Session K58: Delbruck Award Symposium

Dynamic scaling in natural swarms

Collective behavior is widespread in biological systems across many different scales and organisms. As physicists, our hope is that the (complex) details of the individuals are not important when looking at collective properties, and that large scale behavior can be characterized in terms of general laws, much as we do in condensed matter. However, this assumption cannot be given for granted and must be experimentally justified.
With this perspective, I will present experimental evidence of the emergence of dynamic scaling laws in natural swarms. We find that spatio-temporal correlation functions in different swarms can be rescaled by using a single characteristic time, which grows with the correlation length with a dynamical critical exponent z~1. Numerical simulations and previous computations on models of self-propelled particles in the swarming phase give different exponents, suggesting that natural swarms belong to a novel dynamic universality class. This conclusion is strengthened by experimental evidence of non-exponential relaxation, indicating that previously overlooked inertial effects are needed to describe swarm dynamics. The absence of a purely relaxational regime suggests that natural swarms are subject to a near-critical censorship of hydrodynamics.   

Field Potentials in the Fly’s Photoreceptor-LMC Synapse: A Possible Mechanism for Regularizing Vesicle Release

The first synapse in the fly visual system, from photoreceptor to Large Monopolar Cell (LMC), is capable of transmitting information at high rates. The primary physiological mechanism for signal transmission in this system involves the release of synaptic vesicles, leading to graded postsynaptic voltage changes. Previous studies have suggested that measured information rates are difficult to explain on the assumption that the photoreceptor terminal releases vesicles according to Poisson statistics (de Ruyter van Steveninck and Laughlin, Nature 379: 642-645 (1996)). Work by Weckström and Laughlin (J Neurosci 30(28): 9557-9566 (2010)) has demonstrated that the LMC postsynaptic currents generate substantial extracellular potentials in the extracellular space. Here I will explore the possibility that these field potentials play a role in regularizing the statistics of vesicle release, enabling the synapse to transmit information at high rates while maintaining a modest average vesicle release rate.
  

Quantitative genetics and the missing heritability problem

For many heritable traits, including susceptibility to common diseases in humans, regions of the genome uncovered by genetic mapping studies explain only a minority of the heritable contribution to trait variance in the population. This “missing heritability problem” has become a central issue in human genetics, as well as in genetics of quantitative traits more broadly. We set out to investigate genetics of quantitative traits and the causes of missing heritability issues in yeast, a simple genetic model organism. To do so, we set up a very large panel of progeny from a cross between two divergent parent strains. We then used genomic approaches to accurately estimate different sources of heritable variation for 46 quantitative traits and to detect underlying genomic regions with high statistical power. We find that the detected regions explain nearly the entire additive contribution to heritable variation for the traits studied. We also quantify the contribution to heritability of nonlinear gene-gene interactions. Our results are consistent with the hypothesis that missing heritability arises primarily from a very large number of genetic factors with very small individual effects. These factors can be discovered in studies with sufficiently large sample sizes, although the optimal study designs depend on the population frequency spectra of the factors. I will describe ongoing work to measure the contributions of factors with different population frequencies.   

The Information Bottleneck Theory of Deep Neural Networks

Multilayered Deep forward Neural Networks (DNN), trained by Stochastic Gradient Decent (SGD), perform amazingly well on multiple supervised learning tasks. Understanding why and how is still a major scientific challenge. In this line of work, we show that large-scale layered networks, when trained with SGD, achieve -- later by layer - the Information Bottleneck optimal universal (architecture independent) tradeoff between sample complexity and accuracy, for problems which are successively refineable in the information theoretic sense. In that sense, DNN's are provably optimal universal learning machines. Moreover, this optimality is achieved through stochastic relaxation via the noisy gradients to \textit{locally} Gibbs distributions on the weights of the network. The theory provides ample new predictions: interpretation of the hidden layers; equivalent architectures for a given task; mechanisms for self-organized hierarchical representations; exactly solvable models and relations to symmetries and invariants; mechanisms for transfer learning; new biologically plausible learning principles, and more. In this talk I will describe some of these predictions and relate them to Bill Bialek's insights and scientific achievements.   

Max Delbrück Prize in Biological Physics Talk: Precision and emergence in the physics of biological function

Life is more than the sum of its parts: functions crucial for life emerge from interactions among hundreds or thousands of microscopic components. Less obvious, perhaps, is that the mechanisms of life are extraordinarily precise: our visual system counts single photons, many signaling systems are limited by the random arrivals of individual molecules, and more. Observations of extreme precision suggest a theoretical framework in which biological systems have been exquisitely tuned, optimizing performance in the presence of physical constraints. Observations of emergence suggest a different theoretical framework, in which functional behaviors are collective, and hence perhaps insensitive to microscopic details. I will give examples of both approaches, in systems ranging from a developing embryo to large networks of neurons, and from computation in the visual system to flocks of birds. At the end I will try to reconcile the two points of view. I hope to make clear why I believe that a more unified, and unifying, theoretical physics of biological systems is within reach.