Ard Louis

Abstract:

Understanding how genotypes map onto phenotypes, fitness, and eventually organisms is arguably the next major missing piece in a fully predictive theory of evolution. We refer to this generally as the problem of the genotype-phenotype map. Though we are still far from achieving a complete picture of these relationships, our current understanding of simpler questions, such as the structure induced in the space of genotypes by sequences mapped to molecular structures, has revealed important facts that deeply affect the dynamical description of evolutionary processes. Empirical evidence supporting the fundamental relevance of features such as phenotypic bias is mounting as well, while the synthesis of conceptual and experimental progress leads to questioning current assumptions on the nature of evolutionary dynamics-cancer progression models or synthetic biology approaches being notable examples. This work delves with a critical and constructive attitude into our current knowledge of how genotypes map onto molecular phenotypes and organismal functions, and discusses theoretical and empirical avenues to broaden and improve this comprehension. As a final goal, this community should aim at deriving an updated picture of evolutionary processes soundly relying on the structural properties of genotype spaces, as revealed by modern techniques of molecular and functional analysis.

Abstract:

Deep neural networks (DNNs) generalise remarkably well in the overparameterised regime, suggesting a strong inductive bias towards functions with low generalisation error. We empirically investigate this bias by calculating, for a range of architectures and datasets, the probability PSGD(f∣S) that an overparameterised DNN, trained with stochastic gradient descent (SGD) or one of its variants, converges on a function f consistent with a training set S. We also use Gaussian processes to estimate the Bayesian posterior probability PB(f∣S) that the DNN expresses f upon random sampling of its parameters, conditioned on S. Our main findings are that PSGD(f∣S) correlates remarkably well with PB(f∣S) and that PB(f∣S) is strongly biased towards low-error and low complexity functions. These results imply that strong inductive bias in the parameter-function map (which determines PB(f∣S)), rather than a special property of SGD, is the primary explanation for why DNNs generalise so well in the overparameterised regime. While our results suggest that the Bayesian posterior PB(f∣S) is the first order determinant of PSGD(f∣S), there remain second order differences that are sensitive to hyperparameter tuning. A function probability picture, based on PSGD(f∣S) and/or PB(f∣S), can shed light on the way that variations in architecture or hyperparameter settings such as batch size, learning rate, and optimiser choice, affect DNN performance.

Abstract:

Abstract: