• Dr. M. Vanitha (ITP-CAS)
    05 May 2017, Friday: 15:45--16:45
    Introductory seminar on nonlinear dynamics of DNA

    In this introductory seminar talk, I will summarise my previous research accomplishments on base pair opening in DNA double helical molecule in the form of solitons. Base pair opening in an inhomogenous double helical molecular chain with flexible strands is studied by solving the governing dynamical equations analytically and numerically. The results show that inhomogeneity in stacking and hydrogen bonds in localized and periodic forms and the helicity do not alter the amplitude of the bubble soliton. However the flexibility of the strands makes the amplitude to damp. On the otherhand the velocity of the bubble soliton is unaltered. Further the above effects introduce small fluctuation in the tail of the bubble soliton without affecting the robust nature of the soliton during propagation. Based on this well-established nonlinear dynamical framework to account for DNA denaturation--annealing dynamics, I will briefly discuss our initial effort to understand the dynamics of coordinated neural activity in the observed neural dynamics from place cell recordings from mice, including mouse models that are relevant to autism and mental illness in general. I will conclude my talk with the current topic of my interest on networks.

  • Dr. Zhijian Wang (Zhejiang University)
    20 February 2017, Monday: 15:30--16:30
    Experimental evidence on winning strategies of prisoner's dilemma

    The iterated prisoner's dilemma game (IPD), relating to the social cooperation, is a basic model in social science. For long time, the mutual cooperation strategy is regarded as the best winning strategy in IPD. But, as recently discovered by Press and Dyson (2012), the zero-determinant (ZD) extortionate strategy can enforce a linear relationship between a pair of players' scores, and can enforce and exploit cooperation, providing the extortionate strategist with a score advantage, and consequently higher scores than those from the mutual cooperation. This result overturned several decades of consensus about the IPD. In laboratory experiments in which human subjects were paired with computer co-players (extortionate ZD strategists), we demonstrate that the extortionate ZD strategies indeed enforce a unilateral control of the reward. When the experimental setting is sufficiently long and the computerized nature of the opponent is known to human subjects, the extortionate ZD strategy outperforms --- significantly more extortionate strategists finally obtain an average score higher than that from mutual cooperation. At the end of the talk, we will discuss the potential existence of such extortionate strategists in the real life and the social network, the evolutionary processes of human when facing the extorters in experiments, and how to model the processes of the human learning to be extorted in the deep learning view.

    Press, W. H. and Dyson, F. J. Iterated prisoner's dilemma contains strategies that dominate any evolutionary opponent. Proc. Natl Acad. Sci. USA 109, 10409--10413 (2012).
    Hilbe, C. et al. Extortion subdues human players but is finally punished in the prisoner's dilemma. Nature Communications, 5, 3976 (2014).
    Wang, Z. et al. Extortion can outperform generosity in the iterated Prisoner's Dilemma. Nature Communications, 7, 11125 (2016)

  • Dr. Chuang Wang (Harvard University, USA)
    13 February 2017, Monday: 10:30--11:30
    Online learning for high dimensional data processing: Exact dynamics and phase transitions

    We study the dynamics of an online algorithm for learning a sparse leading eigenvector from samples generated from a spiked covariance model. This algorithm combines the classical Oja's method for online principal component analysis with an element-wise nonlinearity at each iteration to promote sparsity. In the high-dimensional limit, the joint empirical measure of the underlying sparse eigenvector and its estimate provided by the algorithm is shown to converge weakly to a deterministic, measure-valued process. This scaling limit is characterized as the unique solution of a nonlinear PDE, and it provides exact information regarding the asymptotic performance of the algorithm. For example, performance metrics such as the cosine similarity and the misclassification rate in sparse support recovery can be obtained by examining the limiting dynamics. A steady-state analysis of the nonlinear PDE also reveals an interesting phase transition phenomenon. Although our analysis is asymptotic in nature, numerical simulations show that the theoretical predictions are accurate for moderate signal dimensions. Moreover, such analysis framework can be applied to more complicated situations, for example, low-rank subspace tracking problem using partially observations. Similar PDEs/ODEs and phase transition phenomenon are observed.

