Abstract: The present invention provides solutions for reducing biases in text-to-image diffusion models, particularly biases related to gender, race, and their intersections in occupational prompts. The invention introduces a fairness framework based on distributional alignment, comprising two core technical solutions: (1) a distributional alignment loss that adjusts the output of the model toward user-defined target distributions, and (2) an adjusted direct finetuning (adjusted DFT) of the model's sampling process using an adjusted gradient to optimize losses based on generated images. These techniques reduce bias while supporting diverse perspectives on fairness, such as age-controlled debiasing across multiple concepts. The method's scalability allows for debiasing multiple prompts simultaneously, improving the inclusivity of diffusion model outputs across varied demographics.

Abstract: The present invention provides solutions for identifying influential training images in diffusion models by calculating importance scores through gradient-based attribution. The system uses a novel Diffusion-Tracing with the Randomly projected After Kernel (D-TRAK) method for identifying and scoring the influence of individual training data points on the outputs of diffusion models, thereby enabling the accurate and interpretable attribution of data in generative models. This approach allows for the identification of training images that have a significant positive or negative influence on a specific generated output image. By focusing on the final checkpoint data, the system reduces computational costs while providing accurate attribution of image generation results. This invention has applications in copyright protection, and model transparency, particularly in identifying the contribution of specific training data to generated outputs.

Abstract: The present application is directed to stranded resource recovery in a cloud computing environment. A resource utilization signal at each of a plurality of nodes that each hosts corresponding virtual machines (VMs) is measured. Based on each resource utilization signal, a set of candidate nodes is identified. Each candidate node comprises a stranded resource that is unutilized due to utilization of a bottleneck resource. The identification includes calculating an amount of the stranded resource at each candidate node. From a plurality of VMs hosted at the set of candidate nodes, a set of candidate VMs is identified for migration for stranded resource recovery. The identification includes calculating a score for each candidate VM based on a degree of imbalance between the stranded resource and the bottleneck resource at a candidate node hosting the candidate VM. Migration of at least one candidate VM in the set of candidate VMs is initiated.

Abstract: Provided herein are systems, methods, and computer-readable media for implementing differentially private principal component analysis (PCA) for vertically partitioned data. An example system can include a first client possessing a first column vector, and a second client possessing a second column vector. The first client can be configured to discretize the first column vector to obtain a first discretized column vector and the second client can be configured to discretize the second column vector to obtain a second discretized column vector. The first client can be configured to introduce a first noise to the first discretized column vector and the second client can be configured to introduce a second noise to the second discretized column vector to obtain a PCA result.

Abstract: An optical integrated device and a preparation method therefor.

Abstract: Disclosed are a downlink signal processing method and apparatus, a base station, and a non-transitory computer-readable medium. The downlink signal processing method may includes performing resource element mapping on an unmodulated downlink signal, and performing modulation on the downlink signal that has been subjected to the resource element mapping to obtain modulated data.

Abstract: Provided is an information backhaul method, which is applied to a networking unit in a network system. The network system includes multiple stages of networking units that are cascaded, and each of the multiple stages of networking units includes multiple optical ports. The information backhaul method includes: acquiring cascade information of a networking unit in a current stage, where the cascade information of the networking unit in the current stage includes optical port information and identification information of the networking unit in the current stage; and feeding back the cascade information of the networking unit in the current stage. Further provided are a data allocation method, a networking unit, a data allocation controller, a network system and a computer-readable storage medium.

Abstract: System, methods, and non-transitory computer-readable medium are provided for watermarking a diffusion model. For example, a method for watermarking a diffusion model may include generating one or more training data elements. In some aspects, the one or more trainings data elements may include target images. Moreover, the target images may include pre-defined watermark information. Further, the method may include training the diffusion model to predict the target images using training data including the one or more training data elements.

Abstract: An optical coupling system and a preparation method therefor are provided. The preparation method for the optical coupling system includes: providing a substrate, mounting an optical chip on the substrate, and forming a first intermediate member; fixing an optical fiber in a fixing structure, forming an optical fiber assembly; providing a connection assembly that includes a chip-side connector and an optical fiber-side connector; fixedly connecting the optical fiber-side connector to the optical fiber assembly, connecting the other end of the optical fiber-side connector to the chip-side connector, and forming a second intermediate member; adjusting the position of the second intermediate member relative to the first intermediate member; and when the optical chip and the optical fiber achieve optimal optical coupling, stopping the adjustment of the relative position, and fixing the chip-side connector to the substrate.

