DQ-STP: An Efficient Sparse On-Device Training Processor Based on Low-Rank Decomposition and Quantization for DNN.
Baoting LiDanqing ZhangPengfei ZhaoHang WangXuchong ZhangHongbin SunNanning ZhengPublished in: IEEE Trans. Circuits Syst. I Regul. Pap. (2024)
Keyphrases
- low rank
- tensor decomposition
- low rank matrix
- rank minimization
- low rank matrices
- low rank subspace
- training process
- missing data
- nuclear norm
- robust principal component analysis
- low rank representation
- matrix factorization
- convex optimization
- linear combination
- singular value decomposition
- group sparsity
- matrix completion
- kernel matrices
- semi supervised
- matrix decomposition
- kernel matrix
- high order
- low rank approximation
- regularized regression
- training set
- training data
- minimization problems
- sparse matrix
- binary matrices
- singular values
- high dimensional
- affinity matrix
- binary matrix
- stochastic gradient descent
- data matrix
- sparse representation
- low rank and sparse
- active learning