Tsen-Wei Hsu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Tsen-Wei Hsu has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
B. N. Lin
Department of Physics, National Tsing Hua University, Hsinchu 300, Taiwan
1 shared publicationC. H. Chang
Department of Physics, National Central University, Chungli 320, Taiwan
1 shared publicationC. H. Lin
Department of Chemistry, Tamkang University, Tamsui 251, Taiwan
1 shared publicationH. C. Chiang
Department of Physics, National Tsing Hua University, Hsinchu 300, Taiwan
1 shared publicationH. C. Ku
Department of Physics, National Tsing Hua University, Hsinchu 300, Taiwan
1 shared publicationH. M. Luo
Department of Physics, National Tsing Hua University, Hsinchu 300, Taiwan
1 shared publication
Latest Publications
- Open Access
- 9 November 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Embedded Computing Systems
- Vol. 22 (6) , 1-22
SG-Float: Achieving Memory Access and Computing Power Reduction Using Self-Gating Float in CNNs
Convolutional neural networks (CNNs) are essential for advancing the field of artificial intelligence. However, since these networks are highly demanding in terms of memory and computation, implementing CNNs can be challenging. To make CNNs more accessible to energy-constrained devices, researchers are exploring new algorithmic techniques and hardware designs that can reduce memory and computation requirements. In this work, we present self-gating float (SG-Float), algorithm hardware co-design of a novel binary number format, which can significantly reduce memory access and computing power requirements in CNNs. SG-Float is a self-gating format that uses the exponent to self-gate the mantissa to zero, exploiting the characteristic of floating-point that the exponent determines the magnitude of a floating-point value and the error tolerance property of CNNs. SG-Float represents relatively small values using only the exponent, which increases the proportion of ineffective mantissas, corresponding to reducing mantissa multiplications of floating-point numbers. To minimize the accuracy loss caused by the approximation error introduced by SG-Float, we propose a fine-tuning process to determine the exponent thresholds of SG-Float and reclaim the accuracy loss. We also develop a hardware optimization technique, called the SG-Float buffering strategy, to best match SG-Float with CNN accelerators and further reduce memory access. We apply the SG-Float buffering strategy to vector-vector multiplication processing elements (PEs), which NVDLA adopts, in TSMC 40nm technology. Our evaluation results demonstrate that SG-Float can achieve up to 35% reduction in memory access power and up to 54% reduction in computing power compared with AdaptivFloat, a state-of-the-art format, with negligible power and area overhead. Additionally, we show that SG-Float can be combined with neural network pruning methods to further reduce memory access and mantissa multiplications in pruned CNN models. Overall, our work shows that SG-Float is a promising solution to the problem of CNN memory access and computing power.
- 9 November 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Embedded Computing Systems
- Vol. 22 (6) , 1-22
Open AccessSG-Float: Achieving Memory Access and Computing Power Reduction Using Self-Gating Float in CNNs
Convolutional neural networks (CNNs) are essential for advancing the field of artificial intelligence. However, since these networks are highly demanding in terms of memory and computation, implementing CNNs can be challenging. To make CNNs more accessible to energy-constrained devices, researchers are exploring new algorithmic techniques and hardware designs that can reduce memory and computation requirements. In this work, we present self-gating float (SG-Float), algorithm hardware co-design of a novel binary number format, which can significantly reduce memory access and computing power requirements in CNNs. SG-Float is a self-gating format that uses the exponent to self-gate the mantissa to zero, exploiting the characteristic of floating-point that the exponent determines the magnitude of a floating-point value and the error tolerance property of CNNs. SG-Float represents relatively small values using only the exponent, which increases the proportion of ineffective mantissas, corresponding to reducing mantissa multiplications of floating-point numbers. To minimize the accuracy loss caused by the approximation error introduced by SG-Float, we propose a fine-tuning process to determine the exponent thresholds of SG-Float and reclaim the accuracy loss. We also develop a hardware optimization technique, called the SG-Float buffering strategy, to best match SG-Float with CNN accelerators and further reduce memory access. We apply the SG-Float buffering strategy to vector-vector multiplication processing elements (PEs), which NVDLA adopts, in TSMC 40nm technology. Our evaluation results demonstrate that SG-Float can achieve up to 35% reduction in memory access power and up to 54% reduction in computing power compared with AdaptivFloat, a state-of-the-art format, with negligible power and area overhead. Additionally, we show that SG-Float can be combined with neural network pruning methods to further reduce memory access and mantissa multiplications in pruned CNN models. Overall, our work shows that SG-Float is a promising solution to the problem of CNN memory access and computing power.
- 1 June 2000
- Published by American Physical Society (APS) in Physical Review B
- Vol. 61 (21) , 14825-14832
Variation of anomalous Pr ordering and crystal symmetry for the oxygenated systemStructural, transport, magnetic, calorimetric, and neutron data were reported for the oxygen-annealed or 1212-type system Powder x-ray Rietveld analysis indicates that due to subtle oxygen distribution in the plane for these oxygenated cuprates, two structural symmetry transitions were observed, from orthorhombic (CuO chain) O(I) phase (space group Pmmm) for to tetragonal 1212 T phase for and then to a different type of orthorhombic 1212 O(II) phase (Cmmm) for Resistivity data indicate that Pr-in-Ba is not favorable for metallic state and no superconductivity can be detected for these insulating cuprates. Magnetic-susceptibility, heat-capacity, and neutron-diffraction data show that, regardless of the structural transitions, the anomalous Néel temperature decreases monotonically and smoothly from 18 K for to 2.5 K for with the same c-axis antiferromagnetic Pr spin alignment. The increasing Pr-O bond length observed with decreasing indicates that this unusual Pr magnetic ordering is closely correlated with the wave-function overlap between orbital and eight orbital in the bilayer.- 1 June 2000
- Published by American Physical Society (APS) in Physical Review B
- Vol. 61 (21) , 14825-14832
Variation of anomalous Pr ordering and crystal symmetry for the oxygenated systemStructural, transport, magnetic, calorimetric, and neutron data were reported for the oxygen-annealed or 1212-type system Powder x-ray Rietveld analysis indicates that due to subtle oxygen distribution in the plane for these oxygenated cuprates, two structural symmetry transitions were observed, from orthorhombic (CuO chain) O(I) phase (space group Pmmm) for to tetragonal 1212 T phase for and then to a different type of orthorhombic 1212 O(II) phase (Cmmm) for Resistivity data indicate that Pr-in-Ba is not favorable for metallic state and no superconductivity can be detected for these insulating cuprates. Magnetic-susceptibility, heat-capacity, and neutron-diffraction data show that, regardless of the structural transitions, the anomalous Néel temperature decreases monotonically and smoothly from 18 K for to 2.5 K for with the same c-axis antiferromagnetic Pr spin alignment. The increasing Pr-O bond length observed with decreasing indicates that this unusual Pr magnetic ordering is closely correlated with the wave-function overlap between orbital and eight orbital in the bilayer.