4 Matching Annotations
  1. Nov 2018
    1. Rethinking floating point for deep learning

      【网络的压缩加速问题】

      Facebook人工智能研究院的Jeff Johnson改进了一种新颖的浮点数表示法(posit),使其更加适用于神经网络的训练和推理,并在FPGA上进行了对比实验。和IEEE-754浮点数标准相比,本论文基于改进的浮点数系统,可以实现低bit神经网络训练和高效推理,不再需要后续的量化压缩过程就可以部署在嵌入式等资源受限终端。该论文提出的方法区别于神经网络模型的剪枝、量化等常规思路,直接从浮点数表示这个更加基本、底层的角度尝试解决模型的压缩加速问题,是一个很新颖的方式,且效果不错,值得深入研究。除了论文,作者还给出了代码实现和博客文章,帮助理解。

  2. Oct 2018
    1. Narratives that describe time as uniform and evolving throughout history towards more accelerated states have also been critiqued for theirpotential to reinforce social inequalities (Sharma 2014) and for justifyingthe appropriation of natural resources in unsustainable ways (Bastian 2012).

      This loosely couples with the degrowth discourses around steady state economies and possible political ecologies

  3. Oct 2017
    1. ‘Truly it would seem as if “Man strews the earth with ruin.”4 But this conclusion is too flattering to human vanity. Man's most permanent memorial is a rubbish-heap, and even that is doomed to be obliterated’ (Sherlock, 1922, p. 343

      CO2 atmospheric concentration used as simple indicator for many years to track great acceleration / progression in Anthropocence, this now joined by long list of other indicators, escalating at an alarming rate, population, water use/ shortage, paper consumption, global warming, increase in number and ferocity of storms .......

    2. In 1873, the Italian geologist and priest Antonio Stoppani suggested that our technologies, infrastructures, and patterns of land use had created fundamental changes in Earth’s systems, propelling us into what he called an ‘anthropozoic era’

      Note : Read over Article again by Will Steffen, Paull J Crutzen & John R McNeill. [] (https://www.pik-potsdam.de/news/public-events/archiv/alter-net/former-ss/2007/05-09.2007/steffen/literature/ambi-36-08-06_614_621.pdf)

      Explore development of Anthropocence. How do we track progression of Anthropocene? CO2 Emissions??