6 results about "Discrete transformation" patented technology

A black box code search model backdoor attack method based on learnable discrete code transformation comprises the steps that a learnable backdoor generator is constructed, and malicious codes with backdoors are generated on the premise that internal parameters of a damaged model are not accessed through the learnable discrete code transformation. Firstly, an agent model capable of simulating victim model behaviors is trained through query-response data; secondly, on the proxy model, a discrete selection process is differentiable by using a re-parameterization sampling technology, and a backdoor generator is trained in combination with a multi-objective loss function so as to realize effectiveness and concealment of backdoor implantation; according to the method, the limitation of a black box is successfully bypassed by constructing the proxy model, and a new possibility is provided for an attacker. A backdoor generation process is integrated into a differentiable training framework through learnable discrete transformation, so that end-to-end learnability is realized, and an attack strategy can be automatically optimized.

The application discloses a thermal superstructure manufacturing method, device and system based on discrete transformation thermology, and belongs to the technical field of heat conduction and heat flow regulation. The method comprises the following steps: carrying out discrete simplex grid dissection on a design domain to obtain a grid, and specifying a hard constraint for an internal function boundary according to a regulation function; defining a piecewise linear mapping on the grid and solving by two-stage optimization, wherein a negative Jacobian determinant is punished in the first stage to eliminate folding to obtain a bijective mapping, and anisotropy is inhibited by minimizing conformal distortion in the second stage; and finally, the thermal conductivity tensor field of the physical domain is derived by using a Jacobian matrix J, microstructure matching and filling are carried out, and additive manufacturing is adopted. The application solves the problems of folding and extreme parameters which are prone to occur in complex geometric transformation thermology design, can realize efficient heat management and omnidirectional heat regulation under any complex shape, and the error of experimental verification is less than 5 percent.

The application discloses an ADS-B spoofing attack detection method based on a discrete cosine patch attention mechanism, and comprises the following steps: step 1, screening and cleaning the collected ADS-B signal sequence, retaining continuous flight data with time sequence characteristics for network training; step 2, dividing the continuous flight data into a plurality of time windows according to channels, and extracting frequency domain features from the time windows by using a cosine discrete transformation; step 3, combining the time windows with frequency domain features and the time windows with time features, inputting the obtained time windows with frequency domain and time domain features into a self-attention mechanism to obtain instance-level features; step 4, normalizing the instance-level features by using anti-instance regularization; step 5, measuring the deviation of the normalized features from normal mode features learned in the training phase; step 6, summarizing the measured deviation, calculating the overall anomaly score of each signal segment, and then classifying the signals by using the anomaly score.

A decoder includes circuits configured to receive a bitstream; determine a first, second, and third region of a current block; and, according to a geometric partitioning mode, decode the current block. Related devices, systems, techniques, and articles are also described. The decoder can determine, from a signal contained in the bitstream, a type of encoding transformation to decode the first, second, and / or third regions. The type of encoding transformation characterizes at least one inverse block cosine discrete transformation and one inverse adaptive block cosine discrete transformation. The decoding of the current block can include using the determined transformation type as the inverse transformation for each of the regions.

In order to solve the limitation of a traditional transform domain watermarking technology in the aspect of resisting traditional attacks and geometric attacks, the invention discloses a robust watermarking scheme based on discrete Hankel Transform (DHT), the robust watermarking scheme is remarkably different from a one-dimensional quantization index modulation method susceptible to noise interference, four-dimensional feature vectors of DHT coefficients are projected to a four-dimensional feature space, and the four-dimensional quantization index modulation method based on the DHT is obtained. The inherent stability and error correction capability of the high-dimensional quantization lattice are utilized, so that the resistance to noise interference and geometric attack is enhanced, and high-dimensional quantization index modulation is realized. The method not only has good invisibility and robustness, but also can effectively solve the problem of weak digital watermark resistance of the color image, and is suitable for occasions of digital media copyright protection.

The invention belongs to the technical field of JADCE algorithms, and relates to an efficient and accurate JADCE algorithm based on model-driven deep learning, the key of GFRA is to realize accurate joint activity detection and channel estimation, and due to sporadic traffic of Internet of Things equipment, the problem can be regarded as a typical sparse signal recovery problem. In order to solve the problem, a traditional compressed sensing algorithm can be utilized, but the defects of high calculation complexity and lack of robustness for different sensing matrixes exist. The invention provides a memory-enhanced iterative shrinkage threshold algorithm network framework for JADCE, and the performance of SSRe is effectively improved by learning discrete transformation and enhancing residual domain information interaction. Particularly, a high-throughput short-term memory mechanism is introduced into an expanded multi-stage deep network, so that the information richness and interpretability of the network are ensured. Simulation results show that the provided network has significant advantages in the aspects of detection and estimation performance, and the existing model-driven deep expansion network.