Abstract: A method and apparatus for classification using a neural network. A classification apparatus includes at least one processor and a memory. The memory is configured to store a classifier and a preprocessor including a defensive noise generator. The at least one processor generates defensive noise from an input image through the defensive noise generator in the preprocessor, generates a combined image by combining the input image and the defensive noise, and outputs a classification result by inputting the combined image to the classifier.

Abstract: A display device may include: a substrate including a display area and a non-display area; and pixels in the display area, and each including sub-pixels. Each sub-pixel may include a pixel circuit layer, and a display element layer including at least one light emitting element. The display element layer may include: a first electrode on the pixel circuit layer; a second electrode on the first electrode and electrically insulated from the first electrode; the light emitting element including a first end portion coupled to the first electrode and a second end portion coupled to the second electrode, and between the first electrode and the second electrode; an intermediate layer enclosing at least one area of the light emitting element, and on the first electrode; a connection line electrically connected to the second electrode. The second electrode may be on the intermediate layer.

Abstract: A display device comprises a first electrode, a second electrode disposed to be spaced apart from the first electrode and face the first electrode, a first insulating layer disposed to cover the first electrode and the second electrode, a second insulating layer disposed on at least a part of the first insulating layer and exposing at a part of a region where the first electrode and the second electrode overlaps the first insulating layer and at least one light emitting element on the exposed first insulating layer between the first electrode and the second electrode, wherein the second insulating layer includes at least one opening exposing the first insulating layer and disposed to be spaced apart from each other on a region where the first electrode and the second electrode face each other, and a bridge portion between the openings, and the light emitting element is disposed on the opening.

Abstract: A method and apparatus with pruning are disclosed. A method is performed by an apparatus including a processor, the method includes determining weight importance of a trained neural network, receiving a constraint condition related to an operation resource, and determining, in accordance with the constraint condition, a pruning mask for maximizing the weight importance of the trained neural network.

Abstract: A method and apparatus with neural network control are provided. In one general aspect, a method is performed by at least one processor of an apparatus, the method includes selecting neural networks from among available neural networks with respective input image resolutions, including plural resolutions, wherein the selected neural networks are selected based on a sum of amounts of mutual information between the selected neural networks being minimal and based on a limiting condition of a use of the available neural networks, and configuring the available neural networks to perform an inference on input data, wherein the configuring is based on the combination of the selected neural networks.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network. One of the methods includes obtaining a batch of training items and a ground truth assignment; processing the training items in the batch using the neural network to generate respective embeddings for each of the training items; and adjusting the current values of the network parameters by performing an iteration of a neural network training procedure to optimize an objective function that penalizes the neural network for generating embeddings that do not result in, for each possible clustering assignment other than the ground truth assignment, the oracle clustering score being higher than a clustering score for the possible clustering assignment by at least a structured margin between the possible clustering assignment and the ground truth assignment.

Abstract: A method includes obtaining a color transformation matrix that removes color distortion of an original color histogram of a target color space for an input image of an original color space; obtaining a color transformation image of the original color space from which color distortion of the input image is removed using the color transformation matrix; and obtaining a space transformation image with a minimum target loss value between an output vector of a neural network for one of candidate images and each of a plurality of candidate class vectors of the input image, the candidate images being spaced apart from the color transformation image of the original color space by a distance less than a threshold.

Abstract: A method and apparatus for classification using a neural network. A classification apparatus includes at least one processor and a memory. The memory is configured to store a classifier and a preprocessor including a defensive noise generator. The at least one processor generates defensive noise from an input image through the defensive noise generator in the preprocessor, generates a combined image by combining the input image and the defensive noise, and outputs a classification result by inputting the combined image to the classifier.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network. One of the methods includes obtaining a batch of training items and a ground truth assignment; processing the training items in the batch using the neural network to generate respective embeddings for each of the training items; and adjusting the current values of the network parameters by performing an iteration of a neural network training procedure to optimize an objective function that penalizes the neural network for generating embeddings that do not result in, for each possible clustering assignment other than the ground truth assignment, the oracle clustering score being higher than a clustering score for the possible clustering assignment by at least a structured margin between the possible clustering assignment and the ground truth assignment.