Semi-supervised learning of statistical models for natural language understanding

Natural language understanding is to specify a computational model that maps sentences to their semantic mean representation. In this paper, we propose a novel framework to train the statistical models without using expensive fully annotated data. In particular, the input of our framework is a set of sentences labeled with abstract semantic annotations. These annotations encode the underlying embedded semantic structural relations without explicit word/semantic tag alignment. The proposed framework can automatically induce derivation rules that map sentences to their semantic meaning representations. The learning framework is applied on two statistical models, the conditional random fields (CRFs) and the hidden Markov support vector machines (HM-SVMs). Our experimental results on the DARPA communicator data show that both CRFs and HM-SVMs outperform the baseline approach, previously proposed hidden vector state (HVS) model which is also trained on abstract semantic annotations. In addition, the proposed framework shows superior performance than two other baseline approaches, a hybrid framework combining HVS and HM-SVMs and discriminative training of HVS, with a relative error reduction rate of about 25% and 15% being achieved in F-measure.

Hyper-random Phenomena: Definition and Description

The paper is dedicated to the theory which describes physical phenomena in non-constant statistical conditions. The theory is a new direction in probability theory and mathematical statistics that gives new possibilities for presentation of physical world by hyper-random models. These models take into consideration the changing of object’s properties, as well as uncertainty of statistical conditions.

Reliability for Beta Models

In the area of stress-strength models there has been a large amount of work as regards estimation of the reliability R = Pr(X2 < X1 ) when X1 and X2 are independent random variables belonging to the same univariate family of distributions. The algebraic form for R = Pr(X2 < X1 ) has been worked out for the majority of the well-known distributions including Normal, uniform, exponential, gamma, weibull and pareto. However, there are still many other distributions for which the form of R is not known. We have identified at least some 30 distributions with no known form for R. In this paper we consider some of these distributions and derive the corresponding forms for the reliability R. The calculations involve the use of various special functions.

A New Class of Processes for Formalizing and Generalizing Individual-Based Models: The Semi-Semi-Markov Processes

2000 Mathematics Subject Classification: 60K15, 60K20, 60G20,60J75, 60J80, 60J85, 60-08, 90B15.

Analysing Conjoint Analysis Data by a Random Coefficient Regression Model

2000 Mathematics Subject Classification: 62J12, 62K15, 91B42, 62H99.

Comparison of Multivariate and Univariate Models for Genetic Evaluation of Milk Yield based on Test Day Data

2000 Mathematics Subject Classification: 62H12, 62P99

Spectral width optimization in random DFB fiber laser

Lasers with random distributed feedback (DFB) owing to Rayleigh scattering in optical fibers [1] have attracted a great interest: a number of papers demonstrating new laser schemes and applications have been proposed [2-7] recently. Moreover, the generation output power and, more generally, generation power distribution could be described both analytically and numerically within simple balance models [8-9]. However, spectral properties of random DFB fiber lasers are not studied except some attempt made in [10]. Generation spectrum of random DFB fiber laser is quite broad (more than 1 nm), and physical mechanisms of its formation and broadening are still unclear. There is no any practical solution up to date to minimize the generation spectrum width. Here we experimentally show the way to minimize the generation spectral width. © 2013 IEEE.

Latent Space and Social Psychological Models of Diffusion

The problem of social diffusion has animated sociological thinking on topics ranging from the spread of an idea, an innovation or a disease, to the foundations of collective behavior and political polarization. While network diffusion has been a productive metaphor, the reality of diffusion processes is often muddier. Ideas and innovations diffuse differently from diseases, but, with a few exceptions, the diffusion of ideas and innovations has been modeled under the same assumptions as the diffusion of disease. In this dissertation, I develop two new diffusion models for "socially meaningful" contagions that address two of the most significant problems with current diffusion models: (1) that contagions can only spread along observed ties, and (2) that contagions do not change as they spread between people. I augment insights from these statistical and simulation models with an analysis of an empirical case of diffusion - the use of enterprise collaboration software in a large technology company. I focus the empirical study on when people abandon innovations, a crucial, and understudied aspect of the diffusion of innovations. Using timestamped posts, I analyze when people abandon software to a high degree of detail.

To address the first problem, I suggest a latent space diffusion model. Rather than treating ties as stable conduits for information, the latent space diffusion model treats ties as random draws from an underlying social space, and simulates diffusion over the social space. Theoretically, the social space model integrates both actor ties and attributes simultaneously in a single social plane, while incorporating schemas into diffusion processes gives an explicit form to the reciprocal influences that cognition and social environment have on each other. Practically, the latent space diffusion model produces statistically consistent diffusion estimates where using the network alone does not, and the diffusion with schemas model shows that introducing some cognitive processing into diffusion processes changes the rate and ultimate distribution of the spreading information. To address the second problem, I suggest a diffusion model with schemas. Rather than treating information as though it is spread without changes, the schema diffusion model allows people to modify information they receive to fit an underlying mental model of the information before they pass the information to others. Combining the latent space models with a schema notion for actors improves our models for social diffusion both theoretically and practically.

