" Lossless" CompressionofDeepNeuralNetworks: AHigh-dimensionalNeuralTangentKernelApproach
–Neural Information Processing Systems
In this paper, building upon recent advances inneural tangent kernel (NTK) andrandom matrix theory (RMT), we provide anovelcompression approach towide and fully-connecteddeepneural nets.
Neural Information Processing Systems
Feb-7-2026, 16:43:46 GMT