The shell model is a powerful theoretical framework for studying the nuclear structure. Unfortunately, the exponential scaling of the many-body space with the number of nucleons or the number of valence levels strongly restricts its applicability. The Quantum Monte-Carlo (QMC) methods may then be considered as a possible alternative to the direct diagonalization of the Hamiltonian. They are based on a stochastic reformulation of the Schrödinger equation to reduce the many-body problem to a set of one-body problems, numerically tractable, describing independent particles that evolve in fluctuating external fields. The originality of the QMC scheme proposed in the present thesis is the use of a variational approach, with symmetry restoration before variation, to guide the Brownian motion and to constrain it in order to control the sign/phase problem that generally occurs in the QMC samplings for fermions. The "yrast" spectroscopy we obtain for sd- and fp-shell nuclei with realistic residual interactions agree remarkably well with the results from an exact diagonalization of the Hamiltonian. Moreover, an openness towards strongly correlated electronic systems is presented through new QMC schemes recently developed for the two-dimensional Hubbard model. In contrast with the traditional samplings, they guarantee positive-weighted trajectories regardless the on-site interaction strength or the doping of the lattice. We demonstrate that these schemes are in fact related to the stochastic approach applied to the nuclear shell model. The origin of the systematic errors that emerge in these methods, although free from sign/phase problem with the Hubbard Hamiltonian, is also discussed.