The empirical case study focuses on how the changing value of an innovation, introduced by the innovations' network externalities, influences when people abandon the innovation. In it, I find that people are least likely to abandon an innovation when other people in their neighborhood currently use the software as well. The effect is particularly pronounced for supervisors' current use and number of supervisory team members who currently use the software. This case study not only points to an important process in the diffusion of innovation, but also suggests a new approach -- computerized collaboration systems -- to collecting and analyzing data on organizational processes.

Testing Wavefunction Collapse Models using Parametric Heating of a Trapped Nanosphere

We propose a mechanism for testing the theory of collapse models such as continuous spontaneous localization (CSL) by examining the parametric heating rate of a trapped nanosphere. The random localizations of the center-of-mass for a given particle predicted by the CSL model can be understood as a stochastic force embodying a source of heating for the nanosphere. We show that by utilising a Paul trap to levitate the particle and optical cooling, it is possible to reduce environmental decoher- ence to such a level that CSL dominates the dynamics and contributes the main source of heating. We show that this approach allows measurements to be made on the timescale of seconds, and that the free parameter λcsl which characterises the model ought to be testable to values as low as 10^{−12} Hz.

Undersampled Critical Branching Processes on Small-World and Random Networks Fail to Reproduce the Statistics of Spike Avalanches

The power-law size distributions obtained experimentally for neuronal avalanches are an important evidence of criticality in the brain. This evidence is supported by the fact that a critical branching process exhibits the same exponent t~3=2. Models at criticality have been employed to mimic avalanche propagation and explain the statistics observed experimentally. However, a crucial aspect of neuronal recordings has been almost completely neglected in the models: undersampling. While in a typical multielectrode array hundreds of neurons are recorded, in the same area of neuronal tissue tens of thousands of neurons can be found. Here we investigate the consequences of undersampling in models with three different topologies (two-dimensional, small-world and random network) and three different dynamical regimes (subcritical, critical and supercritical). We found that undersampling modifies avalanche size distributions, extinguishing the power laws observed in critical systems. Distributions from subcritical systems are also modified, but the shape of the undersampled distributions is more similar to that of a fully sampled system. Undersampled supercritical systems can recover the general characteristics of the fully sampled version, provided that enough neurons are measured. Undersampling in two-dimensional and small-world networks leads to similar effects, while the random network is insensitive to sampling density due to the lack of a well-defined neighborhood. We conjecture that neuronal avalanches recorded from local field potentials avoid undersampling effects due to the nature of this signal, but the same does not hold for spike avalanches. We conclude that undersampled branching-process-like models in these topologies fail to reproduce the statistics of spike avalanches.

Essays on Machine Learning and Hedonic Models

Thesis (Ph.D.)--University of Washington, 2016-08

Does the abundance of hoverfly mimics (Syrphidae) depend on the numbers of their hymenopteran models ?

We tested the prediction that, if hoverflies are Batesian mimics, this may extend to behavioral mimicry such that their numerical abundance at each hour of the day (the daily activity pattern) is related to the numbers of their hymenopteran models. After accounting for site, season, microclimatic responses and for general hoverfly abundance at three sites in north-west England, the residual numbers of mimics were significantly correlated positively with their models 9 times out of 17, while 16 out of 17 relationships were positive, itself a highly significant non-random pattern. Several eristaline flies showed significant relationships with honeybees even though some of them mimic wasps or bumblebees, perhaps reflecting an ancestral resemblance to honeybees. There was no evidence that good and poor mimics differed in their daily activity pattern relationships with models. However, the common mimics showed significant activity pattern relationships with their models, but the rarer mimics did not. We conclude that many hoverflies show behavioral mimicry of their hymenopteran models.

