Spinal cord injury therapy through active learning

Therapy employing epidural electrostimulation holds great potential for improving therapy for patients with spinal cord injury (SCI) (Harkema et al., 2011). Further promising results from combined therapies using electrostimulation have also been recently obtained (e.g., van den Brand et al., 2012). The devices being developed to deliver the stimulation are highly flexible, capable of delivering any individual stimulus among a combinatorially large set of stimuli (Gad et al., 2013). While this extreme flexibility is very useful for ensuring that the device can deliver an appropriate stimulus, the challenge of choosing good stimuli is quite substantial, even for expert human experimenters. To develop a fully implantable, autonomous device which can provide useful therapy, it is necessary to design an algorithmic method for choosing the stimulus parameters. Such a method can be used in a clinical setting, by caregivers who are not experts in the neurostimulator's use, and to allow the system to adapt autonomously between visits to the clinic. To create such an algorithm, this dissertation pursues the general class of active learning algorithms that includes Gaussian Process Upper Confidence Bound (GP-UCB, Srinivas et al., 2010), developing the Gaussian Process Batch Upper Confidence Bound (GP-BUCB, Desautels et al., 2012) and Gaussian Process Adaptive Upper Confidence Bound (GP-AUCB) algorithms. This dissertation develops new theoretical bounds for the performance of these and similar algorithms, empirically assesses these algorithms against a number of competitors in simulation, and applies a variant of the GP-BUCB algorithm in closed-loop to control SCI therapy via epidural electrostimulation in four live rats. The algorithm was tasked with maximizing the amplitude of evoked potentials in the rats' left tibialis anterior muscle. These experiments show that the algorithm is capable of directing these experiments sensibly, finding effective stimuli in all four animals. Further, in direct competition with an expert human experimenter, the algorithm produced superior performance in terms of average reward and comparable or superior performance in terms of maximum reward. These results indicate that variants of GP-BUCB may be suitable for autonomously directing SCI therapy.

Neural representation of auditory temporal structure

Neurons in the songbird forebrain nucleus HVc are highly sensitive to auditory temporal context and have some of the most complex auditory tuning properties yet discovered. HVc is crucial for learning, perceiving, and producing song, thus it is important to understand the neural circuitry and mechanisms that give rise to these remarkable auditory response properties. This thesis investigates these issues experimentally and computationally.

Extracellular studies reported here compare the auditory context sensitivity of neurons in HV c with neurons in the afferent areas of field L. These demonstrate that there is a substantial increase in the auditory temporal context sensitivity from the areas of field L to HVc. Whole-cell recordings of HVc neurons from acute brain slices are described which show that excitatory synaptic transmission between HVc neurons involve the release of glutamate and the activation of both AMPA/kainate and NMDA-type glutamate receptors. Additionally, widespread inhibitory interactions exist between HVc neurons that are mediated by postsynaptic GABA_A receptors. Intracellular recordings of HVc auditory neurons in vivo provides evidence that HV c neurons encode information about temporal structure using a variety of cellular and synaptic mechanisms including syllable-specific inhibition, excitatory post-synaptic potentials with a range of different time courses, and burst-firing, and song-specific hyperpolarization.

The final part of this thesis presents two computational approaches for representing and learning temporal structure. The first method utilizes comput ational elements that are analogous to temporal combination sensitive neurons in HVc. A network of these elements can learn using local information and lateral inhibition. The second method presents a more general framework which allows a network to discover mixtures of temporal features in a continuous stream of input.

Localization of sources of human evoked responses

The evoked response, a signal present in the electro-encephalogram when specific sense modalities are stimulated with brief sensory inputs, has not yet revealed as much about brain function as it apparently promised when first recorded in the late 1940's. One of the problems has been to record the responses at a large number of points on the surface of the head; thus in order to achieve greater spatial resolution than previously attained, a 50-channel recording system was designed to monitor experiments with human visually evoked responses.

Conventional voltage versus time plots of the responses were found inadequate as a means of making qualitative studies of such a large data space. This problem was solved by creating a graphical display of the responses in the form of equipotential maps of the activity at successive instants during the complete response. In order to ascertain the necessary complexity of any models of the responses, factor analytic procedures were used to show that models characterized by only five or six independent parameters could adequately represent the variability in all recording channels.

