Abstract: A document processing system processes a document image to identify document image regions including floating images, structured data units, and unstructured floating text. A first masked image is generated by deleting any floating images from the document image and a second masked image is generated by deleting any structured data units from the first masked image. The structured data units and the unstructured floating text are thus identified serially one after another. Textual data is extracted from the structured data units and the unstructured floating text by processing the corresponding document image regions via optical character recognition (OCR). Entities are extracted from the textual data using natural language processing (NLP) techniques.

Abstract: The system obtains a hierarchy representing a construction site, where the hierarchy includes a parent node and a child node, and obtains a first computer model of a first construction site including a first multiplicity of virtual objects. Based on the hierarchy representing the construction site, and the first computer model of the first construction site, the system creates a first map between the hierarchy representing the construction site and the first computer model of the first construction site. The system trains an artificial intelligence based on the first map, and then provides to the artificial intelligence a second computer model of a second construction site. The system receives from the artificial intelligence a second map between the hierarchy representing the construction site and the second computer model.

Abstract: The system obtains a hierarchy representing a construction site, and multiple queries associated with a node in the hierarchy, where a query among the multiple queries requests an indication of a stage of construction associated with an object at the construction site represented by the node, and where the query among the multiple queries requests an answer from a binary set. The system obtains a recording of the object at the construction site and presents the recording and the multiple queries to a user. The system receives the answer from the binary set from the user, and creates an artificial intelligence training dataset based on the answer associated with the binary set and the recording of the object.

Abstract: The system obtains a first recording of a construction site captured at a first time, and a first virtual trajectory of a first virtual camera, and creates a first data structure representing the first recording. The system obtains a second recording of the construction site captured at a second time, and creates a second data structure representing the second recording. The system establishes a correspondence between the first and second data structures by creating a map between a second image in the second recording and a first image in the first recording by efficiently searching the first data structure for the first image similar to the second image. The system obtains a first location of the first virtual camera corresponding to the first image and, based on the map between the first and second image, determines a second location of a second virtual camera corresponding to the second image.

Abstract: The system obtains a floor plan associated with a construction site, and a three-dimensional computer model of the construction site, a trajectory through the construction site, and a recording of the construction site. The system establishes a correspondence between the trajectory and the floor plan. Based on the correspondence between the trajectory and the floor plan, the system creates a virtual camera associated with the three-dimensional computer model and a virtual trajectory, where the virtual trajectory influences virtual objects visible to the virtual camera, and the virtual objects visible to the virtual camera match objects visible to a camera making the recording. The system determines a transformation of the camera between an initial image and a subsequent image in the recording, and applies the transformation to the virtual camera to obtain a virtual trajectory of the virtual camera associated with the three-dimensional computer model.

Abstract: Technologies for fiber array unit (FAU) lid designs are disclosed. In one embodiment, channels in the lid allow for suction to be applied to fibers that the lid covers, pulling the fibers into place in a V-groove. The suction can hold the fibers in place as the fiber array unit is mated with a photonic integrated circuit (PIC) die. Additionally or alternatively, channels can be on pitch, allowing for pulling the FAU towards a PIC die as well as sensing the position and alignment of the FAU to the PIC die. In another embodiment, a warpage amount of a PIC die is characterized, and a FAU lid with a similar warpage is fabricated, allowing for the FAU to position fibers correctly relative to waveguides in the PIC die. In another embodiment, a FAU has an extended lid, which can provide fiber protection as well as position and parallelism tolerance control.

Abstract: Techniques are described herein for detecting an invalid (e.g., spoof) email before it is received by an intended recipient. In some embodiments, the techniques may involve, upon receiving an electronic communication directed to an intended recipient, determining, based on information included in the electronic communication, a claimed source entity, and determining a domain associated with the email communication. The techniques may further involve determining an owner entity associated with the domain and then determining that the electronic communication is valid based on a comparison between the owner entity and the claimed source entity. Upon determining that the electronic communication is not valid, the techniques may further comprise performing one or more mitigation techniques.

Abstract: A method and system are disclosed for collecting and maintaining road event information, where point-based road event location data from transportation authorities, construction companies, vehicle sensors, or combinations thereof, is susceptible to location sensor errors, digital map errors, and/or map mismatching errors. Errors less than a selected threshold are filtered by categorizing or grouping reports of point-based locations into segments along links in a representation of a road, providing improved accuracy of reporting of road events.

Abstract: An improved apparatus 100 for cutting and squeezing sachets/pouches is disclosed, having a cutting and sliding assembly 200 configured to allow linear movement of one or more sachets/pouches and cut the one or more sachets/pouches as the one or more sachets/pouches undergo linear movement: and a roller assembly 300 configured to pull the one or more sachets/pouches through the cutting and sliding assembly. and squeeze the one or more sachets/pouches as the one or more sachets are pulled. the squeezing resulting in extracting contents of the one or more sachets/pouches. The cutting and sliding assembly 200 and the roller assembly 300 are accommodated within a container assembly 400 such that the squeezed out contents are collected within a container of the container assembly 400.

Abstract: This disclosure describes techniques and mechanisms to retroactively identifying, classifying, categorizing, and/or remediating campaigns by an email threat defense system. The described techniques may perform a time-series analysis on record data associated with emails and identify campaigns that have bypassed threat detection mechanisms. The described techniques may extract and correlate features of the record data in order to label and determine whether a campaign is malicious. Where the email campaign is malicious, remedial action(s) can occur. Accordingly, the described techniques may remediate false negatives in a network and improve network security.

