Abstract: In an embodiments a method includes obtaining a neural network (INN), the neural network having a plurality of neural layers, each layer being capable of being executed according to different implementation solutions and impacting a required memory allocation for the execution of the neural network and/or an execution time of the neural network, defining a maximum execution time threshold of the neural network and/or a maximum required memory allocation threshold for the execution of the neural network, determining an optimal required memory allocation size for the execution of the neural network from possible implementation solutions for each layer of the neural network, determining an optimal execution time of the neural network from the possible implementation solutions for each layer of the neural network and estimating a performance loss or a performance gain in terms of execution time and required memory allocation for each implementation solution of each layer of the neural network.

Abstract: An artificial neural network includes a unit cell. The unit cell includes a first binary two-dimensional convolution layer configured to receive an input tensor and to generate a first tensor. A first batch normalization layer is configured to receive the first tensor and to generate a second tensor. A concatenation layer is configured to generate a third tensor by concatenating the input tensor and the second tensor. A second binary two-dimensional convolution layer is configured to receive the third tensor and to generate a fourth tensor. A second batch normalization layer is configured to generate an output tensor based on the fourth tensor.

Abstract: According to one aspect, there is proposed a method for transforming a trained artificial neural network including a binary convolution layer followed by a pooling layer then a batch normalization layer, the method includes obtaining the trained artificial neural network and transforming the trained artificial neural network such that the order of the layers of the trained artificial neural network is modified by displacing the batch normalization layer after the convolution layer.

Abstract: A method can be used to process an initial set of data through a convolutional neural network that includes a convolution layer followed by a pooling layer. The initial set is stored in an initial memory along first and second orthogonal directions. The method includes performing a first filtering of the initial set of data by the convolution layer using a first sliding window along the first direction. Each slide of the first window produces a first set of data. The method also includes performing a second filtering of the first sets of data by the pooling layer using a second sliding window along the second direction.

Abstract: According to one aspect, a method is provided for detecting events or elements in physical signals, including at least one implementation of a reference artificial neural network, at least one implementation of an auxiliary artificial neural network distinct from the reference artificial neural network. The auxiliary artificial neural network being simplified relative to the reference artificial neural network. At least one assessment of a probability of presence of the event or the element by the implementation of the reference artificial neural network or by the implementation of the auxiliary artificial neural network, where the reference artificial neural network is implemented when the probability of presence of the event or the element is greater than a threshold, and wherein the auxiliary artificial neural network is implemented when the probability of presence of the event or the element is below the threshold.

Abstract: According to one aspect, the disclosure proposes a method for detecting events or features in physical signals by implementing an artificial neural network. The method includes evaluating the probability of presence of the event or feature by implementing the artificial neural network. The method includes implementing the artificial neural network in a nominal mode and to which a physical signal having a first so-called nominal resolution is fed, as long as the probability of the presence of the event or feature is below a threshold. The method further includes implementing the artificial neural network in a reduced consumption mode with a reduced resolution, as long as the probability of the presence of the event or feature is above the threshold. The reduced resolution is lower than the first resolution.

Abstract: According to one aspect, a method is proposed for detecting events or elements in physical signals by implementing an artificial neural network. The method includes an assessment of a probability of the presence of the event or the element by an implementation of the neural network. The implementation of the neural network according to a nominal mode takes as input a physical signal having a first resolution, called nominal resolution, when the probability of presence of the event or the element is greater than a threshold. The implementation of the neural network according to a low power mode takes as input a physical signal having a second resolution, called reduced resolution, lower than the first resolution, when the probability of presence of the event or the element is below the threshold.

Abstract: A method can be performed prior to implementation of a neural network by a processing unit. The neural network comprising a succession of layers and at least one operator applied between at least one pair of successive layers. A computational tool generates an executable code intended to be executed by the processing unit in order to implement the neural network. The computational tool generates at least one transfer function between the at least one pair of layers taking the form of a set of pre-computed values.

Abstract: In accordance with an embodiment, a method for determining an overall memory size of a global memory area configured to store input data and output data of each layer of a neural network includes: for each current layer of the neural network after a first layer, determining a pair of elementary memory areas based on each preceding elementary memory area associated with a preceding layer, wherein: the two elementary memory areas of the pair of elementary memory areas respectively have two elementary memory sizes, each of the two elementary memory areas are configured to store input data and output data of the current layer of the neural network, the output data is respectively stored in two different locations, and the overall memory size of the global memory area corresponds to a smallest elementary memory size at an output of the last layer of the neural network.

Abstract: According to an aspect, a method is proposed for defining placements, in a volatile memory, of temporary scratch buffers used during an execution of an artificial neural network, the method comprising: determining an execution order of layers of the neural network, defining placements, in a heap memory zone of the volatile memory, of intermediate result buffers generated by each layer, according to the execution order of the layers, determining at least one free area of the heap memory zone over the execution of the layers, defining placements of temporary scratch buffers in the at least one free area of the heap memory zone according to the execution order of the layers.

