Latent variable models for neural data analysis

The brain is perhaps the most complex system to have ever been subjected to rigorous scientific investigation. The scale is staggering: over 10^11 neurons, each making an average of 10^3 synapses, with computation occurring on scales ranging from a single dendritic spine, to an entire cortical area. Slowly, we are beginning to acquire experimental tools that can gather the massive amounts of data needed to characterize this system. However, to understand and interpret these data will also require substantial strides in inferential and statistical techniques. This dissertation attempts to meet this need, extending and applying the modern tools of latent variable modeling to problems in neural data analysis.

It is divided into two parts. The first begins with an exposition of the general techniques of latent variable modeling. A new, extremely general, optimization algorithm is proposed - called Relaxation Expectation Maximization (REM) - that may be used to learn the optimal parameter values of arbitrary latent variable models. This algorithm appears to alleviate the common problem of convergence to local, sub-optimal, likelihood maxima. REM leads to a natural framework for model size selection; in combination with standard model selection techniques the quality of fits may be further improved, while the appropriate model size is automatically and efficiently determined. Next, a new latent variable model, the mixture of sparse hidden Markov models, is introduced, and approximate inference and learning algorithms are derived for it. This model is applied in the second part of the thesis.

The second part brings the technology of part I to bear on two important problems in experimental neuroscience. The first is known as spike sorting; this is the problem of separating the spikes from different neurons embedded within an extracellular recording. The dissertation offers the first thorough statistical analysis of this problem, which then yields the first powerful probabilistic solution. The second problem addressed is that of characterizing the distribution of spike trains recorded from the same neuron under identical experimental conditions. A latent variable model is proposed. Inference and learning in this model leads to new principled algorithms for smoothing and clustering of spike data.

The photochemistry of dioxorhenium(V)

The excited-state properties of trans-ReO₂(py)₄⁺ (ReO₂⁺) in acetonitrile solution have been investigated. The excited-state absorption spectrum of ReO₂⁺ is dominated by bleaching of the ground state MLCT and d-d systems. The reduction potential of ReO₂^2+/+* is estimated from emission and electrochemical data to be -0.7 V (SSCE). The ReO₂⁺ excited state efficiently reduces methylviologen and other pyridinium and olefin acceptors. The resulting Re(VI) species oxidizes secondary alcohols and silanes. Acetophenone is the product of sec-phenethyl alcohol oxidation.

The emission properties of ReO₂⁺ in aqueous solutions of anionic and nonionic surfactants have been investigated. The emission and absorption maxima of ReO₂⁺ are dependent on the water content of its environment. Emission lifetimes vary over four orders of magnitude upon shifting from aqueous to nonaqueous environments. The emission lifetime has a large (8.6) isotope effect (k(H₂O)/k(D₂O)) that reflects its sensitivity towards the environment. These properties have been used to develop a model for the interactions of ReO₂⁺ with sodium dodecyl sulfate (SDS). A hydrophobic ReO₂⁺ derivative, ReO₂(3-Ph-py)₄⁺, has been used to probe micelles of nonionic surfactants, and these results are consistent with those obtained with SDS.

The emission properties of ReO₂⁺ in Nafion perfluorosulfonated membranes have been investigated. Absorption and emission spectroscopy indicate that the interior of the membrane is quite polar, similar to ethylene glycol. Two well-resolved emission components show different lifetimes and different isotope effects, indicative of varying degrees of solvent accessibility. These components are taken as evidence for chemically distinct regions in the polymer film, assigned as the interfacial region and the ion cluster region.

The unsubstituted pyridine complex shows monophasic, τ = 1.7 µs, emission decay when bound to calf thymus DNA. Switching to the 3-Ph-py complex yields a biphasic emission decay (τ₁ = 2.4 µs, τ₂ = 10 µs) indicative of an additional, solvent-inaccessible binding mode. Photoinduced electron transfer to methylviologen leads to oxidative cleavage of the DNA as detected by gel electrophoresis. Electrochemical and spectrophotometric techniques used with organic substrates also can be used to monitor the oxidation of DNA. Abstraction of the ribose 4' hydrogen by ReO₂²⁺ is a possible mechanism.

