Abstract: Floating-point numbers are compressed for neural network computations. A compressor receives multiple operands, each operand having a floating-point representation of a sign bit, an exponent, and a fraction. The compressor re-orders the operands into a first sequence of consecutive sign bits, a second sequence of consecutive exponents, and a third sequence of consecutive fractions. The compressor then compresses the first sequence, the second sequence, and the third sequence to remove at least duplicate exponents. As a result, the compressor can losslessly generate a compressed data sequence.

Abstract: A system stores parameters of a feature extraction network and a refinement network. The system receives an input including a degraded image concatenated with a degradation estimation of the degraded image; performs operations of the feature extraction network to apply pre-trained weights to the input to generate feature maps; and performs operations of the refinement network including a sequence of dynamic blocks. One or more of the dynamic blocks dynamically generates per-grid kernels to be applied to corresponding grids of an intermediate image output from a prior dynamic block in the sequence. Each per-grid kernel is generated based on the intermediate image and the feature maps.