  • Dr. Xiangyi Li (University of Zurich, Swiss)
    16 January 2017, Monday: 14:30--15:30
    Environmental fluctuations and nature's sexy responses

    Species that cannot cope with the fluctuations of environments face exaggerated risks of extinction. Bet-hedging is one of the general strategies that can help species survive the unpredictable changes of the environment. Bet-hedgers spread their risk in reproduction by "not putting all eggs in one basket" or by preferring "one bird in the hand" over "two birds in the forest". By hedging one's bets against unpredictable environmental conditions, bet-hedgers enjoy higher geometric mean fitness in the long term at the cost of reduced arithmetic mean fitness at each specific time point, compared to individuals that specifically adapted to one particular environment. Sexual reproduction can be considered as a form of bet-hedging. Species that reproduce sexually pay the cost of producing males (which do not contribute to the population growth directly), but gain the benefit of having diversified offspring. Another common way of surviving in fluctuating environments is by dispersal. Spreading one's offspring in space may increase the chance that at least some of them survive and contribute to future generations. In sexually reproducing species, male and female offspring often have different dispersal probabilities, and their distance distributions of dispersal are also very often different. We build mathematical and simulation models to study the complex strategies produced by natural selection that help sexual species survive in unpredictable environments.

  • Prof. Erik Aurell (Royal Institute of Technology KTH, Stockholm)
    20 October 2016, Thursday: 10:30--11:30
    Progress on continuous-time dynamic cavity method

    Computing the marginal probabilities of a Gibbs-Boltzmann probability distribution has many applications in Physics and outside Physics. Efficient approximate methods have been developed over several decades to achieve this task if the graph of interactions has a locally tree-like form, as it has for random graphs. In Physics these methods are called the cavity method or the Bethe-Peierls method; in computer science they are called Belief Propagation and in Information Theory they are a part of iterative decoding.
    It is an open issue whether similar methods could also be competitive to determine the marginals of the probability distributions of non-equilibrium Physics, e.g. for the dynamics on random graphs. The case of exclusively one-way interactions (fully assymmetric couplings) was studied by Derrida and co-workers already more than 20 years ago, but is obviously special. We and others have over the last five years studied a version of this problem called the dynamic cavity and showed that it works well for the discrete-time (parallel update) kinetic Ising model on a random graph with partly symmetric and partly assymmetric, though, sufficiently weak, couplings (high-temperature regime).
    In this talk I will describe recent progress on the continuous-time version of this problem. The talk is based on: Erik Aurell, Gino Del Ferraro, Eduardo Dominguez, Roberto Mulet, ``A Cavity Master Equation for the continuous time dynamics of discrete spins models" [arXiv:1607.06959] (2016)

  • 11 October 2016, Tuesday: 08:55-12:00 (conference room 6420 of New Building)
    Mini-Workshop on Statistical Inference and Quantum Computing

    08:55 -- 09:00 Opening remarks

    09:00 -- 09:45 Dr. Tatsuro Kawamoto (AIST, Japan)
    Cross-validation estimate of the number of modules in modular networks

    It is a common and important task to extract modular structures from network data. Whereas many papers have been published in the field of computer science and statistical physics, there are many problems remain mystery, particularly for sparse networks. Here we focus on the model selection problem, i.e., the selection of the number of modules. We show that the leave-one-out cross-validation estimate of prediction errors are good measures to determine the number of modules and can be efficiently calculated using belief propagation.

    09:50 -- 10:35 Dr. Yingying Xu (Aalto University, Finland)
    Statistical inference of genome-wide epistasis in bacteria and general discussions

    Potts model in statistical physics has extended its application to biological data in last decay. After great success in prediction of protein structures, Direct Coupling Analysis is tried on genome-wide epistasis inference in bacteria in our research. We will show the results on the major human pathogen Streptococcus pneumonia. However, there are still many interesting open questions on the methodology part, such as how to determine the significant coupling threshold. In the talk, I will present the questions as well and let's have a look together.

    11:00 -- 11:50 Prof. Erik Aurell (KTH, Sweden)
    A global view of quantum computation with noisy components

    This talk is an attempt to estimate the error made in a general quantum computation by the Feynman-Vernon method. I will show how some simple estimates can be obtained for idealized systems, and why such estimates are more difficult to obtain for more advanced schemes such as surface codes. The talk is mainly based on arXiv:1606.09407.

