Various algorithms for optimization and learning in adaptive systems

This thesis discusses various methods for learning and optimization in adaptive systems. Overall, it emphasizes the relationship between optimization, learning, and adaptive systems; and it illustrates the influence of underlying hardware upon the construction of efficient algorithms for learning and optimization. Chapter 1 provides a summary and an overview.

Chapter 2 discusses a method for using feed-forward neural networks to filter the noise out of noise-corrupted signals. The networks use back-propagation learning, but they use it in a way that qualifies as unsupervised learning. The networks adapt based only on the raw input data-there are no external teachers providing information on correct operation during training. The chapter contains an analysis of the learning and develops a simple expression that, based only on the geometry of the network, predicts performance.

Chapter 3 explains a simple model of the piriform cortex, an area in the brain involved in the processing of olfactory information. The model was used to explore the possible effect of acetylcholine on learning and on odor classification. According to the model, the piriform cortex can classify odors better when acetylcholine is present during learning but not present during recall. This is interesting since it suggests that learning and recall might be separate neurochemical modes (corresponding to whether or not acetylcholine is present). When acetylcholine is turned off at all times, even during learning, the model exhibits behavior somewhat similar to Alzheimer's disease, a disease associated with the degeneration of cells that distribute acetylcholine.

Chapters 4, 5, and 6 discuss algorithms appropriate for adaptive systems implemented entirely in analog hardware. The algorithms inject noise into the systems and correlate the noise with the outputs of the systems. This allows them to estimate gradients and to implement noisy versions of gradient descent, without having to calculate gradients explicitly. The methods require only noise generators, adders, multipliers, integrators, and differentiators; and the number of devices needed scales linearly with the number of adjustable parameters in the adaptive systems. With the exception of one global signal, the algorithms require only local information exchange.

Efficient methods for empirical tests of behavioral economics theories in laboratory and field experiments

In the quest for a descriptive theory of decision-making, the rational actor model in economics imposes rather unrealistic expectations and abilities on human decision makers. The further we move from idealized scenarios, such as perfectly competitive markets, and ambitiously extend the reach of the theory to describe everyday decision making situations, the less sense these assumptions make. Behavioural economics has instead proposed models based on assumptions that are more psychologically realistic, with the aim of gaining more precision and descriptive power. Increased psychological realism, however, comes at the cost of a greater number of parameters and model complexity. Now there are a plethora of models, based on different assumptions, applicable in differing contextual settings, and selecting the right model to use tends to be an ad-hoc process. In this thesis, we develop optimal experimental design methods and evaluate different behavioral theories against evidence from lab and field experiments.

We look at evidence from controlled laboratory experiments. Subjects are presented with choices between monetary gambles or lotteries. Different decision-making theories evaluate the choices differently and would make distinct predictions about the subjects' choices. Theories whose predictions are inconsistent with the actual choices can be systematically eliminated. Behavioural theories can have multiple parameters requiring complex experimental designs with a very large number of possible choice tests. This imposes computational and economic constraints on using classical experimental design methods. We develop a methodology of adaptive tests: Bayesian Rapid Optimal Adaptive Designs (BROAD) that sequentially chooses the "most informative" test at each stage, and based on the response updates its posterior beliefs over the theories, which informs the next most informative test to run. BROAD utilizes the Equivalent Class Edge Cutting (EC²) criteria to select tests. We prove that the EC² criteria is adaptively submodular, which allows us to prove theoretical guarantees against the Bayes-optimal testing sequence even in the presence of noisy responses. In simulated ground-truth experiments, we find that the EC² criteria recovers the true hypotheses with significantly fewer tests than more widely used criteria such as Information Gain and Generalized Binary Search. We show, theoretically as well as experimentally, that surprisingly these popular criteria can perform poorly in the presence of noise, or subject errors. Furthermore, we use the adaptive submodular property of EC² to implement an accelerated greedy version of BROAD which leads to orders of magnitude speedup over other methods.

We use BROAD to perform two experiments. First, we compare the main classes of theories for decision-making under risk, namely: expected value, prospect theory, constant relative risk aversion (CRRA) and moments models. Subjects are given an initial endowment, and sequentially presented choices between two lotteries, with the possibility of losses. The lotteries are selected using BROAD, and 57 subjects from Caltech and UCLA are incentivized by randomly realizing one of the lotteries chosen. Aggregate posterior probabilities over the theories show limited evidence in favour of CRRA and moments' models. Classifying the subjects into types showed that most subjects are described by prospect theory, followed by expected value. Adaptive experimental design raises the possibility that subjects could engage in strategic manipulation, i.e. subjects could mask their true preferences and choose differently in order to obtain more favourable tests in later rounds thereby increasing their payoffs. We pay close attention to this problem; strategic manipulation is ruled out since it is infeasible in practice, and also since we do not find any signatures of it in our data.

