Efficient Image and Video Representations for Retrieval

Image (Video) retrieval is an interesting problem of retrieving images (videos) similar to the query. Images (Videos) are represented in an input (feature) space and similar images (videos) are obtained by finding nearest neighbors in the input representation space. Numerous input representations both in real valued and binary space have been proposed for conducting faster retrieval. In this thesis, we present techniques that obtain improved input representations for retrieval in both supervised and unsupervised settings for images and videos. Supervised retrieval is a well known problem of retrieving same class images of the query. We address the practical aspects of achieving faster retrieval with binary codes as input representations for the supervised setting in the first part, where binary codes are used as addresses into hash tables. In practice, using binary codes as addresses does not guarantee fast retrieval, as similar images are not mapped to the same binary code (address). We address this problem by presenting an efficient supervised hashing (binary encoding) method that aims to explicitly map all the images of the same class ideally to a unique binary code. We refer to the binary codes of the images as `Semantic Binary Codes' and the unique code for all same class images as `Class Binary Code'. We also propose a new class­ based Hamming metric that dramatically reduces the retrieval times for larger databases, where only hamming distance is computed to the class binary codes. We also propose a Deep semantic binary code model, by replacing the output layer of a popular convolutional Neural Network (AlexNet) with the class binary codes and show that the hashing functions learned in this way outperforms the state­ of ­the art, and at the same time provide fast retrieval times. In the second part, we also address the problem of supervised retrieval by taking into account the relationship between classes. For a given query image, we want to retrieve images that preserve the relative order i.e. we want to retrieve all same class images first and then, the related classes images before different class images. We learn such relationship aware binary codes by minimizing the similarity between inner product of the binary codes and the similarity between the classes. We calculate the similarity between classes using output embedding vectors, which are vector representations of classes. Our method deviates from the other supervised binary encoding schemes as it is the first to use output embeddings for learning hashing functions. We also introduce new performance metrics that take into account the related class retrieval results and show significant gains over the state­ of­ the art. High Dimensional descriptors like Fisher Vectors or Vector of Locally Aggregated Descriptors have shown to improve the performance of many computer vision applications including retrieval. In the third part, we will discuss an unsupervised technique for compressing high dimensional vectors into high dimensional binary codes, to reduce storage complexity. In this approach, we deviate from adopting traditional hyperplane hashing functions and instead learn hyperspherical hashing functions. The proposed method overcomes the computational challenges of directly applying the spherical hashing algorithm that is intractable for compressing high dimensional vectors. A practical hierarchical model that utilizes divide and conquer techniques using the Random Select and Adjust (RSA) procedure to compress such high dimensional vectors is presented. We show that our proposed high dimensional binary codes outperform the binary codes obtained using traditional hyperplane methods for higher compression ratios. In the last part of the thesis, we propose a retrieval based solution to the Zero shot event classification problem - a setting where no training videos are available for the event. To do this, we learn a generic set of concept detectors and represent both videos and query events in the concept space. We then compute similarity between the query event and the video in the concept space and videos similar to the query event are classified as the videos belonging to the event. We show that we significantly boost the performance using concept features from other modalities.

Quantum Gate and Quantum State Preparation through Neighboring Optimal Control

Successful implementation of fault-tolerant quantum computation on a system of qubits places severe demands on the hardware used to control the many-qubit state. It is known that an accuracy threshold Pa exists for any quantum gate that is to be used for such a computation to be able to continue for an unlimited number of steps. Specifically, the error probability Pe for such a gate must fall below the accuracy threshold: Pe < Pa. Estimates of Pa vary widely, though Pa ∼ 10−4 has emerged as a challenging target for hardware designers. I present a theoretical framework based on neighboring optimal control that takes as input a good quantum gate and returns a new gate with better performance. I illustrate this approach by applying it to a universal set of quantum gates produced using non-adiabatic rapid passage. Performance improvements are substantial comparing to the original (unimproved) gates, both for ideal and non-ideal controls. Under suitable conditions detailed below, all gate error probabilities fall by 1 to 4 orders of magnitude below the target threshold of 10−4. After applying the neighboring optimal control theory to improve the performance of quantum gates in a universal set, I further apply the general control theory in a two-step procedure for fault-tolerant logical state preparation, and I illustrate this procedure by preparing a logical Bell state fault-tolerantly. The two-step preparation procedure is as follow: Step 1 provides a one-shot procedure using neighboring optimal control theory to prepare a physical qubit state which is a high-fidelity approximation to the Bell state |β01⟩ = 1/√2(|01⟩ + |10⟩). I show that for ideal (non-ideal) control, an approximate |β01⟩ state could be prepared with error probability ϵ ∼ 10−6 (10−5) with one-shot local operations. Step 2 then takes a block of p pairs of physical qubits, each prepared in |β01⟩ state using Step 1, and fault-tolerantly prepares the logical Bell state for the C4 quantum error detection code.

