality of ideal maximum numbers of the plurality of records in a plurality of blocks, respectively, based on a plurality of blocking keys; determining a duplication factor based on a number of unique records of the plurality of records based on all of the at least one attribute; and generating candidate pairs of the plurality of records for linkage based on the plurality of maximum block sizes and the duplication factor.

Abstract: A method includes receiving, by one or more processors, a clinical image associated with a dental procedure; identifying, by the one or more processors, one or more dental procedure codes based on processing the clinical image using a plurality of AI models; receiving, by the one or more processors, a periodontal chart image; processing, by the one or more processors, the periodontal chart image using optical character recognition to obtain pocket measurements associated with a plurality of teeth along with positional coordinates of each of the pocket measurements; identifying, by the one or more processors, a submitted dental procedure code in a dental claim for the dental procedure; determining, by the one or more processors, whether the submitted dental procedure code corresponds to a visibly detectable procedure; determining, by the one or more processors, whether the submitted dental procedure code matches any of the one or more dental procedure codes based on processing the clinical image when the submi

Abstract: As described herein, various embodiments of the present invention use an output space refinement machine learning model to filter C of the B candidate predictive associations that are referred to herein as dynamically-preselected candidate predictive associations for each prediction input data object, with every prediction input data object being associated with a different subset of the dynamically-preselected candidate predictive associations, where C is less than B and is in some embodiments typically much less than B. As described in greater detail below, this approach reduces the number of computational operations that need to be performed by a final classification machine learning model (referred to herein as a variable-output-space prediction machine learning model), and leads to substantial computational efficiency advantages relative to naïve implementations.