In the second experiment, we compare the main theories of time preference: exponential discounting, hyperbolic discounting, "present bias" models: quasi-hyperbolic (α, β) discounting and fixed cost discounting, and generalized-hyperbolic discounting. 40 subjects from UCLA were given choices between 2 options: a smaller but more immediate payoff versus a larger but later payoff. We found very limited evidence for present bias models and hyperbolic discounting, and most subjects were classified as generalized hyperbolic discounting types, followed by exponential discounting.

In these models the passage of time is linear. We instead consider a psychological model where the perception of time is subjective. We prove that when the biological (subjective) time is positively dependent, it gives rise to hyperbolic discounting and temporal choice inconsistency.

We also test the predictions of behavioral theories in the "wild". We pay attention to prospect theory, which emerged as the dominant theory in our lab experiments of risky choice. Loss aversion and reference dependence predicts that consumers will behave in a uniquely distinct way than the standard rational model predicts. Specifically, loss aversion predicts that when an item is being offered at a discount, the demand for it will be greater than that explained by its price elasticity. Even more importantly, when the item is no longer discounted, demand for its close substitute would increase excessively. We tested this prediction using a discrete choice model with loss-averse utility function on data from a large eCommerce retailer. Not only did we identify loss aversion, but we also found that the effect decreased with consumers' experience. We outline the policy implications that consumer loss aversion entails, and strategies for competitive pricing.

In future work, BROAD can be widely applicable for testing different behavioural models, e.g. in social preference and game theory, and in different contextual settings. Additional measurements beyond choice data, including biological measurements such as skin conductance, can be used to more rapidly eliminate hypothesis and speed up model comparison. Discrete choice models also provide a framework for testing behavioural models with field data, and encourage combined lab-field experiments.

Solar photospheric and coronal abundances from solar energetic particle measurements

Observations of solar energetic particles (SEPs) from 22 solar flares in the 1977-1982 time period are reported. The observations were made by the Cosmic Ray Subsystem on board the Voyager 1 and 2 spacecraft. SEP abundances have been obtained for all elements with 3 ≤ Z ≤ 30 except Li, Be, B. F, Sc, V, Co and Cu. for which upper limits have been obtained. Statistically meaningful abundances of several rare elements (e.g., P, Cl, K, Ti, Mn) have been determined for the first time, and the average abundances of the more abundant elements have been determined with improved precision, typically a factor of three better than the best previous determinations.

Previously reported results concerning the dependence of the fractionation of SEPs relative to photosphere on first ionization potential (FIP) have been confirmed and amplified upon with the new data. The monotonic Z-dependence of the variation between flares noted by earlier studies was found to be interpretable as a fractionation, produced by acceleration of the particles from the corona and their propagation through interplanetary space, which is ordered by the ionic charge-to-mass ratio Q/ M of the species making up the SEPs. It was found that Q/M is the primary organizing parameter of acceleration and propagation effects in SEPs, as evidenced by the dependence on Q/M of time, spatial and energy dependence within flares and of the abundance variability from flare to flare.

An unfractionated coronal composition was derived by applying a simple Q/M fractionation correction to the observed average SEP composition, to simultaneously correct for all Q/M-correlated acceleration/propagation fractionation of SEPs. The resulting coronal composition agrees well with current XUV/X-ray spectroscopic measurements of coronal composition but is of much higher precision and is available for a much larger set of elements. Compared to spectroscopic photospheric abundances, the SEP-derived corona appears depleted in C and somewhat enriched in Cr (and possibly Ca and Ti).

An unfractionated photospheric composition was derived by applying a simple FIP fractionation correction to the derived coronal composition, to correct for the FIP-associated fractionation of the corona during its formation from photospheric material. The resulting composition agrees well with the photospheric abundance tabulation of Grevesse (1984) except for an at least 50% lower abundance of C and a significantly greater abundance of Cr and possibly Ti. The results support the Grevesse photospheric Fe abundance, about 50% higher than meteoritic and earlier solar values. The SEP-derived photospheric composition is not generally of higher precision than the available spectroscopic data, but it relies on fewer physical parameters and is available for some elements (C, N, Ne, Ar) which cannot be measured spectroscopically in the photosphere.

Site-Specific Isotopes in Small Organic Molecules

Stable isotope geochemistry is a valuable toolkit for addressing a broad range of problems in the geosciences. Recent technical advances provide information that was previously unattainable or provide unprecedented precision and accuracy. Two such techniques are site-specific stable isotope mass spectrometry and clumped isotope thermometry. In this thesis, I use site-specific isotope and clumped isotope data to explore natural gas development and carbonate reaction kinetics. In the first chapter, I develop an equilibrium thermodynamics model to calculate equilibrium constants for isotope exchange reactions in small organic molecules. This equilibrium data provides a framework for interpreting the more complex data in the later chapters. In the second chapter, I demonstrate a method for measuring site-specific carbon isotopes in propane using high-resolution gas source mass spectrometry. This method relies on the characteristic fragments created during electron ionization, in which I measure the relative isotopic enrichment of separate parts of the molecule. My technique will be applied to a range of organic compounds in the future. For the third chapter, I use this technique to explore diffusion, mixing, and other natural processes in natural gas basins. As time progresses and the mixture matures, different components like kerogen and oil contribute to the propane in a natural gas sample. Each component imparts a distinct fingerprint on the site-specific isotope distribution within propane that I can observe to understand the source composition and maturation of the basin. Finally, in Chapter Four, I study the reaction kinetics of clumped isotopes in aragonite. Despite its frequent use as a clumped isotope thermometer, the aragonite blocking temperature is not known. Using laboratory heating experiments, I determine that the aragonite clumped isotope thermometer has a blocking temperature of 50-100°C. I compare this result to natural samples from the San Juan Islands that exhibit a maximum clumped isotope temperature that matches this blocking temperature. This thesis presents a framework for measuring site-specific carbon isotopes in organic molecules and new constraints on aragonite reaction kinetics. This study represents the foundation of a future generation of geochemical tools for the study of complex geologic systems.

