One of the most distinctive features of natural intelligence compared to AI is its extraordinary versatility. What allows animals to learn quickly and flexibly, and continually adapt their skills to changing environments and conditions? My group develops theoretical and computational approaches to answer different instances of this question. Areas of focus in the lab include: Statistical and biophysical models of perceptual learning. Context has a powerful effect on the neural representation of sensory inputs. Accordingly, perception adapts to top-down information that is learned through experience, such as the salience or valence of sensory stimuli. We use coarse-grained statistical approaches as well as detailed biophysical models to unravel the brain mechanisms that underpin perceptual learning. We focus on the olfactory system but sometimes our models are sufficiently abstract to be applicable across sensory modalities. Currently, we are interested in understanding the function of adult neurogenesis, a striking form of plasticity specific to the olfactory system. Our goal is to provide insights into the computational principles and biophysical conditions that make adult neurogenesis advantageous for olfactory perceptual learning compared to more common (e.g. synaptic) forms of plasticity. Our theoretical approaches aim at interpreting existing experimental results as well as making predictions that can be tested in future experiments. Bayesian and complexity theories of high-level cognition. We continuously gather noisy data through our senses to make inferences about past, present, and future states of the world. Accessible information, time and resources are limited and constrain the accuracy and complexity of feasible inference strategies. We develop normative theories to understand how efficient inference processes adapt their complexity to environmental uncertainty and task demands. Reconstruction of the functional connectivity of decision circuits. In collaboration with the Padoa Schioppa lab, we use techniques based on the inference of maximum entropy and non-stationary models to dissect the neural circuits underlying value representation and choice formation in the brain. Statistical physics approaches to memory consolidation and retrieval. The continuous acquisition of new knowledge requires effective and flexible ways to store information in the brain. We will use statistical physics approaches based on the inference and simulation of graphical models to tackle different aspects of this problem. In one project, we will identify the attractors of the neural activity (memories) in hippocampal and cortical recordings and will study their evolution during learning and sleep. Our goal is to provide insights into the dynamics of memory consolidation and the factors that may influence this process, such as neurogenesis in the hippocampus. In parallel, by studying the energy landscape of graphical models with different structural and functional properties (e.g. different network topologies, distributions of synaptic weights and neuronal excitabilities), we aim to shed light on the fundamental relationships between network architecture and memory capacity beyond the classical Hopfield model, and to identify the connectivity features that are crucial for maximizing the representation of information in the brain. We hope that our studies will help identify specific therapeutic targets for Alzheimer’s disease and other conditions that disrupt memory and consciousness.