A circuit for implementing a polar decoder is described. The circuit includes a log-likelihood ratio processing circuit. A path metric update circuit is coupled to receive log-likelihood values for decoded bits from the log-likelihood ratio processing circuit, wherein the path metric circuit generates path metric values for the decoded bits. A partial sum calculation circuit is coupled to receive the path metrics; and a sort and cull circuit is coupled to receive a list of child path, wherein the sort and cull circuit eliminates invalid paths from the list of child paths. A method of implementing a polar decoder is also described.

An execution unit for a processor, the execution unit comprising: a look up table; a preparatory circuit configured to determine an index value in dependence upon the operand and search the look up table using the index value to locate an entry comprising a natural logarithm associated with the index value; control circuitry configured to provide a first value determined in dependence upon the operand and a second value determined in dependence upon the operand as inputs to at least one multiplier circuit of the execution unit so as to evaluate terms of a Taylor series expansion of a natural logarithm, wherein the control circuitry is configured to provide the natural logarithm associated with the index value and the terms of the Taylor series expansion as inputs to at least one addition circuit so as to generate a mantissa of a natural logarithm of the operand.

Systems and methods for end-to-end encryption and dynamic resizing and encoding into grouped byte channels are described herein. A query is homomorphically encrypted at a client using dynamic channel techniques. The encrypted query is sent without a private key to a server for evaluation over target data to generate encrypted response without decrypting the encrypted query. The result elements of the encrypted response are grouped, co-located, and dynamically resized and encoded into grouped byte channels using the dynamic channel techniques, without decrypting the encrypted query or the encrypted response. The encrypted response is sent to the client where the client uses the private key and channel extraction techniques associated with the dynamic channel techniques to decrypt and perform channel extraction on the encrypted response to obtain the results of the query without revealing the query or results to a target data owner, an observer, or an attacker.

Systems and methods for efficient fixed-base multi-precision exponentiation are disclosed herein. An example method includes applying a multi-precision exponentiation algorithm to a base number, the multi-precision exponentiation algorithm comprises a pre-generated lookup table used to perform calculations on the base number, the pre-generated lookup table comprising pre-calculated exponentiated values of the base number.

Disclosed are a method and apparatus for accelerating inference of a neural network model, an electronic device, and a medium. The method includes: acquiring image training data, text training data, or speech training data; determining a first neural network model to be accelerated; converting a preset operation on a preset network layer in the first neural network model to a first operation for simulating operational logic of a target operation to obtain a second neural network model; performing, based on the image training data, the text training data, or the speech training data, quantization aware training on the second neural network model by a preset bit width to obtain a third neural network model which is quantized; and converting the first operation of the third neural network model to the target operation, to obtain a target neural network model, which is accelerated, corresponding to the first neural network model.

A non-linear activation function in a neural network may be approximated by one or more linear functions. The input range may be divided into input segments, each of which corresponds to a different exponent in the input range of the activation function and includes input data elements having the exponent. Target accuracies may be assigned to the identified exponents based on a statistics analysis of the input data elements. The target accuracy of an input segment will be used to determine one or more linear functions that approximate the activation function for the input segment. An error of an approximation of the activation function by a linear function for the input segment may be within the target accuracy. The parameters of the linear functions may be stored in a look-up table (LUT). During the execution of the DNN, the LUT may be used to execute the activation function.

Implementations of the disclosure provide logarithm and anti-logarithm operations on a hardware processor based on linear piecewise approximation. An example processor includes a piece wise linear log approximation circuit that receives an input of a floating-point number comprising a sign, an exponent and a mantissa. The piece wise linear log approximation circuit approximates a fractional portion of a fixed point number using a linear approximation of the mantissa of the floating-point number. The piece wise linear log approximation circuit also derives an integer from the exponent.

A method of converting a wide dynamic range (WDR) image to a low dynamic range (LDR) image have the steps of: (i) obtaining a transfer function that has a plurality of sub-functions, each sub-function corresponding to a non-overlapped input interval of the dynamic range of the WDR image; (ii) determining the intensity of each pixel of the LDR image by using the transfer function and at least the intensity value of the corresponding pixel of the WDR image; and (iii) outputting the LDR image.

A method and apparatus for determining an output value representing a picture data by applying a piece-wise linear function on an input value representing a picture data, said piece-wise linear function comprising at least one piece characterized by a slope value ai, an offset value and an interval defined over a range of values comprised between a lower bound and an upper bound. The method comprises: obtaining (50) a coded input value X coded with quantization on n bits; obtaining (51) a coded slope value Ai, said coded slope value Ai representing a mantissa value ai_m coded with m bits and an exponent of 2 value ai_e coded with (Km) bits, K being the total number of bits to code said slope value ai; identifying the (Km) most significant bits of the coded slope, said (Km) most significant bits forming an exponent value ai_e of a decoded slope value; identifying the m least significant bits of the coded slope value, said m least significant bits forming the mantissa ai_m of the decoded slope value; determining (52) the output value y by shifting the product of the mantissa ai_m of the decoded slope and the coded input data X by an integer value equal to (n+mai_ep).

Systems and methods for end-to-end encryption and dynamic resizing and encoding into grouped byte channels are described herein. A query is homomorphically encrypted at a client using dynamic channel techniques. The encrypted query is sent without a private key to a server for evaluation over target data to generate encrypted response without decrypting the encrypted query. The result elements of the encrypted response are grouped, co-located, and dynamically resized and encoded into grouped byte channels using the dynamic channel techniques, without decrypting the encrypted query or the encrypted response. The encrypted response is sent to the client where the client uses the private key and channel extraction techniques associated with the dynamic channel techniques to decrypt and perform channel extraction on the encrypted response to obtain the results of the query without revealing the query or results to a target data owner, an observer, or an attacker.