Our goal is to train a policy for autonomous driving via imitation learning that is robust enough to drive a real vehicle. We find that standard behavior cloning is insufficient for handling complex driving scenarios, even when we leverage a perception system for preprocessing the input and a controller for executing the output on the car: 30 million examples are still not enough. We propose exposing the learner to synthesized data in the form of perturbations to the expert’s driving, which creates interesting situations such as collisions and/or going off the road. Rather than purely imitating all data, we augment the imitation loss with additional losses that penalize undesirable events and encourage progress — the perturbations then provide an important signal for these losses and lead to robustness of the learned model. We show that the ChauffeurNet model can handle complex situations in simulation, and present ablation experiments that emphasize the importance of each of our proposed changes and show that the model is responding to the appropriate causal factors. Finally, we demonstrate the model driving a car in the real world.

ChauffeurNet: Learning to Drive by Imitating the Best and Synthesizing the Worst
Mayank Bansal, Alex Krizhevsky, Abhijit Ogale

Source: arxiv.org

Activity-driven networks (ADNs) have recently emerged as a powerful paradigm to study the temporal evolution of stochastic networked systems. All the information on the time-varying nature of the system is encapsulated into a constant activity parameter, which represents the propensity to generate connections. This formulation has enabled the scientific community to perform effective analytical studies on temporal networks. However, the hypothesis that the whole dynamics of the system is summarized by constant parameters might be excessively restrictive. Empirical studies suggest that activity evolves in time, intertwined with the system evolution, causing burstiness and clustering phenomena. In this paper, we propose a novel model for temporal networks, in which a self-excitement mechanism governs the temporal evolution of the activity, linking it to the evolution of the networked system. We investigate the effect of self-excitement on the epidemic inception by comparing the epidemic threshold of a Susceptible–Infected–Susceptible model in the presence and in the absence of the self-excitement mechanism. Our results suggest that the temporal nature of the activity favors the epidemic inception. Hence, neglecting self-excitement mechanisms might lead to harmful underestimation of the risk of an epidemic outbreak. Extensive numerical simulations are presented to support and extend our analysis, exploring parameter heterogeneities and noise, transient dynamics, and immunization processes. Our results constitute a first, necessary step toward a theory of ADNs that accounts for memory effects in the network evolution.

Modeling Memory Effects in Activity-Driven Networks
Lorenzo Zino, Alessandro Rizzo, and Maurizio Porfiri

SIAM J. Appl. Dyn. Syst., 17(4), 2830–2854. (25 pages)

Source: epubs.siam.org

Experience plays a critical role in crafting high-impact scientific work. This is particularly evident in top multidisciplinary journals, where a scientist is unlikely to appear as senior author if he or she has not previously published within the same journal. Here, we develop a quantitative understanding of author order by quantifying this “chaperone effect,” capturing how scientists transition into senior status within a particular publication venue. We illustrate that the chaperone effect has a different magnitude for journals in different branches of science, being more pronounced in medical and biological sciences and weaker in natural sciences. Finally, we show that in the case of high-impact venues, the chaperone effect has significant implications, specifically resulting in a higher average impact relative to papers authored by new principal investigators (PIs). Our findings shed light on the role played by experience in publishing within specific scientific journals, on the paths toward acquiring the necessary experience and expertise, and on the skills required to publish in prestigious venues.

The chaperone effect in scientific publishing
Vedran Sekara, Pierre Deville, Sebastian E. Ahnert, Albert-László Barabási, Roberta Sinatra, and Sune Lehmann
PNAS published ahead of print December 10, 2018 https://doi.org/10.1073/pnas.1800471115

Source: www.pnas.org

The Multiscale Law of Requisite Variety is a scientific law relating, at each scale, the variation in an environment to the variation in internal state that is necessary for effective response by a system. While this law has been used to describe the effectiveness of systems in self-regulation, the consequences for failure have not been formalized. Here we use this law to consider the internal dynamics of an unstructured system, and its response to a structured environment. We find that, due to its inability to respond, a completely unstructured system is inherently unstable to the formation of structure. And in general, any system without structure above a certain scale is unable to withstand structure arising above that scale. To describe complicated internal dynamics, we develop a characterization of multiscale changes in a system. This characterization is motivated by Shannon information theoretic ideas of noise, but considers structured information. We then relate our findings to political anarchism showing that society requires some organizing processes, even if there is no traditional government or hierarchies. We also formulate our findings as an inverse second law of thermodynamics; while closed systems collapse into disorder, systems open to a structured environment spontaneously generate order.

Taeer Bar-Yam, Owen Lynch, Yaneer Bar-Yam, The inherent instability of disordered systems, arXiv:1812.00450

Source: necsi.edu