Abstract: A method for training a machine learning model is described, comprising receiving, for each perturbation of a plurality of perturbations of model parameters of a starting version of the machine learning model, a change of loss of the machine learning model caused by the perturbation for a set of training data determined by feeding the set of training data to one or more perturbed versions of the machine learning model, estimating a gradient of the loss of the machine learning model with respect to the model parameters from the determined changes of loss and updating the starting version of the machine learning model to an updated version of the machine learning model by changing the model parameters in a direction for which the estimated gradient indicates a reduction of loss.

Abstract: A method and apparatus for forming tungsten features in semiconductor devices is provided. The method includes exposing a top opening of a feature formed in a substrate to a physical vapor deposition (PVD) process to deposit a tungsten liner layer within the feature. The PVD process is performed in a first processing region of a first processing chamber and the tungsten liner layer forms an overhang portion, which partially obstructs the top opening of the feature. The substrate is transferred from the first processing region of the first processing chamber to a second processing region of a second processing chamber without breaking vacuum. The overhang portion is exposed to nitrogen-containing radicals in the second processing region to inhibit subsequent growth of tungsten along the overhang portion. The feature is exposed to a tungsten-containing precursor gas to form a tungsten fill layer over the tungsten liner layer within the feature.

Abstract: A method of coating a piston ring comprises: providing at least first and second piston rings in a coating chamber; spacing the piston rings apart, so that the first piston ring is spaced from and not in contact with the second piston ring, and applying a ta-C coating to the piston rings, whereby the spacing between the adjacent piston rings enables simultaneous coating of upper, outer and lower piston ring surfaces of the first and second piston rings during rotation of rings in a plane co-planar with the coating beam. Piston rings are obtained comprising a substantially hydrogen free ta-C coating of 0.1 to 8 microns on at least their lower surface.

Abstract: Embodiments of the present disclosure disclose an instruction scheduling method, an electronic device and a storage medium. The method comprises: determining at least one target memory access instruction in an instruction set corresponding to a micro-architecture model; determining durations consumed by each target memory access instruction in a plurality of instruction running scenarios; and, performing instruction scheduling on each target memory access instruction on the basis of the durations consumed by each target memory access instruction in the instruction running scenarios. By using the embodiments of the present disclosure, memory access instructions can be scheduled on the basis of the durations consumed by the memory access instructions in different instruction running scenarios, so that the applicability is high.

Abstract: A memory system includes a memory cell array and a controller coupled to the memory cell array. The controller is configured to control applying a first program voltage to a word line to program memory cells in the memory cell array, the memory cells being coupled to the word line, and in response to receiving a suspend command, control applying a positive bias discharge voltage to the word line when the first program voltage ramps down.

Abstract: Methods and apparatus for passivating a target are provided herein. For example, a method includes a) supplying an oxidizing gas into an inner volume of the process chamber; b) igniting the oxidizing gas to form a plasma and oxidize at least one of a target or target material deposited on a process kit disposed in the inner volume of the process chamber; and c) performing a cycle purge comprising: c1) providing air into the process chamber to react with the at least one of the target or target material deposited on the process kit; c2) maintaining a predetermined pressure for a predetermined time within the process chamber to generate a toxic by-product caused by the air reacting with the at least one of the target or target material deposited on the process kit; and c3) exhausting the process chamber to remove the toxic by-product.

Abstract: Provided are a method and system for controlling a voltage. The method includes: acquiring a current output voltage of an inverter; calculating a current voltage error of the current output voltage relative to a given output voltage; inputting the current voltage error into a fuzzy controller to determine a target output voltage of the inverter; determining an amplitude adjustment command based on a difference between an amplitude of the target output voltage and an amplitude of the current output voltage; determining a phase adjustment command based on a difference between a phase of the target output voltage and a phase of the current output voltage; and adjusting an amplitude and a phase of an output voltage of the inverter according to the amplitude adjustment command and the phase adjustment command respectively, to maintain the output voltage of the inverter within a preset range.

Abstract: Disclosed are a downlink signal processing method and apparatus, a base station, and a non-transitory computer-readable medium. The downlink signal processing method may includes performing resource element mapping on an unmodulated downlink signal, and performing modulation on the downlink signal that has been subjected to the resource element mapping to obtain modulated data.

Abstract: Methods for depositing a dielectric oxide layer atop one or more substrates disposed in or processed through a PVD chamber are provided herein. In some embodiments, such a method includes: sputtering source material from a target assembly onto a first substrate while the source material is at a first erosion state and while providing a first amount of RF power to the target assembly to deposit a dielectric oxide layer onto a first substrate having a desired resistance-area; and subsequently sputtering source material from the target assembly onto a second substrate while the source material is at a second erosion state and while providing a second amount of RF power to the target assembly, wherein the second amount of RF power is lower than the first amount of RF power by a predetermined amount calculated to maintain the desired resistance-area.