Abstract: A processing system includes an accelerator unit (AU) configured to perform matrix multiplication operations and subsequent tensor operations for a machine-learning model implemented by the processing system. When performing the matrix multiplication operations, the AU uses matrices having values in a first data format to generate result matrices also having values in the first data format. Further, to perform a subsequent tensor operation, the AU scales and rounds one or more of the values of the result matrix in the first data format such that the values are converted to a second, lower-precision format. The AU then performs the tensor operation using the values in the second, lower-precision format.

Abstract: Methods, apparatus, and processor-readable storage media for generating 3D images using transformer-based and diffusion-based artificial intelligence techniques are provided herein. An example computer-implemented method includes obtaining at least one 2D image; generating one or more point cloud tokens and one or more triplane tokens based at least in part on at least portions of the at least one 2D image; determining multiple 3D features associated with the at least portions of the at least one 2D image by processing at least a portion of the one or more point cloud tokens and at least a portion of the one or more triplane tokens using one or more transformer-based artificial intelligence techniques; and generating a 3D image associated with the at least portions of the at least one 2D image by processing at least a portion of the multiple 3D features using one or more diffusion-based artificial intelligence techniques.

Abstract: A multi-element filtration vessel contains multiple filtration elements. The elements have cylindrical housings and an annular header. At least a portion of the annular header surrounds the cylindrical housings at one end. The annular header has a periodically varying cross-sectional width and thickness such that the annular header has alternating thinner and thicker regions. The elements are packed with thinner sections of the header aligned. This allows for close packing of the elements without unduly compromising the mechanical strength of the headers.

Abstract: An apparatus to facilitate a block normalized adder (BNA) with four-deep systolic parameter support is disclosed. The apparatus includes matrix acceleration hardware comprising: data processing units, wherein the respective data processing units comprise normalization adding circuitry including: multiplier circuits; shifters to receive multiplier outputs from the multiplier circuits and apply block normalization; and an accumulator circuit to accumulate shifter outputs from the shifters with an intermediate value, wherein the accumulator circuit to generate an accumulator output in a high precision and unnormalized format to pass to a next stage of the matrix acceleration hardware; a bypass data structure to cause a global source operand to bypass the data processing units; and an adder circuit to add the accumulator output from a final data processing unit of the data processing units with the global source operand and apply round to nearest even to generate a final output.

Abstract: A drying apparatus is based on supercritical fluid. The drying apparatus includes: an upper cover; a base, arranged below the upper cover and the base and the upper cover; a substrate tray, arranged on the base; a first fluid supply tube, arranged on the top wall of the upper cover; a fluid disturbance plate, arranged below the first fluid supply tube; a second fluid supply tube, arranged on a first side wall of the upper cover; and a fluid discharge tube, arranged on a second side wall of the upper cover. The inner space of the closed chamber can be minimized by using the drying apparatus, thereby saving the usage amount of the supercritical fluid, and reducing the usage costs.

Abstract: Methods, apparatus, and computer-readable media for executing operations on single-instruction-multiple-data (SIMD) units. A lane of a wave executing on a SIMD unit executes a dot-product instruction during a clock cycle of the lane using operands stored in vector general purpose registers (VGPRs) of the lane. The dot-product instruction generates a convolution result for a data element of input data and applies one or more transitional operations to the convolution result to generate an output data element. The transitional operations include any one or any combination of a reformatting operation, a rectifying operation, and a clamping operation. The dot-product instruction may be a dual dot-product instruction that generates convolution results for two data elements and applies the transitional operations to produce two output data elements during the clock cycle of the lane.

Abstract: A method for coordinated control of multiple UVs based on an environment-aligned LLM is provided, including: obtaining, by each of the UVs, a local observation; transmitting, for each of the UVs, the local observation to a prompt system, and obtaining, based on the local observation, a policy prompt and a communication prompt for each of the UVs; executing a centralized training phase, including: obtaining a global state; generating, for each of the UVs, an executable action sequence and an action execution probability of each action in the executable action sequence; obtaining, for each of the UVs, a next action, executing the next action, and obtaining a reward value; and inputting the global state and the reward value into a critic network, and optimizing parameters of a policy network; executing a distributed execution phase, including: for each of the UVs, obtaining an action probability distribution function; and obtaining the next action.

