Santa Fe Institute

About: Santa Fe Institute is a nonprofit organization based out in Santa Fe, New Mexico, United States. It is known for research contribution in the topics: Population & Context (language use). The organization has 558 authors who have published 4558 publications receiving 396015 citations. The organization is also known as: SFI.

The existence of a priori distinctions between learning algorithms

Abstract: This is the second of two papers that use off-training set (OTS) error to investigate the assumption-free relationship between learning algorithms. The first paper discusses a particular set of ways to compare learning algorithms, according to which there are no distinctions between learning algorithms. This second paper concentrates on different ways of comparing learning algorithms from those used in the first paper. In particular this second paper discusses the associated a priori distinctions that do exist between learning algorithms. In this second paper it is shown, loosely speaking, that for loss functions other than zero-one (e.g., quadratic loss), there are a priori distinctions between algorithms. However, even for such loss functions, it is shown here that any algorithm is equivalent on average to its “randomized” version, and in this still has no first principles justification in terms of average error. Nonetheless, as this paper discusses, it may be that (for example) cross-validation has better head-to-head minimax properties than “anti-cross-validation” (choose the learning algorithm with the largest cross-validation error). This may be true even for zero-one loss, a loss function for which the notion of “randomization” would not be relevant. This paper also analyzes averages over hypotheses rather than targets. Such analyses hold for all possible priors over targets. Accordingly they prove, as a particular example, that cross-validation cannot be justified as a Bayesian procedure. In fact, for a very natural restriction of the class of learning algorithms, one should use anti-cross-validation rather than cross-validation (!).

A Theory of State Space Reconstruction in the Presence of Noise

Abstract: Takens’ theorem demonstrates that in the absence of noise a multidimensional state space can be reconstructed from a single time series. This theorem does not treat the effect of noise, however, and so gives no guidance about practical considerations for reconstructing a good state space. We study the problem of reconstructing a state space with observational noise, examining the likelihood for a particular state given a series of noisy observations. We define a quantity called the distortion, which is proportional to the covariance of the likelihood function. This is related to the noise amplification, which corresponds to the root-mean-square errors for time series prediction with an ideal model. We prove that in the low noise limit minimizing the distortion is equivalent to minimizing the noise amplification.

Limitations in predicting the space radiation health risk for exploration astronauts.

Abstract: Despite years of research, understanding of the space radiation environment and the risk it poses to long-duration astronauts remains limited. There is a disparity between research results and observed empirical effects seen in human astronaut crews, likely due to the numerous factors that limit terrestrial simulation of the complex space environment and extrapolation of human clinical consequences from varied animal models. Given the intended future of human spaceflight, with efforts now to rapidly expand capabilities for human missions to the moon and Mars, there is a pressing need to improve upon the understanding of the space radiation risk, predict likely clinical outcomes of interplanetary radiation exposure, and develop appropriate and effective mitigation strategies for future missions. To achieve this goal, the space radiation and aerospace community must recognize the historical limitations of radiation research and how such limitations could be addressed in future research endeavors. We have sought to highlight the numerous factors that limit understanding of the risk of space radiation for human crews and to identify ways in which these limitations could be addressed for improved understanding and appropriate risk posture regarding future human spaceflight.

Gene-culture coevolution and the nature of human sociality.

Abstract: Human characteristics are the product of gene–culture coevolution, which is an evolutionary dynamic involving the interaction of genes and culture over long time periods. Gene–culture coevolution is a special case of niche construction. Gene–culture coevolution is responsible for human other-regarding preferences, a taste for fairness, the capacity to empathize and salience of morality and character virtues.