One type of equivalent source for the responses which meets these specifications is the electrostatic dipole. Two different dipole models were studied: the dipole in a homogeneous sphere and the dipole in a sphere comprised of two spherical shells (of different conductivities) concentric with and enclosing a homogeneous sphere of a third conductivity. These models were used to determine nonlinear least squares fits of dipole parameters to a given potential distribution on the surface of a spherical approximation to the head. Numerous tests of the procedures were conducted with problems having known solutions. After these theoretical studies demonstrated the applicability of the technique, the models were used to determine inverse solutions for the evoked response potentials at various times throughout the responses. It was found that reliable estimates of the location and strength of cortical activity were obtained, and that the two models differed only slightly in their inverse solutions. These techniques enabled information flow in the brain, as indicated by locations and strengths of active sites, to be followed throughout the evoked response.

Responses to auditory stimuli in macaque lateral intraparietal area

The lateral intraparietal area (LIP) of macaque posterior parietal cortex participates in the sensorimotor transformations underlying visually guided eye movements. Area LIP has long been considered unresponsive to auditory stimulation. However, recent studies have shown that neurons in LIP respond to auditory stimuli during an auditory-saccade task, suggesting possible involvement of this area in auditory-to-oculomotor as well as visual-to-oculomotor processing. This dissertation describes investigations which clarify the role of area LIP in auditory-to-oculomotor processing.

Extracellular recordings were obtained from a total of 332 LIP neurons in two macaque monkeys, while the animals performed fixation and saccade tasks involving auditory and visual stimuli. No auditory activity was observed in area LIP before animals were trained to make saccades to auditory stimuli, but responses to auditory stimuli did emerge after auditory-saccade training. Auditory responses in area LIP after auditory-saccade training were significantly stronger in the context of an auditory-saccade task than in the context of a fixation task. Compared to visual responses, auditory responses were also significantly more predictive of movement-related activity in the saccade task. Moreover, while visual responses often had a fast transient component, responses to auditory stimuli in area LIP tended to be gradual in onset and relatively prolonged in duration.

Overall, the analyses demonstrate that responses to auditory stimuli in area LIP are dependent on auditory-saccade training, modulated by behavioral context, and characterized by slow-onset, sustained response profiles. These findings suggest that responses to auditory stimuli are best interpreted as supramodal (cognitive or motor) responses, rather than as modality-specific sensory responses. Auditory responses in area LIP seem to reflect the significance of auditory stimuli as potential targets for eye movements, and may differ from most visual responses in the extent to which they arc abstracted from the sensory parameters of the stimulus.

Analysis of temporal structure in spike trains of visual cortical area MT

The temporal structure of neuronal spike trains in the visual cortex can provide detailed information about the stimulus and about the neuronal implementation of visual processing. Spike trains recorded from the macaque motion area MT in previous studies (Newsome et al., 1989a; Britten et al., 1992; Zohary et al., 1994) are analyzed here in the context of the dynamic random dot stimulus which was used to evoke them. If the stimulus is incoherent, the spike trains can be highly modulated and precisely locked in time to the stimulus. In contrast, the coherent motion stimulus creates little or no temporal modulation and allows us to study patterns in the spike train that may be intrinsic to the cortical circuitry in area MT. Long gaps in the spike train evoked by the preferred direction motion stimulus are found, and they appear to be symmetrical to bursts in the response to the anti-preferred direction of motion. A novel cross-correlation technique is used to establish that the gaps are correlated between pairs of neurons. Temporal modulation is also found in psychophysical experiments using a modified stimulus. A model is made that can account for the temporal modulation in terms of the computational theory of biological image motion processing. A frequency domain analysis of the stimulus reveals that it contains a repeated power spectrum that may account for psychophysical and electrophysiological observations.

Some neurons tend to fire bursts of action potentials while others avoid burst firing. Using numerical and analytical models of spike trains as Poisson processes with the addition of refractory periods and bursting, we are able to account for peaks in the power spectrum near 40 Hz without assuming the existence of an underlying oscillatory signal. A preliminary examination of the local field potential reveals that stimulus-locked oscillation appears briefly at the beginning of the trial.