Abstract: A computer system and associated processes are disclosed for grouping similar real estate properties into contiguous neighborhoods, and for generating neighborhood-specific models capable of estimating property values within their respective neighborhoods. A clustering component uses various sources of property-level data to group properties based on measures of property similarity. For example, the clustering component may measure property similarity based on how frequently specific properties are designated as comparable in appraisal reports. A model generator uses a machine learning process to determine, for specific neighborhoods, correlations between property attributes and values, and uses these correlations to generate the neighborhood specific models.

Abstract: A method for constructing a structural semantic map under an underground weak-light and low-texture environment is provided. The method includes: fusing traditional methods with a parameter line detection and verification model of structural semantics of a Transformer; and establishing a geometric primitive half-plane search method guided by the direction of structural information. The method also includes establishing a neighborhood greedy expansion algorithm based on a geometric primitive model; and optimizing geometric primitive poses and boundaries one by one. The method further includes fusing a point cloud map with structural information, and establishing a semantic map with geometric structure primitives.

Abstract: Disclosed are preparation for an ethinylestradiol-cyclodextrin complex and use thereof. The ethinylestradiol-cyclodextrin complex utilizes the properties of cyclodextrin that the inner cavity is hydrophobic and the outer surface is hydrophilic to increase the water solubility of the ethinylestradiol-cyclodextrin complex that is also an ethinylestradiol drug, so that the solubility of ethinylestradiol is greatly increased and the bioavailability of water-insoluble ethinylestradiol is improved, thereby providing the possibility for changing the administration mode and homogeneous injection of ethinylestradiol.

Abstract: A method in an illustrative embodiment includes blocking cells with predetermined attributes in a first table by performing semantic segmentation on the first table, to obtain a blocked first table. The method further includes converting the blocked first table to a first text. The method further includes generating a second text in a form of natural language based on an understanding of the first text. The method further includes generating a second table corresponding to the first table based on the second text. In this way, the cells containing sensitive information can be identified and blocked, thereby retaining other useful information of the table while protecting privacy. Moreover, using the text in a form of natural language as an intermediate processing result can increase the transparency of table generation, lower the risk of data distortion, and improve the accuracy and consistency of synthetic data while protecting data privacy.

Abstract: A display substrate includes a substrate and a display region, an isolation region, a peripheral region and a hole on the substrate. The display region surrounds the hole, the isolation region is between the display region and the peripheral region, and the peripheral region surrounds and adjoins the hole. The isolation region has an isolation column and a first inorganic structure. The isolation column is at a side of the substrate, and the first inorganic structure is at a side of the isolation column away from the substrate. The peripheral region has a structure layer on the substrate and including a first organic structure and a second inorganic structure along a direction perpendicular to the substrate. The first inorganic structure, the second inorganic structure and the first organic structure all are made of an insulation material. The first inorganic structure and the second inorganic structure encapsulate the display substrate.

Abstract: The present disclosure generally relates to systems and methods for generating subdivision boundaries based on extracted subdivision information from parcel data. In some embodiments, the subdivision boundary system can classify, extract, and standardize information from parcel data to generate “cleaned” parcel data. This cleaned parcel data can be used to generate subdivision boundaries that merge parcels that are likely to belong to the same subdivision within the same boundary. In some embodiments, the subdivision boundary system can complete and correct subdivision boundaries by refilling missing subdivision names. In some embodiments, a boundary system can access the subdivision information (e.g., subdivision names) stored in the subdivision data store and other information (e.g., map information, geographical information) to generate a boundary corresponding to a subdivision.

