Abstract: In one example aspect, the present disclosure provides an example computer-implemented method for generating feedback signals for training a machine-learned agent model. The example method can include obtaining an output of a machine-learned agent model, the output including a next state feature generated by the machine-learned agent model based on a sequence of preceding states. The example method can include processing, using a machine-learned reward model, the output and the sequence of preceding states to generate a quality indicator indicating a quality of the next state feature in view of the preceding states. The machine-learned reward model could be trained by retrieving reference data from a reference data source and computing one or more quality indicators in view of a respective training input and output(s), and the reference data. The example method can include outputting the quality indicator to a model trainer for updating the machine-learned agent model.

Abstract: The present invention provides a polyamic acid aqueous solution composition capable of polymerizing polyamic acid in water rather than in an organic solvent, as well as achieving a high imidization rate during low-temperature curing.

Abstract: The present application provides a polyamic acid aqueous solution composition prepared of a polyimide, wherein the composition enables the polymerization of hydrophobic-based monomers into a polyamic acid in water and is improved in transparency, eco-friendliness, storage stability, and the like when cured.

Abstract: A method for training a neural network includes receiving a training input audio sequence including a sequence of input frames defining a hotword that initiates a wake-up process on a user device. The method further includes obtaining a first label and a second label for the training input audio sequence. The method includes generating, using a memorized neural network and the training input audio sequence, an output indicating a likelihood the training input audio sequence includes the hotword. The method further includes determining a first loss based on the first label and the output. The method includes determining a second loss based on the second label and the output. The method further includes optimizing the memorized neural network based on the first loss and the second loss associated with the training input audio sequence.

Abstract: The present invention relates to a method for manufacturing a meat substitute for crustacean meat. A method for manufacturing a meat substitute for crustacean meat, according to one embodiment of the present invention, comprises (S100) a step for dissolving starch in water to prepare a starch gel, and (S200) a step for adding the starch gel into an external nozzle of a 3D printer, which has a double nozzle in which an internal nozzle is inserted and arranged inside the external nozzle, adding a fish meat composition into the internal nozzle, then discharging the starch gel and fish meat composition at the same time and 3D-printing into a predetermined form.

Abstract: In a hybrid electric vehicle and a touchpoint learning method therefor, while a motor is driven, an engagement oil pressure of an engine clutch is adjusted step-by-step, to learn a touchpoint, thereby reducing touchpoint learning time and improving learning accuracy. The touchpoint learning method may include: determining whether a touchpoint learning entry condition previously set is satisfied or not; controlling a first motor connected to one end of an engine clutch to be in a stopped state when the learning entry condition is satisfied; maintaining a second motor connected to the other end of the engine clutch at a preset speed; changing an engagement oil pressure of the engine clutch step-by-step; and learning a touchpoint of the engine clutch based on a torque change of the second motor.

Abstract: The present invention relates to a method for manufacturing plant-based meat with artificial muscle fiber inserted therein. The method for manufacturing plant-based meat with artificial muscle fiber inserted therein, according to an embodiment of the present invention, comprises: a step of preparing an artificial muscle fiber composition by mixing alginic acid, carrageenan, and glucomannan in distilled water (S100); and a step of injecting the artificial muscle fiber composition into an inner nozzle of a 3D printer having double nozzles in which the inner nozzle is inserted inside an outer nozzle, injecting a vegetable protein composition into the outer nozzle, and then conducting 3D printing (S200).

Abstract: A method by a convergence production device for simultaneously producing 2D content and 3D content according to the present disclosure including the steps of: generating a canvas for generating the content; disposing an element on the canvas; registering an event corresponding to the element; registering (1) an action that is a condition for performing a function corresponding to the event, (2) the function representing an attribute change and a specific action to be performed with respect to a target if the event is activated, and (3) the target that becomes a purpose of the function; and executing a result, if the event is detected, based on the action, and the result includes the function and the target, and the content includes the 2D content and the 3D content.

Abstract: The present specification provides a method for a terminal performing visual coding, the method including: generating a page for the visual coding and placing an asset on the page; setting a target asset that is a target of the visual coding on the basis of the placed asset; setting a user motion associated with an interaction with a user; setting a result associated with the target asset on the basis of the user motion; and displaying the result of the target asset on the basis of the user motion inputted, in which the result includes: a “function” for controlling size, position and state value of the target asset; a “computation” for computing variables associated with the target asset; and a “function page” representing a movement of the page.

Abstract: A method for training hotword detection includes receiving a training input audio sequence including a sequence of input frames that define a hotword that initiates a wake-up process on a device. The method also includes feeding the training input audio sequence into an encoder and a decoder of a memorized neural network. Each of the encoder and the decoder of the memorized neural network include sequentially-stacked single value decomposition filter (SVDF) layers. The method further includes generating a logit at each of the encoder and the decoder based on the training input audio sequence. For each of the encoder and the decoder, the method includes smoothing each respective logit generated from the training input audio sequence, determining a max pooling loss from a probability distribution based on each respective logit, and optimizing the encoder and the decoder based on all max pooling losses associated with the training input audio sequence.

