Abstract: Deep learning models, such as diffusion models, can synthesize images from noise. Diffusion models implement a complex denoising process involving many denoising operations. It can be a challenge to understand the mechanics of diffusion models. To better understand how and when structure is formed, saliency maps and concept formation intensity can be extracted from the sampling network of a diffusion model. Using the input map and the output map of a given denoising operation in a sampling network, a noise gradient map representative of the predicted noise of a given denoising operation can be determined. The noise gradient maps from the denoising operations at different indices can be combined to generate a saliency map. A concept formation intensity value can be determined from a noise gradient map. Concept formation intensity values from the denoising operations at different indices can be plotted.

Abstract: A wiring harness for a battery and inverter system includes a main cable. It further includes an inverter connector configured to plug the main cable into an inverter module. It further includes a first spur of the main cable terminating in a first-type battery device connector. It further includes a second spur of the main cable terminating in a second type battery device connector.

Abstract: Machine learning models can process a video and generate outputs such as action segmentation assigning portions of the video to a particular action, or action classification assigning an action class for each frame of the video. Some machine learning models can accurately make predictions for short videos but may not be particularly suited for performing action segmentation for long duration, structured videos. An effective machine learning model may include a hybrid architecture involving a temporal convolutional network and a bi-directional graph neural network. The machine learning model can process long duration structured videos by using a temporal convolutional network as a first pass action segmentation model to generate rich, frame-wise features. The frame-wise features can be converted into a graph having forward edges and backward edges. A graph neural network can process the graph to refine a final fine-grain per-frame action prediction.

Abstract: Disclosed herein are systems, apparatuses, methods, and non-transitory computer readable media related to a display construct coupled to a structure (e.g., a vision window). The structure can be a supportive structure such as a fixture. The display construct is configured to facilitate media display and is at least partially transparent. The vision window may be a tintable window, e.g., a window in which its tint is electrically controllable (e.g., an electrochromic window). Various interactive capabilities with the display construct are disclosed (e.g., via a touch screen).

Abstract: Disclosed herein are systems, apparatuses, methods, and non-transitory computer readable media related to display constructs that can be fixed or movable. The display constructs can be configured to facilitate media display at opposite sides. The display constructs may be is at least partially transparent and can be viewed therethrough. Various interactive capabilities with the display construct are disclosed, as well as various installation component and kits. Further disclosed are wireless chargers embedded in fixtures and/or real assets of a facility.

Abstract: An air handling unit (AHU) for a heating, ventilation, air conditioning, and refrigeration (HVACR) system includes a housing and a combustion heater disposed within the housing. The housing includes a combustion section with a first channel and a second channel. The combustion heater includes heat exchanger tubes and a tube support that supports heat exchanger tubes within the combustion section. The tube support slidably disposed in the first channel and the second channel. The combustion heater configured to be slidably removable from the AHU. An AHU for an HVACR system includes a housing with a fan section and a fan assembly disposed within the housing. The fan assembly including a pair of grooves slidably disposed on a pair of rails of the housing. The fan assembly configured to be both slidably removable from the AHU and liftably removable from the AHU.

Abstract: Provided are therapeutic compositions containing microbial populations for prevention, treatment and reduction of symptoms associated with a dysbiosis of a mammalian subject such as a human.

Abstract: A structured casing built within an exterior door in an exterior doorway. The casing comprises of an access tray with a handle and an inner tray flap on the inside of the enclosed casing that are connected by an assembled bracket system, an access door with a handle and a locking device on the interior back, opposing interior and exterior side panels, an interior and exterior casing top, and a casing bottom. Opening the access tray sets the assembled bracket system in motion causing the inner tray flap to position at a 90-degree angle, forming a flat package receiving space. Closing the access tray further sets the assembled bracket system in motion causing the inner tray flap to return to its starting position releasing the package to the bottom where the package can then be retrieved by unlocking and opening the access door from inside a home or building.

Abstract: A liquid crystal device that is flexible and includes a cholesteric liquid crystal material erased by flexing. Totally free of electronics and electrically conductive electrodes, the liquid crystal device takes advantage of a reverse mode effect observed in cholesteric liquid crystals whereby pressure applied to the device changes a reflective texture to a transmissive texture to create an image. The image is erased and the device reinitialized by flexing to create reflective texture surroundings.

Abstract: Disclosed is a technical solution to process a video that captures actions to be performed for completing a task based on a chronological sequence of stages within the task. An example system may identify an action sequence from an instruction for the task. The system inputs the action sequence into a trained model (e.g., a recurrent neural network), which outputs the chronological sequence of stages. The RNN may be trained through self-supervised learning. The system may input the video and the chronological sequence of stages into another trained model, e.g., a temporal convolutional network. The other trained model may include hidden layers arranged before an attention layer. The hidden layers may extract features from the video and feed the features into the attention layer. The attention layer may determine attention weights of the features based on the chronological sequence of stages.

