The main subject of this thesis is the physics of spin glasses. After their introduction in the 70s in order to describe dilute magnetic alloys, spin-glass models have been considered prototype models to understand the behavior of supercooled liquids. Among the systems that can be described and analyzed using the language of disordered systems, there are problems of combinatorial optimization. In the first part of the thesis, we consider spin-glass models with Kac interactions in order to investigate the supercooled phase of glass-forming liquids. Afterwards, we show how some features of spin-glass models can be described by ubiquitous results of Random Matrix Theory in connection with Extreme Value Statistics. Finally, from the interaction of spin-glass theory and computer science, we put forward a new algorithm of immediate application in Financial problems.