Abstract: Techniques for an email-security system to screen emails, extract information from the emails, analyze the information, assign probability scores to the emails, and classify the emails as likely fraudulent or not. The system may analyze emails for users and identify fraudulent emails by analyzing the contents of the emails. The system may evaluate the contents of the emails to determine probability score(s) which may further determine an overall probability score. The system may then classify the email as fraudulent, or not, and may perform actions including blocking the email, allowing the email, flagging the email, etc. In some instances, the screened emails may include legitimate brand domain addresses, names, images, URL(s), and the like. However, the screened emails may contain a reply-to domain address that matches a free email service provider domain. In such instances, the email-security system may assign a probability score indicative that the screened email is fraudulent.

Abstract: Techniques for an email-security system to detect multi-stage email scam attacks, and engage an attacker to obtain additional information. The system may analyze emails for users and identify scam emails by analyzing metadata of the emails. The system may then classify the scam emails into particular classes from among a group of scam-email classes. The system may then engage the attacker that sent the scam email. In some instances, the scam emails may be multi-stage attacks, and the system may automatically engage the attacker to move to the next stage of the scam attack. For instance, the system may send a lure email that is responsive to the particular scam class to prompt or provoke the attacker to send more sensitive information, such as a phone number, a bank account, etc. The system may then harvest this sensitive information of the attacker, and use that information for various remedial actions.

Abstract: The method can include optionally training a transportation modality classification model; determining a transportation modality of a trip; and optionally triggering an action based on the transportation modality. However, the method can additionally or alternatively include any other suitable elements. The method functions to facilitate a classification of a transportation modality for trips based on location data (e.g., collected at a mobile device). Additionally or alternatively, the method can function to facilitate content provisions based on a trip classification.

Abstract: Provisioning of day-1 configuration templates is automated. Users are provided access to a Configuration Template User Interface (UI) for creating configuration templates. Input is received on the Configuration Template UI to create the configuration templates. A trigger is received by a configuration manager to initiate configuration of network devices or functions. The trigger includes one or more Template Identifiers (IDs) corresponding to the one or more configuration templates. Configuration files are generated based on the trigger and one or more Template IDs corresponding to the one or more configuration templates. Configuration files are sent from the configuration manager to the one or more network devices or functions to configure the one or more network devices or functions.

Abstract: Techniques are disclosed for framing images and/or video streams. Such techniques may include performing face detection on an image, determining a gaze direction within image content associated with a detected face, and defining a cropping window for the image based on the detected face and the determined gaze direction of the detected face. Thereafter, the image may be cropped according to the cropping window.

Abstract: A system and method of open-world semi-supervised satellite object detection involves a machine learning engine configured with a training component and an inference component. The method of object detection detects objects in satellite imagery represents a solution for difficult challenges of arbitrary orientations, wide variation in object sizes, large number of densely packed objects, and highly complex background. A transformer network detects unknown objects in the satellite image. The transformer network includes a rotation-aware pyramidal pseudo-labeling operation that captures scale-specific pyramidal features at oriented box regions for pseudo-labeling unknown objects in the satellite image. A semi-supervised learning pipeline learns a new set of object classes to be detected. A prediction head outputs the satellite image annotated with a predicted object class for an unknown object.

Abstract: Techniques are described for protecting services containing web applications from malicious attacks. A data flow that includes a stack trace and a process is received from a web application. The stack trace is chained with the process to generate a data flow graph. A data lake is then accessed to determine whether the data flow graph is in the data lake. If the data flow graph is not in the data lake, then the data flow can be considered safe. If the data flow graph is not already in the data lake, then a determination is made as to whether the data flow includes a command that accepts user input. If it is determined that the data flow does include a command that accepts user input, then the data flow can be considered to be malicious and the process is terminated to protect the web application from a malicious attack.

Abstract: An example operation may include one or more of detecting, by a blockchain (BC) node, a contract update on a blockchain, determining, by the BC node, at least one charge affected by the contract update, executing, by the BC node, a smart contract to re-generate an invoice based on the at least one charge, and recording the invoice on a ledger of the blockchain.

Abstract: Provided are systems, methods, and programming for facilitating real-time determination of a process completion likelihood. In some embodiments, data including an update to a system, the update occurring at a first time, wherein updates to the system are permitted until an expiration time may be obtained, a set of fixed descriptors of the system may be retrieved and/or received, and a set of status updates describing the system at prior times may be obtained. Each status update of the set of one or more status updates includes at least (i) an update to the first system and (ii) a time that the respective status update occurred. Based on the data, the fixed descriptors, and the status updates, using a trained machine learning model, a failure/success score indicating a likelihood that, at the expiration time, the system satisfies a threshold condition may be computed and stored in memory.

Abstract: Techniques and architecture are described for detecting a compromised mailbox as an email account compromise (EAC) involved in lateral phishing, lateral scam, lateral BEC, outbound scam, lateral and inbound fraudulent money transfer requests. For example, the techniques and architecture provide a method that comprises scanning, by a pre-filter, electronic mail messages (emails) within an organization, wherein the emails originate within the organization. The pre-filter analyzes the emails with respect to known fraudulent email practices and determines that an email is a questionable email. A retrospective behavior engine analyzes the questionable email with respect to one or more historical traits to provide a feature set. Based at least in part on the feature set, the verdict correlation engine determines that the questionable email belongs in a class of emails from multiple classes of emails. Based at least in part on the class, the verdict correlation engine performs a responsive action.