Properties of small defect clusters such as vacancies or interstitials underpin any multi-scale model predicting the evolution of materials under extreme conditions [1,2]. This study aims at improving our understanding of the free energy landscape of point defects in body-centered-cubic metals up to the melting temperature by resorting to atomistic simulations based on electronic structure calculations. The phase space is sampled using adaptive molecular dynamics methods [3-5] and ab initio atomic forces. Moreover, taking advantage of versatility of adaptive free energy methods we combine the ab-initio force field with interatomic machine learning based interactions. The present approach merging ab initio – free energy – machine learning methods, can provide quantities such as formation free energies or diffusion transition rates as key input parameter for any subsequent multi-scale simulation. We exemplify this approach in the case of point defects in tungsten and iron. Joint work with A. M. Goryaeva, T. D. Swinburne and J. B. Maillet. [1] M.-C. Marinica, F. Willaime, and J.-P. Crocombette, Phys. Rev. Lett. 108, 025501 (2012). [2] R. Alexander, M.-C. Marinica, L. Proville, F. Willaime, K. Arakawa, M. R. Gilbert, S. L. Dudarev, Phys. Rev. B 94, 024103 (2016). [3] T. Lelièvre, M. Rousset, G. Stoltz. J. Chem. Phys., 126, 134111, (2007). [4] T.D Swinburne, M.-C. Marinica, M.-C., Phys. Rev. Lett. 120, 135503 (2018) [5] L. Cao, G. Stoltz, T. Lelièvre, M.-C. Marinica, and M. Athènes, J. Chem. Phys. 140, 104108, (2014).