Multimode optical fibers are the backbone of telecommunication and medical imaging. When light with high intensity travels through a multimode fiber, photons and matter start to interact and propa-gation becomes nonlinear. The nonlinear propagation of light results in variations in the spatial and temporal distribution of the light. Therefore, at the end of the fiber spatial distribution of the light and/or its wavelength changes. In this thesis, nonlinear interactions in multimode fiber are explored and controlled for laser, compu-ting and machine learning applications. Novel approaches for scalable energy-efficient optical compu-ting, learning, and controlling nonlinear dynamics with machine learning tools, ultrashort pulse gener-ation with superior beam qualities, high peak power and stability are demonstrated. Firstly, high-power ultrashort pulse generation in multimode laser cavities is explored. By studying pulse dynamics, significant improvements are achieved and near-single mode output beam profiles are demonstrated. Later, a novel all-fiber laser design is presented to achieve stable ultrashort pulses with a compact and low-cost laser cavity. In the second half of this thesis, machine learning tools are uti-lized to acquire the relation between the nonlinear frequency generation and the initial excitation of the multimode fibers to create tunable frequency sources. Advanced artificial neural network designs are implemented to learn nonlinear light propagation in multimode fibers to replace time consuming conventional simulations. Finally, the nonlinear interactions shaping the propagating beam distribu-tion are employed to process information to perform optical computing with multimode fibers. Multimode fibers are an ideal testbed to investigate complex nonlinear dynamics in nature. We be-lieve the demonstrated applications and the achieved results are just a subset of the capabilities of the optical fiber. These approaches can be used as a steppingstone to demonstrate advanced applica-tions with light.