Understanding planet formation through high precision photometry and spectroscopy

From studies of protoplanetary disks to extrasolar planets and planetary debris, we aim to understand the full evolution of a planetary system. Observational constraints from ground- and space-based instrumentation allows us to measure the properties of objects near and far and are central to developing this understanding. We present here three observational campaigns that, when combined with theoretical models, reveal characteristics of different stages and remnants of planet formation. The Kuiper Belt provides evidence of chemical and dynamical activity that reveals clues to its primordial environment and subsequent evolution. Large samples of this population can only be assembled at optical wavelengths, with thermal measurements at infrared and sub-mm wavelengths currently available for only the largest and closest bodies. We measure the size and shape of one particular object precisely here, in hopes of better understanding its unique dynamical history and layered composition.

Molecular organic chemistry is one of the most fundamental and widespread facets of the universe, and plays a key role in planet formation. A host of carbon-containing molecules vibrationally emit in the near-infrared when excited by warm gas, T~1000 K. The NIRSPEC instrument at the W.M. Keck Observatory is uniquely configured to study large ranges of this wavelength region at high spectral resolution. Using this facility we present studies of warm CO gas in protoplanetary disks, with a new code for precise excitation modeling. A parameterized suite of models demonstrates the abilities of the code and matches observational constraints such as line strength and shape. We use the models to probe various disk parameters as well, which are easily extensible to others with known disk emission spectra such as water, carbon dioxide, acetylene, and hydrogen cyanide.

Lastly, the existence of molecules in extrasolar planets can also be studied with NIRSPEC and reveals a great deal about the evolution of the protoplanetary gas. The species we observe in protoplanetary disks are also often present in exoplanet atmospheres, and are abundant in Earth's atmosphere as well. Thus, a sophisticated telluric removal code is necessary to analyze these high dynamic range, high-resolution spectra. We present observations of a hot Jupiter, revealing water in its atmosphere and demonstrating a new technique for exoplanet mass determination and atmospheric characterization. We will also be applying this atmospheric removal code to the aforementioned disk observations, to improve our data analysis and probe less abundant species. Guiding models using observations is the only way to develop an accurate understanding of the timescales and processes involved. The futures of the modeling and of the observations are bright, and the end goal of realizing a unified model of planet formation will require both theory and data, from a diverse collection of sources.

Part I - The radiation of elastic waves from a spherical cavity in a half space. Part II - Precision determination of focal depths and epicenters of earthquakes.

Part I: The dynamic response of an elastic half space to an explosion in a buried spherical cavity is investigated by two methods. The first is implicit, and the final expressions for the displacements at the free surface are given as a series of spherical wave functions whose coefficients are solutions of an infinite set of linear equations. The second method is based on Schwarz's technique to solve boundary value problems, and leads to an iterative solution, starting with the known expression for the point source in a half space as first term. The iterative series is transformed into a system of two integral equations, and into an equivalent set of linear equations. In this way, a dual interpretation of the physical phenomena is achieved. The systems are treated numerically and the Rayleigh wave part of the displacements is given in the frequency domain. Several comparisons with simpler cases are analyzed to show the effect of the cavity radius-depth ratio on the spectra of the displacements.

Part II: A high speed, large capacity, hypocenter location program has been written for an IBM 7094 computer. Important modifications to the standard method of least squares have been incorporated in it. Among them are a new way to obtain the depth of shocks from the normal equations, and the computation of variable travel times for the local shocks in order to account automatically for crustal variations. The multiregional travel times, largely based upon the investigations of the United States Geological Survey, are confronted with actual traverses to test their validity.

It is shown that several crustal phases provide control enough to obtain good solutions in depth for nuclear explosions, though not all the recording stations are in the region where crustal corrections are considered. The use of the European travel times, to locate the French nuclear explosion of May 1962 in the Sahara, proved to be more adequate than previous work.

A simpler program, with manual crustal corrections, is used to process the Kern County series of aftershocks, and a clearer picture of tectonic mechanism of the White Wolf fault is obtained.

Shocks in the California region are processed automatically and statistical frequency-depth and energy depth curves are discussed in relation to the tectonics of the area.