Pozvaní prednášajúci

Johannes Lengler, ETH Zürich: Optimization is Hard, but Hillclimbing is Easy --- Right?

Hillclimbing is an integral part of every optimization algorithm. One would expect that our standard hillclimbing algorithms can solve nice, unimodal functions efficiently. For example, strictly monotone functions: consider the search space S={0,1}^n of n-bit strings; then a strictly monotone function f: S -> R is a function that satisfies f(x) < f(y) for all different strings x,y in S such that x_i <= y_i holds for all 1 <= i <= n. Strictly montone functions are the easiest imaginable testbed for hillclimbers: The function f always prefers one-bits over zero-bits, so there are always short and clearly labelled paths from any start point to the unique global optimum 11..1. Surprisingly, it has become clear in the last years that many of our standard hillclimbers fail miserably on some strictly monotone functions. In many instances, they need exponential time to find the optimum. This devastating result has deepened our understanding of hillclimbing techniques, and has given new insights into the mechanisms that are behind some hillclimbing techniques, like selective pressure, step size policies, and population effects. These insights have also improved our understanding of optimization beyond hillclimbing, in particular of optimization in dynamic environments.

Dr. Johannes Lengler has an unusually broad research spectrum, reaching from pure mathematics to neuroscience. He studied mathematics at Saarland (Germany) and Warwick University (UK) and received his PhD in algebraic number theory at Saarland University in 2010. Afterwards, he joined ETH Zürich (Switzerland) as a postdoc, by now as senior scientist, and simultaneously entered the fields of computer science and of neuroscience. Within computer science, his main research areas include nature-inspired optimization algorithms, random graph models for large real-world networks, and stochastic processes in graphs and networks. Within neuroscience, his focus lies on neuroplasticity, memory, and rapid learning. Other publications of Lengler are concerned with efficient communication protocols, the power of decision-makers with limited choices, computational geometry, and quantum computing.

Lukáš Vrábel (Babcock & Bonbright): Predikcia cien nehnuteľností v Kaliforni

Prednáška predstaví našu snahu za posledný rok a pol o rozbehnutie startupu na oceňovanie nehnuteľností v USA. Startup s názvom Babcock & Bonbright bol budovaný v rámci portfolia Central Europe AI - startupového štúdia zameraného na využitie umelej inteligencie na inováciu vo finančnej oblasti a zdravotníctve.

Prezentácia bude zameraná hlavne na technologický aspekt projektu - spojenie štrukturovaných dát, NLP, a spracovania obrazu do komplexného systému na odhad hodnoty obytných domov a bytov pomocou strojového učenia. Pozornosť bude venovaná aj spracovaniu mapových dát, architektúry systému a návrhu datových pipeline.

Lukáš Vrábel vyštudoval Obor "Umelá Inteligencia" na Fakulte Informačných technologií VUT Brno, kde následne pracoval ako výskumník v oblasti formálnych jazykov a gramatík. Okrem akademických skúseností načerpal prax aj z komerčnej sféry - vo firme Seznam.cz, kde pôsobil na rôznych pozíciách od výskumníka až po vedúceho oddelenia výskumu, sa vrátil k umelej inteligencii a strojovému učeniu. Do startupovej scény vstúpil pri zakladaní Brnenskej pobočky pre americko-európske startup studio Central Europe AI, kde zároveň pracoval na aplikovaní strojového učenia na odhad hodnoty nemovitostí. Aktuálne pracuje ako konzultant na volnej nohe, pomáha rôznym firmám vypracovávať data-science projekty a rozvíjať datovo-výzkumné týmy.