Influence of composition on the structure and properties of Fe-Pd-P and Ni-Pd-P amorphous alloys

Ternary alloys of nickel-palladium-phosphorus and iron-palladium- phosphorus containing 20 atomic % phosphorus were rapidly quenched from the liquid state. The structure of the quenched alloys was investigated by X-ray diffraction. Broad maxima in the diffraction patterns, indicative of a glass-like structure, were obtained for 13 to 73 atomic % nickel and 13 to 44 atomic % iron, with palladium adding up to 80%.

Radial distribution functions were computed from the diffraction data and yielded average interatomic distances and coordination numbers. The structure of the amorphous alloys could be explained in terms of structural units analogous to those existing in the crystalline Pd₃P, Ni₃P and Fe₃P phases, with iron or nickel substituting for palladium. A linear relationship between interatomic distances and composition, similar to Vegard's law, was shown for these metallic glasses.

Electrical resistivity measurements showed that the quenched alloys were metallic. Measurements were performed from liquid helium temperatures (4.2°K) up to the vicinity of the melting points (900°K- 1000°K). The temperature coefficient in the glassy state was very low, of the order of 10^-4/°K. A resistivity minimum was found at low temperature, varying between 9°K and 14°K for Ni_x-Pd_80-x -P₂₀ and between 17°K and 96°K for Fe_x-Pd_80-x -P₂₀, indicating the presence of a Kondo effect. Resistivity measurements, with a constant heating rate of about 1.5°C/min,showed progressive crystallization above approximately 600°K.

The magnetic moments of the amorphous Fe-Pd-P alloys were measured as a function of magnetic field and temperature. True ferromagnetism was found for the alloys Fe₃₂-Pd₄₈-P₂₀ and Fe₄₄-Pd₃₆-P₂₀ with Curie points at 165° K and 380° K respectively. Extrapolated values of the saturation magnetic moments to 0° K were 1.70 µ_B and 2.10 µ_B respectively. The amorphous alloy Fe₂₃-Pd₅₇-P₂₀ was assumed to be superparamagnetic. The experimental data indicate that phosphorus contributes to the decrease of moments by electron transfer, whereas palladium atoms probably have a small magnetic moment. A preliminary investigation of the Ni-Pd-P amorphous alloys showed that these alloys are weakly paramagnetic.

Formal methods for control synthesis in partially observed environments : application to autonomous robotic manipulation

Modern robots are increasingly expected to function in uncertain and dynamically challenging environments, often in proximity with humans. In addition, wide scale adoption of robots requires on-the-fly adaptability of software for diverse application. These requirements strongly suggest the need to adopt formal representations of high level goals and safety specifications, especially as temporal logic formulas. This approach allows for the use of formal verification techniques for controller synthesis that can give guarantees for safety and performance. Robots operating in unstructured environments also face limited sensing capability. Correctly inferring a robot's progress toward high level goal can be challenging.

This thesis develops new algorithms for synthesizing discrete controllers in partially known environments under specifications represented as linear temporal logic (LTL) formulas. It is inspired by recent developments in finite abstraction techniques for hybrid systems and motion planning problems. The robot and its environment is assumed to have a finite abstraction as a Partially Observable Markov Decision Process (POMDP), which is a powerful model class capable of representing a wide variety of problems. However, synthesizing controllers that satisfy LTL goals over POMDPs is a challenging problem which has received only limited attention.

This thesis proposes tractable, approximate algorithms for the control synthesis problem using Finite State Controllers (FSCs). The use of FSCs to control finite POMDPs allows for the closed system to be analyzed as finite global Markov chain. The thesis explicitly shows how transient and steady state behavior of the global Markov chains can be related to two different criteria with respect to satisfaction of LTL formulas. First, the maximization of the probability of LTL satisfaction is related to an optimization problem over a parametrization of the FSC. Analytic computation of gradients are derived which allows the use of first order optimization techniques.

The second criterion encourages rapid and frequent visits to a restricted set of states over infinite executions. It is formulated as a constrained optimization problem with a discounted long term reward objective by the novel utilization of a fundamental equation for Markov chains - the Poisson equation. A new constrained policy iteration technique is proposed to solve the resulting dynamic program, which also provides a way to escape local maxima.

The algorithms proposed in the thesis are applied to the task planning and execution challenges faced during the DARPA Autonomous Robotic Manipulation - Software challenge.

Resonance cones and mode conversion in a warm magnetized bounded plasma

The warm plasma resonance cone structure of the quasistatic field produced by a gap source in a bounded magnetized slab plasma is determined theoretically. This is initially determined for a homogeneous or mildly inhomogeneous plasma with source frequency lying between the lower hybrid frequency and the plasma frequency. It is then extended to the complicated case of an inhomogeneous plasma with two internal lower hybrid layers present, which is of interest to radio frequency heating of plasmas.

