Awalin Sopan

Development of new machine learning models is typically done on manually curated data sets, making them unsuitable for evaluating the models' performance during operations, where the evaluation needs to be performed automatically on incoming streams of new data. Unfortunately, pure reliance on a fully automatic pipeline for monitoring model performance makes it difficult to understand if any observed performance issues are due to model performance, pipeline issues, emerging data distribution… Mehr zeigen

Development of new machine learning models is typically done on manually curated data sets, making them unsuitable for evaluating the models' performance during operations, where the evaluation needs to be performed automatically on incoming streams of new data. Unfortunately, pure reliance on a fully automatic pipeline for monitoring model performance makes it difficult to understand if any observed performance issues are due to model performance, pipeline issues, emerging data distribution biases, or some combination of the above. With this in mind, we developed a web-based visualization system that allows the users to quickly gather headline performance numbers while maintaining confidence that the underlying data pipeline is functioning properly. It also enables the users to immediately observe the root cause of an issue when something goes wrong. We introduce a novel way to analyze performance under data issues using a data coverage equalizer. We describe the various modifications and additional plots, filters, and drill-downs that we added on top of the standard evaluation metrics typically tracked in machine learning (ML) applications, and walk through some real world examples that proved valuable for introspecting our models. Weniger anzeigen

In this paper we consider the following problem: Given a set of $n$ Player1 sites in the plane and their Delaunay triangulation ${\cal D}$, place minimum possible Player2 sites such that in the resulting Delaunay triangulation ${\cal D}'$ of the sites of both Players, the neighborship between Player1 sites are as less as possible. We first consider placing minimum number of Player2 sites such that no two Player1 sites are neighbors in ${\cal D}'$. We show that to isolate a Player1 site $p$, two… Mehr zeigen

In this paper we consider the following problem: Given a set of $n$ Player1 sites in the plane and their Delaunay triangulation ${\cal D}$, place minimum possible Player2 sites such that in the resulting Delaunay triangulation ${\cal D}'$ of the sites of both Players, the neighborship between Player1 sites are as less as possible. We first consider placing minimum number of Player2 sites such that no two Player1 sites are neighbors in ${\cal D}'$. We show that to isolate a Player1 site $p$, two Player2 ites are both necessary and sufficient if $p$ is in the convex hull of ${\cal D}$, otherwise three Player2 sites are both necessary and sufficient. This gives a liner time algorithm to individually isolate all Player1 sites by 3n-h$ Player2 sites, where $h$ is the number of sites in the convex hull of ${\cal D}$. Then we give two more algorithms for this problem. The next algorithm runs also in linear time and uses $3(n-1)-h$ Player2 sites, but is much simpler. Our next algorithm uses $5|{\cal M}|$ sites, where $|{\cal M}|$ is the size of a maximum matching in ${\cal D}$, and runs in $O(\sqrt{n}m)$ time, where $m$ is the number of edges of ${\cal D}$. Then we consider isolating sites by components, where a component in ${\cal D'}$ is a maximal connected subset of sites of the same player. We show that it is possible to place $n$ Player2 sites such that in ${\cal D'}$ the number of components among Player1 sites is higher than that among Player2 sites. The above problems are related to at least two existing well known research topics: Voronoi games, where the strategy of each of the two players is to place sites such that in the resulting Voronoi diagram some certain criteria is optimized for each player, and proximity graphs, where this problem is known as minimum stabbing set of Delaunay circles. Our bound of $5|{\cal M}|$ for the first problem would work better than known bound for the minimum stabbing set of Delaunay circles if M has a smaller size. Weniger anzeigen

Engineering and research blog post