Incorporate @ceriottm final feedback

Author: Luthaf

paper/paper.md

@@ -25,25 +25,26 @@ bibliography: paper.bib
 
 # Summary
 
-The number of materials or molecules that can be created by combining
-different chemical elements in various proportions and spatial arrangements is
-enormous. Computational chemistry can be used to generate databases containing
-billions of potential structures [@Ruddigkeit2012], and predict some of the
-associated properties [@Montavon2013; @Ramakrishnan2014]. Unfortunately, the
-very large number of structures makes exploring such database — to understand
+The number of materials or molecules that can be created by combining different
+chemical elements in various proportions and spatial arrangements is enormous.
+Computational chemistry can be used to generate databases containing billions of
+potential structures [@Ruddigkeit2012], and predict some of the associated
+properties [@Montavon2013; @Ramakrishnan2014]. Unfortunately, the very large
+number of structures makes exploring such database — to understand
 structure-property relations or find the *best* structure for a given
-application — a daunting task. In the recent years, multiple molecular
-*descriptors* [@Behler2007; @Bartok2013; @Willatt2019] have been developed to
-compute structural similarities between materials or molecules, incorporating
-physically-relevant information and symmetries. These descriptors can be used
-for unsupervised machine learning applications, such as clustering or
-classification of the different structures, and high-throughput screening of
-database for specific properties [@Maier2007; @De2017; @Hautier2019].
-Unfortunately, the dimensionality of most descriptors is very high, which makes
-the resulting classifications, clustering or mapping very hard to visualize.
-Additional dimensionality reduction algorithm [@Schlkopf1998; @Ceriotti2011;
-@McInnes2018] can reduce the number of relevant dimensions to a handful,
-creating 2D or 3D maps of the full database.
+application — a daunting task. In recent years, multiple molecular
+*representations* [@Behler2007; @Bartok2013; @Willatt2019] have been developed
+to compute structural similarities between materials or molecules, incorporating
+physically-relevant information and symmetries. The features associated with
+these representations can be used for unsupervised machine learning
+applications, such as clustering or classification of the different structures,
+and high-throughput screening of database for specific properties [@Maier2007;
+@De2017; @Hautier2019]. Unfortunately, The dimensionality of these features (as
+well as most of other descriptors used in chemical and materials informatics) is
+very high, which makes the resulting classifications, clustering or mapping very
+hard to visualize. Additional dimensionality reduction algorithm
+[@Schlkopf1998; @Ceriotti2011; @McInnes2018] can reduce the number of relevant
+dimensions to a handful, creating 2D or 3D maps of the full database.
 
 ![The Qm7b database [@Montavon2013] visualized with chemiscope](screenshot.png)
 
@@ -55,11 +56,11 @@ point corresponds to a chemical entity. The axes, color, and style of each point
 can be set to represent a property or a structural descriptor to visualize
 structure-property relations directly. Structural descriptors are not computed
 directly by chemiscope, but must be obtained from one of the many codes
-implementing such descriptors [@librascal; @QUIP]. Since the most common
+implementing general-purpose atomic representation [@librascal; @QUIP] or more specialized descriptors. Since the most common
 descriptors can be very high dimensional, it can be convenient to apply a
 dimensionality reduction algorithm that maps them to a lower-dimensional space
 for easier visualization. For example the sketch-map algorithm [@Ceriotti2011]
-was used with the Smooth Overlap of Atomic Positions descriptor [@Bartok2013] to
+was used with the Smooth Overlap of Atomic Positions representation [@Bartok2013] to
 generate the visualization in Figure 1. The right panel displays the
 three-dimensional structure of the chemical entities, possibly including
 periodic repetition for crystals. Visualizing the chemical structure can help to
@@ -99,14 +100,14 @@ slower, while still handling 100k points easily.
 The use of web technologies makes chemiscope usable from different operating
 systems without the need to develop, maintain and package the code for each
 operating system. It also means that we can provide an online service at
-http://chemiscope.org allowing users to visualize their own dataset without
-any local installation. Chemiscope is implemented as a library of re-usable
+http://chemiscope.org allowing users to visualize their own dataset without any
+local installation. Chemiscope is implemented as a library of re-usable
 components linked together via callbacks. This makes it easy to modify the
 default interface to generate more elaborate visualizations, for example,
 displaying multiple maps generated with different parameters of a dimensionality
 reduction algorithm. Chemiscope can also be distributed in a standalone mode,
 where the code and a predefined dataset are merged together as a single HTML
-file. This standalone mode is useful for archival purposes, for example, as
+file. This standalone mode is useful for archival purposes, for example as
 supplementary information for a published article and for use in corporate
 environments with sensitive datasets.