Biophysical and network mechanisms of high frequency extracellular potentials in the rat hippocampus

A fundamental question in neuroscience is how distributed networks of neurons communicate and coordinate dynamically and specifically. Several models propose that oscillating local networks can transiently couple to each other through phase-locked firing. Coherent local field potentials (LFP) between synaptically connected regions is often presented as evidence for such coupling. The physiological correlates of LFP signals depend on many anatomical and physiological factors, however, and how the underlying neural processes collectively generate features of different spatiotemporal scales is poorly understood. High frequency oscillations in the hippocampus, including gamma rhythms (30-100 Hz) that are organized by the theta oscillations (5-10 Hz) during active exploration and REM sleep, as well as sharp wave-ripples (SWRs, 140-200 Hz) during immobility or slow wave sleep, have each been associated with various aspects of learning and memory. Deciphering their physiology and functional consequences is crucial to understanding the operation of the hippocampal network.

We investigated the origins and coordination of high frequency LFPs in the hippocampo-entorhinal network using both biophysical models and analyses of large-scale recordings in behaving and sleeping rats. We found that the synchronization of pyramidal cell spikes substantially shapes, or even dominates, the electrical signature of SWRs in area CA1 of the hippocampus. The precise mechanisms coordinating this synchrony are still unresolved, but they appear to also affect CA1 activity during theta oscillations. The input to CA1, which often arrives in the form of gamma-frequency waves of activity from area CA3 and layer 3 of entorhinal cortex (EC3), did not strongly influence the timing of CA1 pyramidal cells. Rather, our data are more consistent with local network interactions governing pyramidal cells' spike timing during the integration of their inputs. Furthermore, the relative timing of input from EC3 and CA3 during the theta cycle matched that found in previous work to engage mechanisms for synapse modification and active dendritic processes. Our work demonstrates how local networks interact with upstream inputs to generate a coordinated hippocampal output during behavior and sleep, in the form of theta-gamma coupling and SWRs.

Rydberg states of diatomic and polyatomic molecules using model potentials

A simple model potential is used to calculate Rydberg series for the molecules: nitrogen, oxygen, nitric oxide, carbon monoxide, carbon dioxide, nitrogen dioxide, nitrous oxide, acetylene, formaldehyde, formic acid, diazomethane, ketene, ethylene, allene, acetaldehyde, propyne, acrolein, dimethyl ether, 1, 3-butadiene, 2-butene, and benzene. The model potential for a molecule is taken as the sum of atomic potentials, which are calibrated to atomic data and contain no further parameters. Our results agree with experimentally measured values to within 5-10% in all cases. The results of these calculations are applied to many unresolved problems connected with the above molecules. Some of the more notable of these problems are the reassignment of states in carbon monoxide, the first ionization potential of nitrogen dioxide, the interpretation of the V state in ethylene, and the mystery bands in substituted ethylenes, the identification of the R and R’ series in benzene and the determination of the orbital scheme in benzene from electron impact data.

Conditioned responses in the reticular formation

Unit activity was recorded from the midbrain and pons of 40 freely moving rats in an appetitive classical conditioning situation. Responses to auditory stimuli were observed from 100 units before and during a conditioning procedure in which presentation of food occurred 1 sec after the onset of the auditory stimulus. Conditioned unit responses (i.e., spike rate accelerations or decelerations) were considered to be positive when 1) no similar responses appeared prior to conditioning, and 2) latencies were equal to or less than those of sensory responses derived from the inferior colliculus. Such short latency conditioned unit responses were recorded from 11 probes located in the mid-lateral pert of the ventral region of the brain stem. This region was differentiated from paramedian, far lateral and dorsal parts of the brain stem reticular formation. Conditioned unit responses of considerably longer latencies were recorded from 76 probe located in these other regions. Among the longer latency responses interesting differences appeared in experiments conducted after the first conditioning series was completed. With additional training, units in the "reticular activating system" of midbrain and pons tended to yield stabilized responses in the early portion of the CS-US interval closely related in time to the orientation responses evoked by the CS. In contrast, the responses of units in the limbic midbrain tended to stabilize in the later part of the CS-US interval closely related in time to preparatory responses tied to the US. During extinction when the auditory stimulus was no longer followed by presentation of food, many of the responses were reduced to their pre-conditioning levels. However, there was a tendency for units which had displayed short latency responses on the first conditioning day to be more resistant to extinction than units which had displayed longer latency conditioned responses. The data were interpreted as indicating a local correlate of learning in the reticular formation of midbrain end pons and a separation of the midbrain system into at least two areas: 1) the classical "reticular activating system" related to orienting reactions, and 2) the limbic midbrain areas related to drives and rewards. Because the ventral and mid-lateral area with very short latency conditioned responses was not clearly tied to either of these; it was considered as possibly representing a third division.