Abstract: An embodiment senses a raw data sequence by a central processing unit, responsive to the raw data sequence, computes by the central processing unit a transfer data size of the raw data sequence based at least in part on the comparison of a data size of the raw data sequence to a memory size of a graphics processing unit. The embodiment transfers by the central processing unit of the raw data sequence to the graphics processing unit based on the transfer data size. The embodiment trains a foundation model on the raw data sequence where a sliding window algorithm is executed on the raw data sequence by the graphics processing unit, where generating a window of the sliding window algorithm is based on a memory pointer to the raw data sequence.

Abstract: Mechanisms are provided for rendering content in a compacted view. A machine learning computer model is trained by a machine learning process to predict a user attention score for segments of content based on features of the content and historical user attention data. The trained machine learning computer model processes new content to associate with each segment, in a plurality of segments, of the new content, a corresponding user attention score. The segments, in the plurality of segments, of the new content are ranked relative to one another based on the corresponding user attention scores of the segments. A compacted view of the new content is rendered based on the ranking of the segments. A first number of segments are rendered in the compacted view and a second number of segments are not rendered in the compacted view, and are replaced with an inserted user selectable expansion element.

Abstract: Targeted acquisition of data for model training includes identifying attributes of classified samples of a collection of samples classified by a classification model, and generating at least one query based on the identified attributes, the at least one query tailored, based on the attributes, to retrieve additional training data for training the classification model to more accurately classify samples and avoid incorrect sample classification.

Abstract: Debris signature-based robotic cleaning device navigation includes operating the robotic cleaning device in a first mode as part of a vacuum cycle, the device including suction ports configurable for different suction power levels and each port having a suction path along which debris entering through the suction port is collected by the device. In the first operating mode the suction ports are operated at a first suction power level. The navigation also includes changing operation of the device to a second mode and in which the suction ports are operated at a greater suction power, measuring a respective amount of debris collected through each suction port, and selecting a direction in which to navigate the robotic cleaning device based on the debris measurements.

Abstract: Embodiments of the present disclosure provide systems and methods for synchronizing an unaligned audio stream and a corresponding unaligned video stream of real time streaming media. A non-limiting disclosed method comprises performing, using a video classifier, video reference point classification of a video stream based on an audio-video dataset; performing, using an audio classifier, audio reference point classification of the audio stream based on the audio-video dataset. The system correlates object related reference points in video segments of the video stream and in audio segments of the audio stream to identify a set of audio-video synchronization candidates. The system compares context of the set of audio-video synchronization candidates to identify an audio-video synchronization candidate to synchronize the audio stream and video stream based on reference point alignment.

Abstract: Systems and methods for lightweight proxy virtualization of a plurality of sensor data streams in a device are described. A processor can receive a plurality of sensor data streams from a plurality of sensors. The processor can identify missing sensor data in a sensor data stream among the plurality of sensor data streams. The processor can predict a value of the missing sensor data by running a machine learning model trained using sensor data determined based on at least one of a plurality of co-existence probabilities of the plurality of sensor data streams and a plurality of co-prediction accuracies of the plurality of sensor data streams.

Abstract: Embodiments of the present disclosure provide systems and methods for synchronizing an unaligned audio stream and a corresponding unaligned video stream of real time streaming media. A non-limiting disclosed method comprises performing, using a video classifier, video reference point classification of a video stream based on an audio-video dataset; performing, using an audio classifier, audio reference point classification of the audio stream based on the audio-video dataset. The system correlates object related reference points in video segments of the video stream and in audio segments of the audio stream to identify a set of audio-video synchronization candidates. The system compares context of the set of audio-video synchronization candidates to identify an audio-video synchronization candidate to synchronize the audio stream and video stream based on reference point alignment.

Abstract: A method, system, and compute program product are configured to: receive a dataset comprising a multivariate time series that includes plural channels; generate an original forecast of the multivariate time series using a channel-independent backbone and a prediction head; and generate a revised forecast of the multivariate time series using a cross-channel reconciliation head with the original forecast, wherein the cross-channel reconciliation head generates the revised forecast based on correlations between the channels of the multivariate time series.

Abstract: A method for lightweight and efficient long sequence time-series forecasting and representation learning includes segmenting a time-series dataset from a plurality of sensors into a plurality of patches. The method further includes applying gated multilayer perceptron (MLP) mixing across different directions of the patched input time-series. The method further includes capturing local and global and interrelated correlations across the plurality of patches and within the plurality of patches. The method further includes applying a patch-time aggregated hierarchy to guide lowest-level predictions based on aggregated hierarchy signals at a patch-level. The method further includes chaining MLP-mixers in a patch length context aware hierarchy fashion to enhance time-series short and long-term correlation capture.

Abstract: An embodiment includes retrofitting an existing control device with an automation panel that senses manual actuation of a control element of the control device, and automatically actuates the control element in response to a specified control signal. The embodiment collects state data indicative of an actuation state of the control element and context data of conditions at a time that the state data is collected, and generates a training dataset comprising collected state data and sensor data. It then uses this data to train a classification model to predict a control element actuation state based on sensor data. The embodiment deploys the trained classification model to process sensor data and upon detecting a mismatch between a predicted actuation state output from the trained classification model and an actual actuation state of the control element, the embodiment transmits the specified control signal to the automation panel to actuate the control element.