Abstract: A method of generating a path of an object through a virtual environment is provided, the method comprising: receiving image data, at a first instance of time, from a plurality of image capture devices arranged in a physical environment; receiving image data, at an at least one second instance of time after the first instance of time, from a plurality of image capture devices arranged in the physical environment; detecting a location of a plurality of points associated with an object within the image data from each image capture device at the first instance of time and the at least one second instance of time; projecting the location of the plurality of points associated with the object within the image data from each image capture device at the first instance of time and the at least one second instance of time into a virtual environment to generate a location of the plurality of points associated with the object in the virtual environment at each instance of time; and generating a path of the object through

Abstract: A writing device includes a liquid crystal layer including reflective cholesteric liquid crystal material. There are electrically conductive layers between which the liquid crystal layer is disposed. At least one of the electrically conductive layers includes at least two conductive sections. There is a front substrate that is flexible and a back substrate between which the electrically conductive layers and the liquid crystal layer are disposed. Neither the front substrate nor the back substrate includes separated sections. Application of pressure to the front substrate changes a reflectivity of a portion of the cholesteric liquid crystal material to form an image. In one aspect, there is electronic erase circuitry including at least one actuator that independently applies an erase voltage waveform to each of at least two of the conductive sections.

Abstract: Disclosed is a technical solution to calibrate confidence scores of classification networks. A classification network has been trained to receive an input and output a label of the input that indicates a class of the input. The classification network also outputs a confidence score of the label, which indicates a likelihood of the input falling into the class, i.e., a confidence level of the classification network that the label is correct. To calibrate the confidence of the classification network, a logit transformation function may be added into the classification network. The logic transformation function may be an entropy-based function and have learnable parameters, which may be trained by inputting calibration samples into the classification network and optimizing a negative log likelihood based on the labels generated by the classification network and ground-truth labels of the calibration samples. The trained logic transformation function can be used to compute reliable confidence scores.

Abstract: A liquid crystal device that is flexible and includes a cholesteric liquid crystal material erased by flexing. Totally free of electronics and electrically conductive electrodes, the liquid crystal device takes advantage of a reverse mode effect observed in cholesteric liquid crystals whereby pressure applied to the device changes a reflective texture to a transmissive texture to create an image. The image is erased and the device reinitialized by flexing to create reflective texture surroundings.

Abstract: A computer system includes an account data store, a physical account ledger, a virtual account ledger, and an interest distribution subsystem. The account data store includes the mappings between physical accounts and corresponding hierarchies of virtual accounts and interest distribution parameters indicating how interest is distributed between physical accounts and corresponding virtual accounts. The physical account ledger stores balances for the physical accounts and the virtual account ledger stores balances for the virtual accounts. The interest distribution subsystem is configured to receive interest accrual information from the ledgers and determine how to distribute the accrued interest according to the interest distribution parameters.

Abstract: A cholesteric liquid crystal device has an optical response opposite to that of the Prior Art. The liquid crystal device takes advantage of a unique and unusual effect in cholesteric liquid crystals where pressure applied to the device with a pointed stylus, without applying a voltage, creates a transmissive texture as opposed to the planar texture of the Prior Art.

Abstract: An apparatus is provided for deep learning. The apparatus accesses a neural network including an input layer, hidden layers, and an output layer. The apparatus adds an activation function to one or more of the hidden layers of the hidden layers and output layer. The activation function includes a tunable parameter, the value of which can be adjusted during the training of the neural network. The apparatus trains the neural network by inputting training samples into the neural network and determining internal parameters of the neural network based on the training samples. Determining the internal parameters includes determining a value of the tunable parameter based on the training samples. The apparatus may determine two different values of the tunable parameter for two different layers. The activation function may include another tunable parameter. The apparatus can determine a value for the other tunable parameter during the training of the neural network.

Abstract: A method of generating a path of an object through a virtual environment is provided, the method comprising: receiving image data, at a first instance of time, from a plurality of image capture devices arranged in a physical environment; receiving image data, at an at least one second instance of time after the first instance of time, from a plurality of image capture devices arranged in the physical environment; detecting a location of a plurality of points associated with an object within the image data from each image capture device at the first instance of time and the at least one second instance of time; projecting the location of the plurality of points associated with the object within the image data from each image capture device at the first instance of time and the at least one second instance of time into a virtual environment to generate a location of the plurality of points associated with the object in the virtual environment at each instance of time; and generating a path of the object through

Abstract: A magnet erased writing device or eWriter includes cholesteric liquid crystal material, in which a written image is erased using magnetically activated electronic circuitry. By eliminating the mechanical erase button of Prior Art eWriters, the eWriter can be designed with a display footprint that can cover nearly the entire face of the eWriter. An erasing magnet can be used with the device, which can be handheld as, for example, by being part of a stylus. With the erasing magnet separated from the device and the mechanical erase button being omitted, the eWriter avoids unintentional erasures, a drawback of Prior Art eWriters.

Abstract: An apparatus is provided to compress CNN models using a combination of filter pruning and tensor decomposition. For example, the apparatus accesses a trained CNN that includes convolutional tensors. The apparatus prunes the filters of a convolutional tensor to generate a sparse tensor. Further, the apparatus decomposes the sparse tensor to generate a low-rank approximation of the sparce tensor. The low-rank approximation of the sparse tensor includes a core tensor and principal tensors. The apparatus generates a convolutional flow that includes the core tensor and convolutional operations generated based on the principal tensors. The apparatus may replace some or all the convolutional tensors in the trained CNN with the corresponding convolutional flows. The apparatus may fine-tune the updated CNN by re-training the updated CNN. The number of epochs for re-training the updated CNN may be smaller than the number of epochs for training the CNN.