Abstract: Teacher-student learning can be used to train a keyword spotting (KWS) model using augmented training instance(s). Various implementations include aggressively augmenting (e.g., using spectral augmentation) base audio data to generate augmented audio data, where one or more portions of the base instance of audio data can be masked in the augmented instance of audio data (e.g., one or more time frames can be masked, one or more frequencies can be masked, etc.). Many implementations include processing augmented audio data using a KWS teacher model to generate a soft label, and processing the augmented audio data using a KWS student model to generate predicted output. One or more portions of the KWS student model can be updated based on a comparison of the soft label and the generated predicted output.

Abstract: A method for detecting a hotword includes receiving a sequence of input frames that characterize streaming audio captured by a user device and generating a probability score indicating a presence of a hotword in the streaming audio using a memorized neural network. The network includes sequentially-stacked single value decomposition filter (SVDF) layers and each SVDF layer includes at least one neuron. Each neuron includes a respective memory component, a first stage configured to perform filtering on audio features of each input frame individually and output to the memory component, and a second stage configured to perform filtering on all the filtered audio features residing in the respective memory component. The method also includes determining whether the probability score satisfies a hotword detection threshold and initiating a wake-up process on the user device for processing additional terms.

Abstract: A method for training hotword detection includes receiving a training input audio sequence including a sequence of input frames that define a hotword that initiates a wake-up process on a device. The method also includes feeding the training input audio sequence into an encoder and a decoder of a memorized neural network. Each of the encoder and the decoder of the memorized neural network include sequentially-stacked single value decomposition filter (SVDF) layers. The method further includes generating a logit at each of the encoder and the decoder based on the training input audio sequence. For each of the encoder and the decoder, the method includes smoothing each respective logit generated from the training input audio sequence, determining a max pooling loss from a probability distribution based on each respective logit, and optimizing the encoder and the decoder based on all max pooling losses associated with the training input audio sequence.

Abstract: Disclosed is a hybrid electric vehicle and a method of controlling an engine stop for the vehicle. The method includes: determining whether a first condition and a second condition is satisfied; applying a negative torque to the motor when the first condition and the second condition are satisfied; predicting occurrence of reverse rotation of the engine; and applying a feedback torque for satisfying the target behavior of the engine to the motor when the occurrence of the reverse rotation is predicted.

Abstract: Disclosed are a hybrid electric vehicle and a method for controlling an engine speed of a hybrid electric vehicle. In particular, the method includes: determining, when an engine on one end and a first motor are directly connected and while driving in a state where an engine clutch is connected and a second motor is connected to the other end, a compensation value based on the difference in the rotational speed of the engine and the second motor when a control entry is caused, which induces the rotational speed mismatch of the engine and the second motor; and compensating an output torque of the first motor based on the compensation value.

Abstract: A method for detecting a hotword includes receiving a sequence of input frames that characterize streaming audio captured by a user device and generating a probability score indicating a presence of a hotword in the streaming audio using a memorized neural network. The network includes sequentially-stacked single value decomposition filter (SVDF) layers and each SVDF layer includes at least one neuron. Each neuron includes a respective memory component, a first stage configured to perform filtering on audio features of each input frame individually and output to the memory component, and a second stage configured to perform filtering on all the filtered audio features residing in the respective memory component. The method also includes determining whether the probability score satisfies a hotword detection threshold and initiating a wake-up process on the user device for processing additional terms.

Abstract: A power supply for electric vehicles includes: a main battery mounted to a vehicle to supply electric power to a motor generating a driving power for the vehicle; a supplementary battery charged with electric power to be supplied to the main battery; a battery rack provided in a vehicle body to load the supplementary battery; and a power converter configured to convert electricity power of the supplementary battery mounted to the battery rack to charge the main battery.

Abstract: The present invention relates to a heat transfer fluid composition with improved thermal stability, and has the effect of inhibiting the oxidation of a coolant due to hydrogen vehicle stack cooling system components and operating temperature.

Abstract: A method of classifying a cerebrovascular branch in cerebrovascular image is disclosed, the method including receiving, by an analysis device, an cerebrovascular image of a subject; extracting, by the analysis device, a plurality of vascular unit structures from the cerebrovascular image based on geometric features of a 3D model; extracting, by the analysis device, feature values for each of the plurality of vascular unit structures; inputting, by the analysis device, the feature values of each of the plurality of vascular unit structures into a first learning model trained in advance, and classifying chunks to which each of the plurality of vascular unit structures belongs; and inputting, by the analysis device, the feature values of each of the vascular unit structures belonging to same chunk into a second learning model trained in advance, and classifying a vessel branch composed of the vascular unit structures belonging to the same chunk.

Abstract: The present disclosure relates to a cooling liquid composition for an electric vehicle. The purpose of the composition according to the present disclosure is to effectively control heat generated during the operation of electronic components which are constituent elements of the electric vehicle, thereby enhancing the durability of the components and ensuring stability. The present disclosure has the effect that the anti-corrosion properties between components of a cooling system are excellent, while the effect of the generation of precipitates that form when electricity flows is small.