In the first case, the potential is obtained as a sum of multiply reflected warm plasma resonance cones, each of which has a similar structure, but a different size, amplitude, and position. An important interference between nearby multiply-reflected resonance cones is found. The cones are seen to spread out as they move away from the source, so that this interference increases and the individual resonance cones become obscured far away from the source.

In the second case, the potential is found to be expressible as a sum of multiply-reflected, multiply-tunnelled, and mode converted resonance cones, each of which has a unique but similar structure. The effects of both collisional and collisionless damping are included and their effects on the decay of the cone structure studied. Various properties of the cones such as how they move into and out of the hybrid layers, through the evanescent region, and transform at the hybrid layers are determined. It is found that cones can tunnel through the evanescent layer if the layer is thin, and the effect of the thin evanescent layer is to subdue the secondary maxima of cone relative to the main peak, while slightly broadening the main peak and shifting it closer to the cold plasma cone line.

Energy theorems for quasistatic fields are developed and applied to determine the power flow and absorption along the individual cones. This reveals the points of concentration of the flow and the various absorption mechanisms.

Propagation of the fast magnetosonic wave in a tokamak plasma

The propagation of the fast magnetosonic wave in a tokamak plasma has been investigated at low power, between 10 and 300 watts, as a prelude to future heating experiments.

The attention of the experiments has been focused on the understanding of the coupling between a loop antenna and a plasma-filled cavity. Special emphasis has been given to the measurement of the complex loading impedance of the plasma. The importance of this measurement is that once the complex loading impedance of the plasma is known, a matching network can be designed so that the r.f. generator impedance can be matched to one of the cavity modes, thus delivering maximum power to the plasma. For future heating experiments it will be essential to be able to match the generator impedance to a cavity mode in order to couple the r.f. energy efficiently to the plasma.

As a consequence of the complex impedance measurements, it was discovered that the designs of the transmitting antenna and the impedance matching network are both crucial. The losses in the antenna and the matching network must be kept below the plasma loading in order to be able to detect the complex plasma loading impedance. This is even more important in future heating experiments, because the fundamental basis for efficient heating before any other consideration is to deliver more energy into the plasma than is dissipated in the antenna system.

The characteristics of the magnetosonic cavity modes are confirmed by three different methods. First, the cavity modes are observed as voltage maxima at the output of a six-turn receiving probe. Second, they also appear as maxima in the input resistance of the transmitting antenna. Finally, when the real and imaginary parts of the measured complex input impedance of the antenna are plotted in the complex impedance plane, the resulting curves are approximately circles, indicating a resonance phenomenon.

The observed plasma loading resistances at the various cavity modes are as high as 3 to 4 times the basic antenna resistance (~ .4 Ω). The estimated cavity Q’s were between 400 and 700. This means that efficient energy coupling into the tokamak and low losses in the antenna system are possible.

Superconductivity, spin-glass properties, and ferromagnetism in amorphous La-Gd-Au alloys

The superconducting and magnetic properties of splat cooled amorphous alloys of composition (La_100-xGd_x)₈₀Au₂₀ (0 ≤ x ≤ 100) have been studied. The La₈₀Au₂₀ alloys are ideal type II super-conductors (critical temperature T_c = 3.5° K ). The concentration range (x less than 1) where superconductivity and spin-glass freezing n1ight coexist has been studied in detail. The spin-glass alloys (0 less than x less than 70) exhibit susceptibility maxima and thermomagnetic history effects. In the absence of complications due to crystal field and enhanced matrix effects, a phenomenological model is proposed in which the magnetic clusters are treated as single spin entities interacting via random forces using the molecular field approach. The fundamental parameters (such as the strength of the forces and the size of clusters) can be deduced from magnetization measurements. The remanent magnetization is shown to arise from an interplay of the RKKY and dipolar forces. Magnetoresistivity results are found to be consistent with the aforementioned picture. The nature of magnetic interactions in an amorphous matrix is also discussed. The moment per Gd atom (7µ_B) is found to be constant and close to that of the crystalline value throughout the concentration range investigated. Finally, a detail study is made of the critical phenomena and magnetic properties of the amorphous ferromagnet: Gd₈₀Au₂₀. The results are compared with recent theories on amorphous magnetism.

