Abstract: An apparatus for embedding a sentence feature vector according to an embodiment includes a sentence acquisitor configured to acquire a first sentence and a second sentence, each including one or more words; a vector extractor configured to extract a first feature vector corresponding to the first sentence and a second feature vector corresponding to the second sentence by independently inputting each of the first sentence and the second sentence into a feature extraction network; and a vector compressor configured to compress the first feature vector and the second feature vector into a first compressed vector and a second compressed vector, respectively, by independently inputting each of the first feature vector and the second feature vector into a convolutional neural network (CNN)-based vector compression network.

Abstract: A method for training an image classification model according to an embodiment includes training a feature extractor and a rotation angle classifier to predict a rotation angle of each of unlabeled first training images, training the image classification model to predict a label and rotation angle of each of labeled second training images, but predict a uniform label even though an actual rotation angle of each of the second training images is changed, generating a pseudo label based on a training image that satisfy a preset condition among unlabeled candidate images, and training the image classification model to predict a rotation angle of each of the third training images, and predict a label of each of the third training images based on the pseudo label, but predict a uniform label even though an actual rotation angle of each of the third training images is changed.

Abstract: A hyperparameter optimization method performed by a hyperparameter optimization apparatus to optimize hyperparameters of a model includes calculating an evaluation score for each of a plurality of hyperparameter samples constituting a first hyperparameter sample set by applying each of the hyperparameter samples to a target model, performing weighted clustering of the hyperparameter samples by using the calculated evaluation scores as weights and constructing a second hyperparameter sample set based on the clustering result.

Abstract: A method for training an image classification model according to an embodiment includes training a feature extractor and a rotation angle classifier to predict a rotation angle of each of unlabeled first training images, training the image classification model to predict a label and rotation angle of each of labeled second training images, but predict a uniform label even though an actual rotation angle of each of the second training images is changed, generating a pseudo label based on a training image that satisfy a preset condition among unlabeled candidate images, and training the image classification model to predict a rotation angle of each of the third training images, and predict a label of each of the third training images based on the pseudo label, but predict a uniform label even though an actual rotation angle of each of the third training images is changed.

Abstract: An apparatus for embedding a sentence feature vector according to an embodiment includes a sentence acquisitor configured to acquire a first sentence and a second sentence, each including one or more words; a vector extractor configured to extract a first feature vector corresponding to the first sentence and a second feature vector corresponding to the second sentence by independently inputting each of the first sentence and the second sentence into a feature extraction network; and a vector compressor configured to compress the first feature vector and the second feature vector into a first compressed vector and a second compressed vector, respectively, by independently inputting each of the first feature vector and the second feature vector into a convolutional neural network (CNN)-based vector compression network.

Abstract: A hyperparameter optimization method performed by a hyperparameter optimization apparatus to optimize hyperparameters of a model includes calculating an evaluation score for each of a plurality of hyperparameter samples constituting a first hyperparameter sample set by applying each of the hyperparameter samples to a target model, performing weighted clustering of the hyperparameter samples by using the calculated evaluation scores as weights and constructing a second hyperparameter sample set based on the clustering result.

Abstract: A system and method for analyzing a product fabrication process are disclosed. A product yield analysis system according to an exemplary embodiment of the present disclosure includes a data extraction unit that extracts sensor data from a plurality of sensors arranged in equipment for fabricating a product, a reference signal generation unit that generates a reference signal for each of the plurality of sensors from the sensor data, and a sensor detection unit that detects one or more sensors having a correlation with a yield of the product using the sensor data and the reference signal.

Abstract: A video decoding device for a TV receiver, e.g., SDTV, decodes a compressed video signal transmitted from a transmitting side and displays the decoded signal. The device includes a first downsampling part for downsampling input data in a unit of block at a predetermined ratio; a display size converting part for dividing the downsampled result into a first group of data and a second group of data and after downsampling the first group of data, for performing screen format conversion for the downsampled first group of data into a display size; a memory for storing each of the converted first group of data and the second group of data; and an output part for downsampling the second group of data to correspond with the first group of data to thereby perform screen format conversion for the downsampled second group of data, when the second group of data is reproduced, and for outputting the first and second groups of data.

Abstract: A memory structure and method for implementing an on screen display (OSD) is disclosed. The present invention separates a command area, which stores small but frequently updated data causing fatal errors due to erroneous read, from a bitmap area, which stores relatively larger data, less frequently updated than the command area and which is not sensitive to errors. Thus, OSD memory structure and method allows an external host processor to effectively control the OSD process. Also, the command area further comprises a global command and a plurality of local commands, wherein the global command contains common control information to the local commands and control information other than OSD information, thereby allowing the external host processor to effectively control the OSD even when implementing multistandard OSD where a plurality of bitmap and commands for these bitmap are constructed in an external memory.

Abstract: An apparatus for a fast inverse discrete cosine transform (IDCT) by processing only non-zero IDCT input data including non-zero feeding unit connected to each row-column decomposed one dimensional IDCT input for feeding only non-zero input data during each cycle. The first 1-D IDCT and the second 1-D IDCT have the same structure, and 1-D IDCT core with a non-zero feeding unit at its front end. The 1-D IDCT core of the present invention is for performing IDCT operation based on the input scaling method, and the operation of the IDCT core consists of a cosine kernel selection operation using a received non-zero input, multiplication-and-addition, and a storing operation. The non-zero feeding unit for the first 1-D IDCT and the second one are different in their configuration. There are two different methods for implementing the first 1-D IDCT, a stored input feeding method and a zigzag scanned input feeding method.

Abstract: A decoder for a digital TV receiver is disclosed. The present decoder improves the picture quality of a video signal down converted when an SD class TV receiver receives an HD class video signal. Generally, the decoder receives data in block units, converts a format of the data into a format for display, memorizes the data, processes the memorized data, and displays the processed data by converting the format to include more horizontal color signals than vertical color signals, memorizing and reproducing the data in downsampling the data in block units in a given ratio, and storing the data.

Abstract: An inverse discrete cosine transforming (IDCT) system for a digital television receiver performing a variable length decoding and an inverse quantization of an applied bit stream and an IDCT with respect to inversely quantized DCT coefficients, including an integrated IDCT part which performs an inverse discrete cosine transformation of standards DCT coefficients inversely quantized or high definition DCT coefficients inversely quantized.