diff --git a/paper.bib b/paper.bib index 20fd1862bc68e5a099987c6b76e8baa2208e0960..72f0fc25010ade4497bfc0913644ddffbe26971b 100755 --- a/paper.bib +++ b/paper.bib @@ -65,3 +65,29 @@ year={2008}, school={Rice University} } + +@article{chen2017intelligent, + title={Intelligent adaptive sampling guided by Gaussian process inference}, + author={Chen, Yuhang and Peng, Chaoyang}, + journal={Measurement Science and Technology}, + volume={28}, + number={10}, + pages={105005}, + year={2017}, + publisher={IOP Publishing} +} + +@article{takhtaganov2018adaptive, + title={Adaptive Gaussian process surrogates for Bayesian inference}, + author={Takhtaganov, Timur and M{\"u}ller, Juliane}, + journal={arXiv preprint arXiv:1809.10784}, + year={2018} +} + +@online{mathematica_adaptive, + author = {Stephen Wolfram}, + title = {{M}athematica: {A}daptive {P}lotting}, + year = 2011, + url = {http://demonstrations.wolfram.com/AdaptivePlotting/}, + urldate = {2019-09-10} +} diff --git a/paper.md b/paper.md index e0c85e5a5d93478c58b7ecfee1932725df259537..39268606e995279e080f2f356ac9792d0ba2a80a 100755 --- a/paper.md +++ b/paper.md @@ -24,9 +24,9 @@ Even though it is suboptimal, one usually resorts to sampling $X$ on a homogeneo #### Choosing new points based on existing data improves the simulation efficiency. <!-- This should convey the point that it is advantageous to do this. --> -A better alternative which improves the simulation efficiency is to choose new, potentially interesting points in $X$ based on existing data. [@gramacy2004parameter; @de1995adaptive; @castro2008active]<!-- cite i.e. hydrodynamics, Bayesian sampling --> -Baysian optimization works well for high-cost simulations where one needs to find a minimum (or maximum). -If the goal of the simulation is to approximate a contineous function with the least amount of points, the continuity of the approximation is achieved by a greedy algorithm that samples mid-points of intervals with the largest Euclidean distance. <!-- cite literature to support this claim that it is better, Mathematica and MATLAB maybe --> +A better alternative which improves the simulation efficiency is to choose new, potentially interesting points in $X$ based on existing data. [@gramacy2004parameter; @de1995adaptive; @castro2008active; @chen2017intelligent] <!-- cite i.e. hydrodynamics--> +Baysian optimization works well for high-cost simulations where one needs to find a minimum (or maximum). [@@takhtaganov2018adaptive] +If the goal of the simulation is to approximate a contineous function with the least amount of points, the continuity of the approximation is achieved by a greedy algorithm that samples mid-points of intervals with the largest Euclidean distance. [@mathematica_adaptive] <!-- cite literature to support this claim that it is better, Mathematica and MATLAB maybe --> Such a sampling strategy would trivially speedup many simulations. One of the most significant complications here is to parallelize this algorithm, as it requires a lot of bookkeeping and planning ahead.