The Impact of Motor Learning on Motor Behavior and Cortical Dynamics in a Complex Stressful Social Environment

An economy of effort is a core characteristic of highly skilled motor performance often described as being effortless or automatic. Electroencephalographic (EEG) evaluation of cortical activity in elite performers has consistently revealed a reduction in extraneous associative cortical activity and an enhancement of task-relevant cortical processes. However, this has only been demonstrated under what are essentially practice-like conditions. Recently it has been shown that cerebral cortical activity becomes less efficient when performance occurs in a stressful, complex social environment. This dissertation examines the impact of motor skill training or practice on the EEG cortical dynamics that underlie performance in a stressful, complex social environment. Sixteen ROTC cadets participated in head-to-head pistol shooting competitions before and after completing nine sessions of skill training over three weeks. Spectral power increased in the theta frequency band and decreased in the low alpha frequency band after skill training. EEG Coherence increased in the left frontal region and decreased in the left temporal region after the practice intervention. These suggest a refinement of cerebral cortical dynamics with a reduction of task extraneous processing in the left frontal region and an enhancement of task related processing in the left temporal region consistent with the skill level reached by participants. Partitioning performance into ‘best’ and ‘worst’ based on shot score revealed that deliberate practice appears to optimize cerebral cortical activity of ‘best’ performances which are accompanied by a reduction in task-specific processes reflected by increased high-alpha power, while ‘worst’ performances are characterized by an inappropriate reduction in task-specific processing resulting in a loss of focus reflected by higher high-alpha power after training when compared to ‘best’ performances. Together, these studies demonstrate the power of experience afforded by practice, as a controllable factor, to promote resilience of cerebral cortical efficiency in complex environments.

THE INFLUENCE OF MOTIVATION ON EMOTION REGULATION AND MOTOR PERFORMANCE: EXAMINATION OF A NEURO-AFFECTIVE MODEL

Mental stress is known to disrupt the execution of motor performance and can lead to decrements in the quality of performance, however, individuals have shown significant differences regarding how fast and well they can perform a skilled task according to how well they can manage stress and emotion. The purpose of this study was to advance our understanding of how the brain modulates emotional reactivity under different motivational states to achieve differential performance in a target shooting task that requires precision visuomotor coordination. In order to study the interactions in emotion regulatory brain areas (i.e. the ventral striatum, amygdala, prefrontal cortex) and the autonomic nervous system, reward and punishment interventions were employed and the resulting behavioral and physiological responses contrasted to observe the changes in shooting performance (i.e. shooting accuracy and stability of aim) and neuro-cognitive processes (i.e. cognitive load and reserve) during the shooting task. Thirty-five participants, aged 18 to 38 years, from the Reserve Officers’ Training Corp (ROTC) at the University of Maryland were recruited to take 30 shots at a bullseye target in three different experimental conditions. In the reward condition, $1 was added to their total balance for every 10-point shot. In the punishment condition, $1 was deducted from their total balance if they did not hit the 10-point area. In the neutral condition, no money was added or deducted from their total balance. When in the reward condition, which was reportedly most enjoyable and least stressful of the conditions, heart rate variability was found to be positively related to shooting scores, inversely related to variability in shooting performance and positively related to alpha power (i.e. less activation) in the left temporal region. In the punishment (and most stressful) condition, an increase in sympathetic response (i.e. increased LF/HF ratio) was positively related to jerking movements as well as variability of placement (on the target) in the shots taken. This, coupled with error monitoring activity in the anterior cingulate cortex, suggests evaluation of self-efficacy might be driving arousal regulation, thus affecting shooting performance. Better performers showed variable, increasing high-alpha power in the temporal region during the aiming period towards taking the shot which could indicate an adaptive strategy of engagement. They also showed lower coherence during hit shots than missed shots which was coupled with reduced jerking movements and better precision and accuracy. Frontal asymmetry measures revealed possible influence of the prefrontal lobe in driving this effect in reward and neutral conditions. The possible interactions, reasons behind these findings and implications are discussed.