Final Report Summary - OPTINF (Optimization and inference algorithms from the theory of disordered systems: theoretical challenges and applications to large-scale inverse problems in systems biology)
OPINF is concerned with the design of optimization and inference algorithms based on advanced statistical physics methods for disordered systems, and with their interdisciplinary applications in computational biology, statistical inference and machine learning.
On the methodological side, we developed several advanced analytic and algorithmic techniques that allowed us to tackle basic problems for which clear methodological and computational bottlenecks existed.
Main examples range from inverse dynamical problems, to stochastic and combinatorial optimization, to inference and learning. The unifying aspect of our techniques resides in the key computational role played by out-of-equilibrium states which has become in our studies a central conceptual guide for the understanding of hardness in computational problems and for the design of efficient algorithms in many different contexts.
The applications have been many, from computational biology, to information theory, condensed matter physics, inference on dynamical processes over networks and machine learning. To give some examples, we have shown how to take advantage of sequencing data to improve our predictions on proteins structures, how to solve the patient-zero problem in epidemic spreading and how to perform learning in large scale artificial neural networks with realistic limited precision contacts.
All these achievements have opened an array of novel research perspectives in out-of-equilibrium statistical physics and in large scale computational problems.
On the methodological side, we developed several advanced analytic and algorithmic techniques that allowed us to tackle basic problems for which clear methodological and computational bottlenecks existed.
Main examples range from inverse dynamical problems, to stochastic and combinatorial optimization, to inference and learning. The unifying aspect of our techniques resides in the key computational role played by out-of-equilibrium states which has become in our studies a central conceptual guide for the understanding of hardness in computational problems and for the design of efficient algorithms in many different contexts.
The applications have been many, from computational biology, to information theory, condensed matter physics, inference on dynamical processes over networks and machine learning. To give some examples, we have shown how to take advantage of sequencing data to improve our predictions on proteins structures, how to solve the patient-zero problem in epidemic spreading and how to perform learning in large scale artificial neural networks with realistic limited precision contacts.
All these achievements have opened an array of novel research perspectives in out-of-equilibrium statistical physics and in large scale computational problems.