Abstract: A computer-implemented method, a computer program product, and a computer system for synchronizing audio and video using pause gap analysis. A computer splits a video into an audio stream and a video stream. A computer identifies time points at which there is no sound in the audio stream and derives pause gaps in the audio stream. A computer applies a binary classifier to predict sound presence or absence in frames of the video stream and derives pause gaps in the video stream. A computer identifies desynchronization between the pause gaps in the video stream and the pause gaps in the audio stream. A computer aligns the pause gaps in the video stream with the pause gaps in the audio stream, based on metadata of the pause gaps in the video stream.

Abstract: Inadvertent data swaps can be prevented by measuring volume of transactions in distributed computing environment to determine locations for potential data swaps; and managing a correlation between a thread identification (ID) and transaction header (ID) for transactions in the distributed computing environment. In some embodiments, the prevention of data swaps can further include performing a data transmission interruption to avoid data swaps at the locations for potential data swaps. When the thread identification (ID) and transaction header (ID) do not match the potential for data swaps can be high.

Abstract: Approaches presented herein enable maneuvering collaborative robots to rescue persons in a hydrological disaster. A plurality of robots are dispersed in a body of water to spread out and seek victims using cooperative foraging techniques within resource constraints. A location of victims located by a robot using sensing techniques is communicated to other robots. A situational assessment is performed using victim location information to determine a number of robots to deploy to the location. The deployed robots are directed to perform coordinated maneuvers to create a connected floatation unit to support floatation of victims for rescue.

Abstract: Mitigating mistranscriptions resolves errors in a transcription of the audio portion of a video based on a semantic matching with contextualized data electronically garnered from one or more sources other than the audio portion of the video. A mistranscription is identified using a pretrained word embedding model that maps words to an embedding space derived from the contextualizing data. A similarity value for each vocabulary word of a multi-word vocabulary of the pretrained word embedding model is determined in relation to the mistranscription. Candidate words are selected based on the similarity values, each indicating a closeness of a corresponding vocabulary word to the mistranscription. The textual rendering is modified by replacing the mistranscription with a candidate word that, based on average semantic similarity values, is more similar to the mistranscription than is each other candidate word.

Abstract: Targeted acquisition of data for model training includes identifying attributes of classified samples of a collection of samples classified by a classification model, and generating at least one query based on the identified attributes, the at least one query tailored, based on the attributes, to retrieve additional training data for training the classification model to more accurately classify samples and avoid incorrect sample classification.

Abstract: An example operation may include one or more of storing a hierarchical data set, receiving a plurality of predicted outputs from a plurality of nodes in a distributed computing environment, respectively, wherein each predicted output is generated by a different node via execution of a time-series forecasting model on a different subset of lowest level data in the hierarchical data set, combining the plurality of predicted outputs via bottom-up aggregation to generate one or more additional predicted outputs for the time-series forecasting model based on one or more levels above the lowest level in the hierarchical time-series data set, determining error values for the time-series forecasting model at each level of the hierarchical data set based on the received and the one or more additional generated predicted outputs, and modifying a parameter of the time-series forecasting model based on the determined error values.

Abstract: An embodiment includes analyzing text content of a user query to identify via natural language processing (NLP) a query topic. The embodiment maps the query topic to a topic cluster at a node of a hierarchical model of a text database. The embodiment generates query demand data indicative of demand for the topic cluster based on user queries. The embodiment identifies the topic cluster as a topic-cache candidate based on the query demand data. The embodiment compares an amount of memory required for storing text associated with the first topic cluster to available cache memory. The embodiment caches the text of the topic cluster candidate upon determining that there is sufficient available cache memory space.

Abstract: A system and method are disclosed that utilizes an artificial intelligence based virtual proxy node. The virtual proxy node includes an intent resolution model and communicates between a smart audio device and at least one secondary device, wherein the at least one secondary device is configured to be controlled by a smart audio device or smart hub. The virtual proxy node tracks interactions between the smart audio device and the at least one secondary device to derive historical and context data from the tracking interactions. The virtual proxy node uses the historical and context data to predict which secondary device will be successful in responding to the user input command and broadcasts the input command to the virtual proxy node associated with one of the at least one secondary device. The virtual proxy node includes an intent resolution model trained by historical and context data.

Abstract: Methods, systems, and computer program products for facilitating user selection using trend-based joint embeddings are provided herein. A method includes obtaining a selection of an item in an online catalog; determining a compatible item of the plurality of items at least in part by providing the selected at least one item and at least one previously selected item corresponding to the user to a trend-based machine learning model, wherein the trend-based machine learning model is trained on historical data associated with the item in the online catalog and fine-tuned based on current trend data from multiple data sources; receiving feedback in response to outputting the at least one compatible item; identifying one or more attributes related to the at least one compatible item based on the feedback; and using the trend-based machine learning model to determine at least one additional compatible item based on the one or more attributes.

Abstract: Using a trained neural network to transform user ratings into standardized user ratings is provided. Respective attribute-based leniency and strictness rating scores are generated for a plurality of attributes associated with a product category using the trained neural network based on historical user ratings of products in the product category. A set of attributes associated with a product included in the product category is identified. An overall leniency and strictness rating score is determined for the product using the trained neural network based on a set of attribute-based leniency and strictness rating scores for the set of attributes associated with the product included in the product category. A user rating of the product is received. The user rating of the product is adjusted based on the overall leniency and strictness rating score for the product included in the product category to form a standardized user rating for the product.