The role of intracellular calcium perturbations in muscle damage and dysfunction in mouse models of muscular dystrophy

Duchenne muscular dystrophy (DMD) is a neuromuscular disease caused by mutations in the dystrophin gene. DMD is clinically characterized by severe, progressive and irreversible loss of muscle function, in which most patients lose the ability to walk by their early teens and die by their early 20’s. Impaired intracellular calcium (Ca2+) regulation and activation of cell degradation pathways have been proposed as key contributors to DMD disease progression. This dissertation research consists of three studies investigating the role of intracellular Ca2+ in skeletal muscle dysfunction in different mouse models of DMD. Study one evaluated the role of Ca2+-activated enzymes (proteases) that activate protein degradation in excitation-contraction (E-C) coupling failure following repeated contractions in mdx and dystrophin-utrophin null (mdx/utr-/-) mice. Single muscle fibers from mdx/utr-/- mice had greater E-C coupling failure following repeated contractions compared to fibers from mdx mice. Moreover, protease inhibition during these contractions was sufficient to attenuate E-C coupling failure in muscle fibers from both mdx and mdx/utr-/- mice. Study two evaluated the effects of overexpressing the Ca2+ buffering protein sarcoplasmic/endoplasmic reticulum Ca2+-ATPase 1 (SERCA1) in skeletal muscles from mdx and mdx/utr-/- mice. Overall, SERCA1 overexpression decreased muscle damage and protected the muscle from contraction-induced injury in mdx and mdx/utr-/- mice. In study three, the cellular mechanisms underlying the beneficial effects of SERCA1 overexpression in mdx and mdx/utr-/- mice were investigated. SERCA1 overexpression attenuated calpain activation in mdx muscle only, while partially attenuating the degradation of the calpain target desmin in mdx/utr-/- mice. Additionally, SERCA1 overexpression decreased the SERCA-inhibitory protein sarcolipin in mdx muscle but did not alter levels of Ca2+ regulatory proteins (parvalbumin and calsequestrin) in either dystrophic model. Lastly, SERCA1 overexpression blunted the increase in endoplasmic reticulum stress markers Grp78/BiP in mdx mice and C/EBP homologous protein (CHOP) in mdx and mdx/utr-/- mice. Overall, findings from the studies presented in this dissertation provide new insight into the role of Ca2+ in muscle dysfunction and damage in different dystrophic mouse models. Further, these findings support the overall strategy for improving intracellular Ca2+ control for the development of novel therapies for DMD.

Probabilistic Models for Scalable Knowledge Graph Construction

In the past decade, systems that extract information from millions of Internet documents have become commonplace. Knowledge graphs -- structured knowledge bases that describe entities, their attributes and the relationships between them -- are a powerful tool for understanding and organizing this vast amount of information. However, a significant obstacle to knowledge graph construction is the unreliability of the extracted information, due to noise and ambiguity in the underlying data or errors made by the extraction system and the complexity of reasoning about the dependencies between these noisy extractions. My dissertation addresses these challenges by exploiting the interdependencies between facts to improve the quality of the knowledge graph in a scalable framework. I introduce a new approach called knowledge graph identification (KGI), which resolves the entities, attributes and relationships in the knowledge graph by incorporating uncertain extractions from multiple sources, entity co-references, and ontological constraints. I define a probability distribution over possible knowledge graphs and infer the most probable knowledge graph using a combination of probabilistic and logical reasoning. Such probabilistic models are frequently dismissed due to scalability concerns, but my implementation of KGI maintains tractable performance on large problems through the use of hinge-loss Markov random fields, which have a convex inference objective. This allows the inference of large knowledge graphs using 4M facts and 20M ground constraints in 2 hours. To further scale the solution, I develop a distributed approach to the KGI problem which runs in parallel across multiple machines, reducing inference time by 90%. Finally, I extend my model to the streaming setting, where a knowledge graph is continuously updated by incorporating newly extracted facts. I devise a general approach for approximately updating inference in convex probabilistic models, and quantify the approximation error by defining and bounding inference regret for online models. Together, my work retains the attractive features of probabilistic models while providing the scalability necessary for large-scale knowledge graph construction. These models have been applied on a number of real-world knowledge graph projects, including the NELL project at Carnegie Mellon and the Google Knowledge Graph.

Decoherence models for discrete-time quantum walks and their application to neutral atom experiments

We discuss decoherence in discrete-time quantum walks in terms of a phenomenological model that distinguishes spin and spatial decoherence. We identify the dominating mechanisms that affect quantum-walk experiments realized with neutral atoms walking in an optical lattice. From the measured spatial distributions, we determine with good precision the amount of decoherence per step, which provides a quantitative indication of the quality of our quantum walks. In particular, we find that spin decoherence is the main mechanism responsible for the loss of coherence in our experiment. We also find that the sole observation of ballistic-instead of diffusive-expansion in position space is not a good indicator of the range of coherent delocalization. We provide further physical insight by distinguishing the effects of short- and long-time spin dephasing mechanisms. We introduce the concept of coherence length in the discrete-time quantum walk, which quantifies the range of spatial coherences. Unexpectedly, we find that quasi-stationary dephasing does not modify the local properties of the quantum walk, but instead affects spatial coherences. For a visual representation of decoherence phenomena in phase space, we have developed a formalism based on a discrete analogue of the Wigner function. We show that the effects of spin and spatial decoherence differ dramatically in momentum space.