  • Dr. Stefano Bo (Nordic Institute for Theoretical Physics, Stockholm)
    4 October 2016, Tuesday: 10:30--11:30
    Multiple-scale stochastic processes: decimation, averaging and beyond

    The recent experimental progresses in handling microscopic systems have allowed to probe them at levels where fluctuations are prominent, calling for stochastic modeling in a large number of physical, chemical and biological phenomena. Many systems of interest often involve processes taking place on widely separated time scales. For an efficient modeling one usually focuses on the slower degrees of freedom and it is of great importance to accurately eliminate the fast variables in a controlled fashion, carefully accounting for their net effect on the slower dynamics. This procedure in general requires to perform two different operations: decimation and coarse-graining. I will present such procedures and discuss their application to a series of physical, biological and chemical examples. I will then consider functionals of the stochastic trajectories such as residence times, counting statistics, fluxes, entropy production, etc. which have been increasingly studied in recent years. For such functionals, the elimination of the fast degrees of freedom can present additional difficulties and naive procedures can lead to blatantly inconsistent results. Homogenization techniques for functionals are less covered in the literature and I will present them here as natural extensions of the ones employed for the trajectories.

  • Prof. Da-Qing Li (Beihang University)
    22 September 2016, Thursday: 10:30--11:30
    Resilient Infrastructure

    As the lifeline of human society, infrastructures including information networks, transportation networks as well as power grid networks are facing unprecedented challenge on their reliability from various threats. Starting from a localized failure, a cascading failure can propagate with a domino-like way, which may result in catastrophe such as those observed in many realistic networks. For instance, in March 2015, over half number of provinces of Turkey witnessed cascading blackout, which caused tremendous and inevitable damage in transport, air service as well as residential electricity consumption. Similar cascading failures also happen in information system. In Jan 2014, starting from the crash of servers of Tencent, over two-thirds of DNS servers in China resulted in collapse, which caused a major breakdown of web pages for many users. These cascading failures are mostly inconspicuous at their early stage, but wide spreading in an unexpected way, making it difficult to predict and mitigate. In this talk, I will introduce our recent efforts to understand and design a resilient infrastructure.

  • Mr. Qing Zhang (Universite Pierre et Marie Curie, Paris)
    13 September 2016, Tuesday: 10:30--11:30
    Stochasticity and robustness in S-phase duration from genome replication kinetics

    Genome replication, a key physiological process for a living cell, typically relies on intrinsically stochastic initiation by replication origins, causing a variability of replication timing from cell to cell. Over evolution, an organism can control only the propensity of origins to initiate and their position, but it does not eliminate completely this uncertainty. While widely accepted mathematical models of eukaryotic replication as a stochastic process are available, the question of the link between the controllable parameters and the resulting distribution of global replication timing has not been addressed systematically.
    Here, we propose a combined analytical and computational approach to this question. Our calculations give a simple way to understand how positions and strengths of many origins lead to a given distribution of total duration of the replication of a large region, a chromosome or the entire genome. Specifically, the total replication timing can be framed as an extreme-value problem, since it is due to the last region that replicates in each cell. Our calculations lead us to identify two regimes based on the spread between characteristic completion times of all inter-origin regions of a genome. For widely different completion times, timing is set by the single specific region that is typically the last to replicate in all cells (and is hence "fragile"). Conversely, when the completion times of all regions are comparable, an extreme-value estimate shows that the cell-to-cell variability of genome replication timing has universal properties. Comparison with available data shows that the replication program of two yeast species falls in this extreme-value regime.

  • Dr. Shun Kurokawa (Institute of Zoology, Chinese Academy of Sciences, Beijing)
    9 September 2016, Friday: 10:30--11:30
    Evolution of Stubbornness

    The existence of cooperation demands explanation since cooperation is costly to the actor while cooperation is beneficial to the receiver. In repeated interactions, when players adopt the strategy "If you cooperate, then I cooperate. If you defect, then I defect.", the evolution of cooperation is possible. And this is a major mechanism which is called direct reciprocity. Direct reciprocity is based on the assumption that players can use the information about the opponent's behavior; however, the information about the opponent's behavior is sometimes imperfect. And it is not obvious what behavior is most likely to evolve in the case wherein information about the opponent is absent.
    Here, I consider the game in which there are two strategies; one is conditional cooperators and the other is unconditional defectors in a repeated prisoner's dilemma. And by analyzing evolutionarily stable strategy (ESS) analysis, I have revealed that the strategy with stubbornness which behaves "If I cooperate in the last move, then I cooperate also in the next round. If I defect in the last move, then I defect also in the next round." in the case wherein information about the opponent's behavior is absent, has the loosest ESS condition. This indicates that the evolution of stubbornness is favored by natural selection. I also would like to discuss the relationship between this study and studies on win-stay, lose-shift (i.e., Wang, Xu, & Zhou, 2014).