...

2024.md

@@ -27,7 +27,7 @@ tags = ["2024", "juliacn", "meetup", "winter", "见面会", "冬季", "program",
 </tr>
 <tr>
 <td>9:00PM-9:40PM</td>
-<td title="">Symbolic-numerics. New methods we have that do better than traditional numerical solvers, <strong>Chris Rackauckas</strong>, ResearchAffiliate (Co-PI of the Julia Lab)，MIT (在线)</td>
+<td title="High performance computing is usually talked about in the domain of numerical methods: numerically solving PDEs, numerically solving optimization problems, etc. However, the real great sages of performance have always noted that they do something a little different in their domain. In computational fluid dynamics, split solvers can solve for the pressure separately from the velocity, while in robotics many small tricks in the problem formulation improve numerical integration performance by orders of magnitude. These tricks are always regarded as the reason why such domains cannot use general methods, but is that really the case? In this talk we will introduce symbolic-numerics, a form of numerical methods which weave in symbolic manipulation in order to allow for more performance than the numerics on its own. We will showcase the many symbolic-numeric algorithms already possible in the Julia SciML ecosystem, and some of the methods soon to be released, and show how many of the domain-specific tricks that people currently do by hand can be automated in a symbolic-numeric system.">Symbolic-numerics. New methods we have that do better than traditional numerical solvers, <strong>Chris Rackauckas</strong>, ResearchAffiliate (Co-PI of the Julia Lab)，MIT (在线)</td>
 </tr>
 <tr>
 <td>10:00PM-10:40PM</td>