Abstract: A method can be used to determine an overall memory size of a global memory area to be allocated in a memory intended to store input data and output data from each layer of a neural network. An elementary memory size of an elementary memory area intended to store the input data and the output data from the layer is determined for each layer. The elementary memory size is in the range between a memory size for the input data or output data from the layer and a size equal to the sum of the memory size for the input data and the memory size for the output data from the layer. The overall memory size is determined based on the elementary memory sizes associated with the layers. The global memory area contains all the elementary memory areas.

Abstract: According to one aspect, the disclosure proposes a method for updating an artificial neural network including initial weights stored in a memory at least in an integer format, which method includes: a processing unit determining the error gradients at the output of the layers of the neural network, the processing unit retrieving the initial weights from memory, the processing unit updating the initial weights comprising, for each initial weight, a first calculation of a corrected weight, in the integer format of this initial weight, the processing unit replacing the value of the initial weights stored in the memory by the value of the corrected weights.

Abstract: In an embodiment a method for managing a convolutional calculation carried out by a calculation unit adapted to calculate output data on output channels from convolution kernels applied to input data blocks on at least one input channel, wherein calculations on each input data block correspond respectively to an output datum on an output channel, and wherein the calculations with each convolution kernel correspond to the output data on each output channel respectively includes identifying a size of a memory location available in a temporary working memory of the calculation unit, pre-loading in the temporary working memory a maximum number of convolution kernels storable at the size of the memory; and controlling the calculation unit to calculate a set of output data calculable from pre-loaded convolution kernels.

Abstract: An embodiment method for implementing an artificial neural network in an integrated circuit comprises obtaining an initial digital file representative of a neural network configured according to at least one data representation format, then detecting at least one format for representing at least part of the data of the neural network, then converting at least one detected representation format into a predefined representation format so as to obtain a modified digital file representative of the neural network, and then integrating the modified digital file into an integrated circuit memory.

Abstract: A method is provided for monitoring scene detection by an apparatus detecting scenes from among a set of possible reference scenes. It includes an assignment of an identifier to each reference scene, detection of scenes from among the set of possible reference scenes at successive instants of detection with the aid of at least one classification algorithm, and a sliding time filtering processing of these detected current scenes over a filtering window of size M, based on the identifier of each new detected current scene taken into account in the window and a confidence probability associated with this new detected current scene, the output of the filtering processing successively delivering filtered detected scenes.

Abstract: A method can be performed prior to implementation of a neural network by a processing unit. The neural network comprising a succession of layers and at least one operator applied between at least one pair of successive layers. A computational tool generates an executable code intended to be executed by the processing unit in order to implement the neural network. The computational tool generates at least one transfer function between the at least one pair of layers taking the form of a set of pre-computed values.

Abstract: A method of real-time scene detection performed by a wireless communication device includes, performing a first scene detection measurement to determine that the wireless communication device is located in a first scene. The first scene detection measurement is performed at first instant in time. The first scene is a type of environment. The method further includes associating the first scene with a corresponding reference scene of a predetermined set of reference scenes, determining a reference duration associated with the corresponding reference scene, and performing a second scene detection measurement immediately following expiration of the reference duration measured from the first instant in time.

Abstract: A method for real-time detection of at least one scene by an apparatus, from among a set of possible reference scenes, includes acquiring current values of attributes from measurement values supplied by sensors. The method further includes traversing a path through a decision tree. The nodes of the decision tree are respectively associated with the attributes. The traversal considers at each node along the path, the current value of the corresponding attribute, so as to obtain at the output of the path, a scene from among the set of reference scenes. The obtained scene identifying which reference scene is the detected scene. The method further includes developing a confidence index (SC) associated with the identification of the detected scene.

Abstract: In accordance with an embodiment, a method for determining an overall memory size of a global memory area configured to store input data and output data of each layer of a neural network includes: for each current layer of the neural network after a first layer, determining a pair of elementary memory areas based on each preceding elementary memory area associated with a preceding layer, wherein: the two elementary memory areas of the pair of elementary memory areas respectively have two elementary memory sizes, each of the two elementary memory areas are configured to store input data and output data of the current layer of the neural network, the output data is respectively stored in two different locations, and the overall memory size of the global memory area corresponds to a smallest elementary memory size at an output of the last layer of the neural network.

Abstract: In accordance with an embodiment, a method includes reducing a size of at least one initial parameter of each layer of an initial multilayer neural network to obtain for each layer a set of new parameters defining a new neural network, wherein each new parameter of the set of new parameters has its data represented in two portions comprising an integer portion and a fractional portion; implementing the new neural network using a test input data set applied only once to each layer; determining a distribution function or a density function resulting from the set of new parameters for each layer; and based on the determined distribution function or density function, adjusting a size of a memory area allocated to the fractional portion and a size of the memory area allocated to the integer portion of each new parameter associated with each layer.