Abstract: Disclosed herein are image-based object recognition method and system by and in which a learning server performs image-based object recognition based on the learning of environment variable data. The image-based object recognition method includes: receiving an image acquired through at least one camera, and segmenting the image on a per-frame basis; primarily learning labeling results for one or more objects in the image segmented on a per-frame basis; performing primary reasoning by performing object detection in the image through a model obtained as a result of the primary learning; performing data labeling based on the results of the primary reasoning, and performing secondary learning with weights allocated to respective boxes obtained by the primary reasoning and estimated as object regions; and finally detecting one or more objects in the image through a model generated as a result of the secondary learning.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a machine learning model using teacher annealing.

Abstract: Methods and systems for identity governance that provide for the identification of common or unique access items (e.g., identity management artifacts that may grant access). Certain embodiments may leverage representative data structures that represent an enterprise's identity management data to determine common or unique identity management access items represented in those data structures. In other embodiments, a machine learning model may be trained based on identity management data and utilize predictive scores to determine common or unique identity management access items.

Abstract: Systems and methods are provided that train a machine-learned language encoding model through the use of a contrastive learning task. In particular, the present disclosure describes a contrastive learning task where the encoder learns to distinguish input tokens from plausible alternatives. In some implementations, on each training example the proposed method masks out some subset (e.g., 15%) of the original input tokens, replaces the masked tokens with samples from a “generator” (e.g., which may be a small masked language model), and then trains the encoder to predict whether each token comes from the original data or is a replacement produced by the generator.

Abstract: A method for determining a placement for machine learning model operations across multiple hardware devices is described.

Abstract: Methods, apparatus and systems for wirelessly monitoring a micro motion of an object are described. In one example, a described system comprises: a transmitter configured for transmitting a first wireless signal through a wireless multipath channel of a venue; a receiver configured for receiving a second wireless signal through the wireless multipath channel; and a processor. The second wireless signal differs from the first wireless signal due to the wireless multipath channel and a motion of an object in the venue.

Abstract: The disclosed subject matter relates to a microcapsule including a particle core selected from activated carbon (AC), super activated carbon (SAC), MOF composition, multifunctional material or a mixture thereof and a water-soluble polymer shell, including a membrane into which the microcapsule is incorporated, a membrane with exposed AC, SAC, MOF, or multifunctional materials or mixture thereof formed therefrom and methods used is the formation of all of the above.

Abstract: Disclosed herein is a method of forming origami 3-dimensional (3D) graphene on a substrate. The method includes steps of, (a) placing the substrate in a reaction chamber; (b) purging the reaction chamber with a flow of a carry gas; (c) providing a flow of hydrogen along with the carry gas into the reaction chamber; (d) elevating the temperature of the substrate tin the reaction chamber to a level that is close to its melting point; (e) providing a flow of a precursor gas to the reaction chamber for a period of time to deposit a flat graphene layer onto the substrate; and (f) cooling the substrate of step (e) to ambient temperature thereby transforming the flat graphene layer into the origami 3D graphene.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing inputs using recurrent neural networks. One of the systems includes a main recurrent neural network comprising one or more recurrent neural network layers and a respective hyper recurrent neural network corresponding to each of the one or more recurrent neural network layers, wherein each hyper recurrent neural network is configured to, at each of a plurality of time steps: process the layer input at the time step to the corresponding recurrent neural network layer, the current layer hidden state of the corresponding recurrent neural network layer, and a current hypernetwork hidden state of the hyper recurrent neural network to generate an updated hypernetwork hidden state.

Abstract: A method and a spherical field of view gimbal system (“gimbal system”) are provided, comprising: a suspension arm having a first terminal and a second terminal; a first gimbal arm connected to the suspension arm; a second L-shaped gimbal arm connected to the gimbal arm; and a third L-shaped gimbal arm connected to the first gimbal arm; and controller circuit configured to measure the errors in rotational angles and adjust the rotations of the second L-shaped gimbal arm and second L-shaped gimbal arm accordingly.

