Timo Berthold is a Director at FICO, leading the MIP research and development team of the FICO Xpress Solver. In additon, he is a lecturer at the Mathematical Optimization Department of TU Berlin, working of the intersection of academia and industry. Before joining FICO, Timo was a main developer of the open-source MIP and MINLP solver SCIP at Zuse Institue Berlin. Timo is an expert on all aspects of computational mixed-integer linear and nonlinear optimization, heuristic methods, and recently, the integration of ML methods into optimization solvers. He has published over 50 papers in this field, supervised many talented students and won multiple prestigious awards for his research. As a fun-fact, his PhD thesis won the 2014 GOR dissertation award and made the finals of the EURO dissertation award, while at the same time, an accompanying article on his PhD work received a science-communication prize for being the best "Math research explained to the general public" article of the year.

Contents of this video:
0:00 - Intro
1:51 - Family background
2:43 - Memories from German reunification
5:17 - 1990 world cup
5:48 - Avid reader
6:51 - Moving to Lietzen in 1994
9:19 - Attending a Stephen Hawking’s talk in Potsdam in 1999
11:16 - Interest in computer games
13:00 - Shy kid?
13:59 - Passion for Punk Rock
16:01 - Interest in Biology in early teenage years
16:46 - Good student?
18:31 - Reason for choosing Mathematics over Computer Science
20:22 - Moving to Berlin in 2001 to begin university
22:57 - Transition from high school to university
24:36 - Learning about Optimization
26:15 - Influential master’s thesis on primal heuristics for mixed integer programs
29:19 - 7-year PhD journey
31:01 - PhD dissertation on “Heuristic algorithms in global MINLP solvers"
36:14 - Winning an award for public understanding of science
38:17 - Joining Fico in 2014
39:07 - Juggling between work, PhD, raising a young child, and supporting a pregnant wife
40:34 - Typical workday
41:39 - Making impact through various activities
43:31 - Partnership between ZIB and FICO
44:07 - Xpress longevity and its ability to remain competitive over the decades
45:35 - Types of optimization problems Xpress can solve
46:11 - First major commercial solver capable of solving non-convex MINLP problems to optimality
47:22 - Numerical issues in MIP solvers: what can a modeler do?
48:31 - To linearize or not to linearize?
51:35 - Does Xpress incorporate constraint programming techniques?
52:31 - Using ML to improve performance of Xpress
54:27 - The profile of Xpress users and customers
57:13 - Good hardware for running a MIP solver and the path consistency problem
1:00:11 - Free academic licences
1:00:32 - Friendship with Tobias Achterberg from Gurobi
1:01:41 - Fair play in the MIP community
1:03:07 - Becoming a lecturer at TU Berlin in 2022
1:04:00 - Plans for the future
1:05:14 - Encouraging young talents to work on computational MIP
1:06:18 - Concluding remarks

свернуть