The electrical and magnetic properties of amorphous Pd-Cu-P alloys

The amorphous phases of the Pd-Cu-P system has been obtained using the technique of rapidly quenching from the liquid state. Broad maxima in the diffraction pattern were obtained in the X-ray diffraction studies which are indicative of a glass-like structure. The composition range over which the amorphous solid phase is retained for the Pd-Cu-P system is (Pd_100-xCu_x)₈₀P₂₀ with 10 ≤ x ≤ 50 and (Pd₆₅Cu₃₅)_100-yP_y with 15 ≤ y ≤ 24 and (Pd₆₀Cu₄₀)_100-yP_y with 15 ≤ y ≤ 24.

The electrical resistivity for the Pd-Cu-P alloys decreases with temperature as T² at low temperatures and as T at high temperatures up to the crystallization temperature. The structural scattering model of the resistivity proposed by Sinha and the spin-fluctuation resistivity model proposed by Hasegawa are re-examined in the light of the similarity of this result to the Pt-Ni-P and Pd-Ni-P systems. Objections are raised to these interpretations of the resistivity results and an alternate model is proposed consistent with the new results on Pd-Cu-P and the observation of similar effects in crystalline transition metal alloys. The observed negative temperature coefficients of resistivity in these amorphous alloys are thus interpreted as being due to the modification of the density of states with temperature through the electron-phonon interaction. The weak Pauli paramagnetism of the Pd-Cu-P, Pt-Ni-P and Pd-Ni-P alloys is interpreted as being modifications of the transition d-states as a result of the formation of strong transition metal-metalloid bonds rather than a large transfer of electrons from the glass former atoms (P in this case) to the d-band of the transition metal in a rigid band picture.

Observations of 1-6 MeV Jovian electrons at 1 AU and a study of their propagation

The 1-6 MeV electron flux at 1 AU has been measured for the time period October 1972 to December 1977 by the Caltech Electron/Isotope Spectrometers on the IMP-7 and IMP-8 satellites. The non-solar interplanetary electron flux reported here covered parts of five synodic periods. The 88 Jovian increases identified in these five synodic periods were classified by their time profiles. The fall time profiles were consistent with an exponential fall with τ ≈ 4-9 days. The rise time profiles displayed a systematic variation over the synodic period. Exponential rise time profiles with τ ≈ 1-3 days tended to occur in the time period before nominal connection, diffusive profiles predicted by the convection-diffusion model around nominal connection, and abrupt profiles after nominal connection.

The times of enhancements in the magnetic field, │B│, at 1 AU showed a better correlation than corotating interaction regions (CIR's) with Jovian increases and other changes in the electron flux at 1 AU, suggesting that │B│ enhancements indicate the times that barriers to electron propagation pass Earth. Time sequences of the increases and decreases in the electron flux at 1 AU were qualitatively modeled by using the times that CIR's passed Jupiter and the times that │B│ enhancements passed Earth.

The electron data observed at 1 AU were modeled by using a convection-diffusion model of Jovian electron propagation. The synodic envelope formed by the maxima of the Jovian increases was modeled by the envelope formed by the predicted intensities at a time less than that needed to reach equilibrium. Even though the envelope shape calculated in this way was similar to the observed envelope, the required diffusion coefficients were not consistent with a diffusive process.

Three Jovian electron increases at 1 AU for the 1974 synodic period were fit with rise time profiles calculated from the convection-diffusion model. For the fits without an ambient electron background flux, the values for the diffusion coefficients that were consistent with the data were k_x = 1.0 - 2.5 x 10²¹ cm²/sec and k_y = 1.6 - 2.0 x 10²² cm²/sec. For the fits that included the ambient electron background flux, the values for the diffusion coefficients that were consistent with the data were k_x = 0.4 - 1.0 x 10²¹ cm²/sec and k_y = 0.8 - 1.3 x 10²² cm²/sec.

Plasma inverse scattering theory

The object of this report is to calculate the electron density profile of plane stratified inhomogeneous plasmas. The electron density profile is obtained through a numerical solution of the inverse scattering algorithm.

The inverse scattering algorithm connects the time dependent reflected field resulting from a δ-function field incident normally on the plasma to the inhomogeneous plasma density.

Examples show that the method produces uniquely the electron density on or behind maxima of the plasma frequency.

It is shown that the δ-function incident field used in the inverse scattering algorithm can be replaced by a thin square pulse.