Abstract: Provided herein are methods and systems for determining joint information, such as joint angle and range of movement, quickly and accurately without using cumbersome, specialized equipment even when visibility to the joint may be limited. Such methods and systems may be achieved by using computationally efficient approaches that are accurate and robust enough to handle noises and occlusions. The methods and systems provided herein may provide quick, low-cost, on-demand approaches to obtain information about the joint health of a patient. The joint angle information may facilitate diagnosis, treatment, prognosis, and/or rehabilitation of the patient from a joint injury, pain, or discomfort by providing quick, actionable information to the healthcare provider.

Abstract: A method for determining a final architecture for a neural network to perform a particular machine learning task is described.

Abstract: Aspects of the disclosure are directed to jointly searching machine learning model architectures and hardware architectures in a combined space of models, hardware, and mapping strategies. A search strategy is utilized where all models, hardware, and mappings are evaluated together at once via weight sharing and a supernetwork. A multi-objective reward function is utilized with objectives for quality, performance, power, and area.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating output sequences using auto-regressive decoder neural networks. In particular, during generation, adaptive early exiting is used to reduce the time required to generate the output sequence.

Abstract: The invention relates to a method for producing graphene flakes by electrochemical exfoliation, where electrodes (anode and cathode) together with electrolyte constitute an electrochemical circuit with current flowing through it, characterized in that the electrodes are immersed at least partially in the electrolyte and electrolysis is carried out, during which graphene flakes detach from the electrode to be released into the electrolyte solution and then the exfoliated graphene flakes are recovered from the electrolyte solution.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating output sequences using auto-regressive decoder neural networks. In particular, during generation, adaptive early exiting is used to reduce the time required to generate the output sequence.

Abstract: Mechanical joints and processes for using same. The mechanical joint can be configured to provide an articulated connection between a first member and a second member. In some embodiments, the mechanical joint can include a support member and an arm. The support member can define an aperture therethrough and a socket on an inner surface thereof. The arm can include a ball disposed on a first end thereof that can be engaged within and supported by the socket defined by the support member. A second end of the arm can extend through the aperture of the support member such that the arm can rotate relative to the support member about a longitudinal axis of the arm and at least partially rotate about two axes that pass through a center point of the ball that are perpendicular to one another and to the longitudinal axis of the arm.

Abstract: A self-balancing Two-Rotation Driving Mechanism (TRDM) uses parallelogram and has low weight, high load, high accuracy, quick response and good adaptation with multi equipment. A longitudinal rotation motor assembly is constrained fully with a base assembly and drives a longitudinal rotation assembly through connecting rod parallelogram. The lateral rotation motor assembly is assembled in the hub of a longitudinal rotation assembly and rotates an equipment clamping assembly in a lateral direction. Generally, the TRDM can drive equipment clamping assembly in two rotations, with longitudinal rotation in the range of ±45° and lateral rotation in range of ±20°. A rotation limitation is reached when limitation shafts touch the end of respective grooves.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating author vectors. One of the methods includes obtaining a set of sequences of words, the set of sequences of words comprising a plurality of first sequences of words and, for each first sequence of words, a respective second sequence of words that follows the first sequence of words, wherein each first sequence of words and each second sequence of words has been classified as being authored by a first author; and training a neural network system on the first sequences and the second sequences to determine an author vector for the first author, wherein the author vector characterizes the first author.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a document classification neural network. One of the methods includes training an autoencoder neural network to autoencode input documents, wherein the autoencoder neural network comprises the one or more LSTM neural network layers and an autoencoder output layer, and wherein training the autoencoder neural network comprises determining pre-trained values of the parameters of the one or more LSTM neural network layers from initial values of the parameters of the one or more LSTM neural network layers; and training the document classification neural network on a plurality of training documents to determine trained values of the parameters of the one or more LSTM neural network layers from the pre-trained values of the parameters of the one or more LSTM neural network layers.