6 Hits in 3.8 sec

SDQ: Stochastic Differentiable Quantization with Mixed Precision [article]

Xijie Huang, Zhiqiang Shen, Shichao Li, Zechun Liu, Xianghong Hu, Jeffry Wicaksana, Eric Xing, Kwang-Ting Cheng
2022 arXiv   pre-print
In this work, we present a novel Stochastic Differentiable Quantization (SDQ) method that can automatically learn the MPQ strategy in a more flexible and globally-optimized space with smoother gradient  ...  SDQ outperforms all state-of-the-art mixed or single precision quantization with a lower bitwidth and is even better than the full-precision counterparts across various ResNet and MobileNet families, demonstrating  ...  Conclusion We have presented Stochastic Differentiable Quantization (SDQ), a learning based mixed precision quantization framework that can optimize the optimal MPQ strategy automatically.  ... 
arXiv:2206.04459v3 fatcat:2a7y44fxwbaw5hod6w2l3bsp5e

Dynamics of Shuttle Devices [article]

Andrea Donarini
2005 arXiv   pre-print
We have kept the somehow mixed notation with the differential operator ∂ P to keep the notation lighter. In general the operators are acting on all what is on their right.  ...  a quantized oscillator.  ...  We believe that equation (7. 95) contains the relevant dynamics of the SDQS in the coexistence regime, but, despite all the approximations already introduced the problem still looks untractable.  ... 
arXiv:cond-mat/0501242v1 fatcat:b335rswugfbvnbfo7b4l266key

Hanbury Brown–Twiss interferometry and second-order correlations of inflaton quanta

Massimo Giovannini
2011 Physical Review D  
The quantum mechanical results are compared and contrasted with different situations including the one where intensity correlations are the result of a classical stochastic process.  ...  After adapting the description of the quantized scalar and tensor modes of the geometry to the analysis of intensity correlations, the normalized degrees of first-order and second-order coherence are computed  ...  Later on different authors applied it to scalar field matter with particular attention to the quantization of the fluctuations [49, 50] (see also [51] ).  ... 
doi:10.1103/physrevd.83.023515 fatcat:a7oi6jauefflrpb4yuts2s4gz4

Atom–surface diffraction: a trajectory description

R. Guantes, A.S. Sanz, J. Margalef-Roig, S. Miret-Artés
2004 Surface science reports  
In particular, special emphasis on surface rainbows and resonance profiles is given with the help of the catastrophe theory. #  ...  After a short presentation of the fundamentals on classical, semiclassical and Bohmian theories, a very detailed analysis of the scattering of He atoms from both Cu(1 1 a) metallic surfaces (with a ¼ 0  ...  for the differential reflection coefficients.  ... 
doi:10.1016/j.surfrep.2004.02.001 fatcat:nvcvhvmbqfgm7mjjua4ygrre2y

One dimensional bosons: From condensed matter systems to ultracold gases

M. A. Cazalilla, R. Citro, T. Giamarchi, E. Orignac, M. Rigol
2011 Reviews of Modern Physics  
Beginning with results from exactly solvable models and computational approaches, we introduce the concept of bosonic Tomonaga-Luttinger Liquids relevant for one-dimension, and compare it with Bose-Einstein  ...  The mixed phase was instead studied using stochastic series expansions by Sengupta and Pryadko (2007) .  ...  et al., 1996; Prokof'ev and Svistunov, 1998; 2009) and by the stochastic series expansion (SSE) algorithm with nonlocal updates (Sandvik, 1992; . c.  ... 
doi:10.1103/revmodphys.83.1405 fatcat:kseirxg74bhobnlcr6jyygpwdy

Extraction of Urban Environmental Quality Indicators using LiDAR-Based Digital Surface Models

Cláudio Magalhães Carneiro
Beyond his irreplaceable assistance and opinions, he granted me with the necessary independence to conduct my research thinking and follow my own ideas.  ...  of sufficient precision.  ...  An important innovation delivered by this thesis consists on the integration of UEQ indicators resulting from different stakes and domains (mixed scenarios), which allows to derive more precise results  ... 
doi:10.5075/epfl-thesis-5050 fatcat:qobicp45ebfzrkehyvhjx6yn6u