A system for fusion of multimodal receives a spatial input and a temporal input, wherein the spatial input comprises spatial data having spatial embeddings and the temporal input comprises temporal data having temporal embeddings. The spatial embeddings and the temporal embeddings have different time dimensions. A spatial data output with the spatial embeddings having a same time dimension as the temporal embeddings is generated from the spatial data based on a spatial perception model. The spatial perception model is pre-trained. A temporal data output is generated from the temporal data based on a temporal model. The spatial data output and the temporal data output are combined into an output representing dependencies between the spatial input and the temporal input using a fusion model. A desired target variable is obtained from the output and one of an estimated or predicted value is generated based on the desired target value.

An apparatus configured to receive an input dataset, x, indicative of radar signals reflected from targets as received at a plurality of antenna elements; define a matrix, A, formed of direction-of-arrival-angle vectors, a.sub.n, each direction-of-arrival-angle vector representing an expected response at the plurality of antenna elements of radar signals from one of the targets; define a signal amplitude vector s to represent expected complex amplitudes as received in the radar signals; define an objective function based on x, A and s; search for a set of direction of arrival angles for each of the plurality of targets by the repeated evaluation of the objective function for a plurality of candidate matrices based on matrix A; and wherein said search space comprises a plurality of discrete points, z, associated with the direction of arrival angles by a function of sin(θ.sub.k).

In data communications, a suitably designed contrast coding scheme, comprising a process of contrast encoding (108) at a transmitter end (101) and a process of contrast decoding (120) at a receiver end (103), may be used to create contrast between the bit error rates ‘BERs’ experienced by different classes of bits. Contrast coding may be used to tune the BERs experienced by different subsets of bits, relative to each other, to better match a plurality of forward error correction ‘FEC’ schemes (104, 124) used for transmission of information bits (102), which may ultimately provide a communications system (100) having a higher noise tolerance, or greater data capacity, or smaller size, or lower heat.

Recordings from microphones that provide 1.sup.st order Ambisonics signals, so-called B-format signals, offer a limited cognition of sound directivity. Sound sources are perceived broader than they actually are, especially for off-center listening positions, and the sound sources are often located to be coming from the closest speaker positions. In a method and apparatus for enhancing the directivity of 1.sup.st order Ambisonics signals, additional directivity information is extracted (SFA) from the lower order Ambisonics input signal. The additional directivity information is used to estimate higher order Ambisonics coefficients, which are then combined with the coefficients of the input signal. Thus, the directivity of the Ambisonics signal is enhanced, which leads to an increased accuracy of spatial source localization when the Ambisonics signal is decoded to loud speaker signals. The resulting output signal has more energy than the input signal.

Apparatus for predicting a predetermined block of a picture using a plurality of reference samples The apparatus is configured to form a sample value vector out of the plurality of reference samples, derive from the sample value vector a further vector onto which the sample value vector is mapped by a predetermined invertible linear transform, compute a matrix-vector product between the further vector and a predetermined prediction matrix so as to obtain a prediction vector, and predict samples of the predetermined block on the basis of the prediction vector.

A processor can be associated with a memory for storing convolution data. A plurality of M filters from a corresponding plurality of M input channels to a selected one output channel can be provided, wherein each filter can be represented by a corresponding index, m. Each of the M filters can be partitioned into K respective filter partitions, wherein each respective filter partition can be represented by a corresponding index, k. A frequency-domain representation of each filter partition can be provided, wherein each frequency-domain representation of a filter partition comprises N frequency bins and a corresponding frequency-domain filter coefficient, wherein each respective frequency bin can be represented by a corresponding index, n. The memory can store such information in an arrangement suitable for the processor to concurrently receive sufficient information to concurrently convolve a frame of each input signal with the respective filters.

The invention relates to a method for monitoring a process for refining a feedstock of hydrocarbons, in which: a) a signal representative of the transverse relaxation time of the different entities of an effluent resulting from said refining process, in particular an effluent comprising solid entities, is acquired by proton NMR, b) the signal measured is modeled using a mathematical function comprising several components, each component corresponding to a dynamic range of the entities of said effluent, c) the following are extracted from each of the components of the mathematical function: the transverse relaxation time of each of the components, the intensity of each of the components, d) a value of parameter characteristic of said effluent is determined from at least one intensity determined in stage c), e) a signal for controlling the refining process is generated as a function of said characteristic parameter.

Provided is a method for an electronic device to perform inference based on encrypted data received from an external device, using an artificial intelligence (AI) model, the method including: transforming the AI model to perform inference based on encrypted data, generating parameter information including information about at least one parameter for encrypting data to be input to the AI model, based on the transformed AI model, transmitting the parameter information to the external device, receiving, from the external device, data encrypted based on the parameter information, and obtaining an inference result output from the transformed AI model by inputting the encrypted data to the transformed AI model.

A method for monitoring a hydrostatic bearing that is in operation is provided. Frequency domain analysis, time domain analysis and principal components analysis are performed on an operation signal that results from the operation of the hydrostatic bearing, so as to build a Gaussian mixture model. Then, based on a difference between the Gaussian mixture model and a predetermined reference model, an operation state of the hydrostatic bearing can be determined in real time.

A method for inspecting normality of a spindle of a machine tool is provided. Spectral analysis, time domain analysis and principal components analysis are performed on a vibration signal that results from the vibration of the spindle, so as to build a Gaussian mixture model. Then, based on a difference between the Gaussian mixture model and a predetermined reference model, whether the machine tool is operating normally can be determined in real time.