With the development of genetically encoded fluorescent calcium indicators, recording neurons with 2-photon microscopy in an awake, behaving animal has become an important neuroscience technique which allows to correlate behavior and neuronal activity, providing a better understanding of the learning and memory circuits. The main focus of this thesis has been to identify statistical signatures of learning in the neural activity of olfactory bulb granule neuron dendrites recorded using 2-photon calcium imaging as the animal performed an operant conditioning task. We employ self-supervised learning to project dendrite activity time-series into a low-dimensional latent space. Performing supervised classification on compressed representations show that neural activity strongly predicts odors presented to the animal and the reward received. We next identify a robust signature of the animal’s learning encoded in the neural signal. In the second part of the thesis, we aim to ease and accelerate dendrite structures segmentation in microscopic images. Visualization and interaction with complex 3D data can be improved by virtual reality. We allow users to easily provide image annotations, from which we apply a cloud based one-shot learning to segment microscopic and medical images, allowing fast computation and robust accuracy.