Abstract: A method and a system for coordinated control of a UAV swarm based on communication information completion are provided. The method includes: obtaining an information weight based on a local observation and information obtained from other UAVs through an information-level weighting network; obtaining weighted information of each UAV, and generating a global state through an adaptive generation network; obtaining a global action-value function by integrating a local action-value function of each UAV through a mixing network based on the global state and the local action-value function of the UAV; in a distributed execution phase, obtaining, by each UAV, the local action-value function of the UAV based on the local observation and the information weight through a Transformer-based decoder, and performing coordinated decision-making through a policy network.

Abstract: Provided is a collaborative exploration system for an unknown space based on a heterogeneous air-ground robot, comprising: an unmanned ground vehicle (UGV) and a plurality of unmanned aerial vehicles (UAVs) loaded on the UGV; wherein the UGV is equipped with a first perception module and a first computing platform configured for constructing a fine-grained map of an environment; each of the plurality of UAVs is equipped with a second perception module and a second computing platform configure for constructing a lightweight map of the environment; and a wireless communication module is disposed in the UGV and each of the plurality of UAVs, and the UGV and the plurality of UAVs communicate via a semi-centralized self-organizing communication mode; when the system performs cooperative exploration for the unknown space, the plurality of UAVs explore unknown regions around the UGV with the UGV as a center; based on the second perception module and the second computing platform of each of the plurality of UAVs, the

Abstract: The present invention discloses a multi-UAV coordinated control method based on a dynamic directed graph communication structure. In a distributed execution phase, based on local observation values and historical hidden states, a communication structure between UAVs is dynamically adjusted through a multi-key gated communication network using an adjacency trajectory matrix. Local action value functions are calculated and the hidden states are updated based on an updated adjacency trajectory matrix and pre-communication information, and a final control decision is generated and executed. In a centralized training phase, the adjacency trajectory matrix and the local observation value at each moment are recorded to generate a global state, and a global action value function is generated using a mixing network, and network parameters are then updated through rewards. A trained graph collapse network and the mixing network are deployed to each UAV to achieve distributed execution.

Abstract: Provided in the present disclosure are a method and apparatus for determining a position of a client, and a processor. The method includes: a first acquisition step: acquiring at least one piece of first Channel State Information (CSI) from a first communication link for an Access Point (AP) to communicate with the client; a second acquisition step: when a receiving antenna of the AP meets a preset condition, acquiring at least one piece of second CSI from a second communication link for the AP to communicate with the client; a processing step: processing the first CSI to obtain a first phase difference, and processing the second CSI to obtain a second phase difference; and a determination step: determining a position of the client according to the first phase difference and the second phase difference. By respectively acquiring two different groups of CSI, the limitation that a spacing of the AP antenna is limited to half a signal wavelength is lifted.

Abstract: Methods, apparatus, and processor-readable storage media for generating image-based avatars using multi-modal artificial intelligence techniques are provided herein. An example computer-implemented method includes encoding one or more image-related features by processing one or more portions of input image data using at least one multi-channel image encoder; encoding one or more audio-related features by processing one or more portions of input audio data using at least one audio encoder; encoding one or more text-related features by processing one or more portions of input text data using at least one text encoder; and generating at least one image-based avatar by processing, using at least one vision encoder, at least a portion of the one or more image-related features and one or more of at least a portion of the one or more audio-related features and at least a portion of the one or more text-related features.

Abstract: Implementations of the present disclosure relate to a method, an electronic device, and a computer program product for managing an inference process. Here, the inference process is implemented based on a machine learning model. A method includes: determining, based on a computational graph defining the machine learning model, dependency relationships between a set of functions for implementing the inference process; acquiring, in at least one edge device located in an edge computing network, a set of computing units available to execute the inference process; selecting at least one computing unit for executing the set of functions from the set of computing units; and causing the at least one computing unit to execute the set of functions based on the dependency relationships. With example implementations of the present disclosure, the inference process is implemented by making use of a variety of computing units in the edge computing network, thereby improving performance.