For years, frequency standards have been the cornerstone of precision measurement. Among these frequency standards, atomic clocks have set records in both precision and accuracy, and have redefined the second. There is growing interest in more complex molecular systems to complement precision measurements with atoms. The rich internal structure of even the simplest diatomic molecules could provide new avenues for fundamental physics research, including searches for extensions to the Standard Model, dark matter candidates, novel forces or corrections to gravity at short distances, and tests of the variation of fundamental constants. In this thesis, we discuss the fundamental architecture for a precise molecular system based on a strongly forbidden weakly-bound to deeply-bound vibrational transition in 88Sr dimers. We discuss early studies to characterise our system and gain technical and quantum control over the experiment in anticipation of a precise metrological measurement. We, then, demonstrate a record-breaking precision for our 88Sr2 molecular clock ushering in a new era for precision measurement with clocks. Borrowing techniques from previous atomic clock architecture, we measure a ∼32 THz clock transition between two vibrational levels in the electronic ground state, achieving a fractional uncertainty of 4.6 × 10−14 in a new frequency regime. In this current iteration, our molecular clock is fundamentally limited by two-body loss lifetimes of 200 ms and light scattering induced by our high-intensity lattice. Given these limitations, we suggest improvements to combat the effects from both the lattice and two-body collisions in our 1D trap. These include technical improvements to our experiment and strategic choices of particular clock states in our ground electronic potential. We describe in-depth studies of the chemistry and polarizability behaviour of our molecule, which elucidates preferential future directions for a second generation clock system. These empirical results are substantiated by an improved theoretical picture. Ultimately, our molecular system is built in order to probe new physics and as a tool for precision measurement. Leveraging our record-precision clock and our new-found understanding of our molecule, we predict the capacity for our system to place meaningful, competitive constraints on new physics, in particular on Yukawa-type extensions to gravity. These predictions motivate improvements to our current generation clock and set the stage for future measurements with this system.

To solve real-world problems, machines are required to perceive multiple modalities and fuse the information from them. This thesis studies learning to understand and fuse multimodal information. Existing approaches follow a three-stage learning paradigm. The first stage is to train models for each modality. This process for video understanding models is usually based on supervised training, which is not scalable. Moreover, these modality-specific models are updated rather frequently nowadays with improving single-modality perception abilities. The second stage is crossmodal pretraining, which trains a model to align and fuse multiple modalities based on paired multimodal data, such as video-caption pairs. This process is resource-consuming and expensive. The third stage is to further fine-tune or prompt the resulting model from the second stage towards certain downstream tasks. The key bottleneck of conventional methods lies in the continuous feature representation used for non-textual modalities, which is usually costly to align and fuse with text. In this thesis, we investigate the representation and the fusion based on textual concepts. We propose to map non-textual modalities to textual concepts and then fuse these textual concepts using text models. We systematically study various specific methods of mapping and different architectures for fusion. The proposed methods include an end-to-end video-based text generation model with differentiable tokenization for video and audio concepts, a contrastive-model-based architecture with zero-shot concept extractor, a deep concept injection algorithm enabling language models to solve multimodal tasks without any training, and a distant supervision framework learning concepts in a long temporal span. With our concept representation, we empirically demonstrate that without several orders of magnitude more cost for the crossmodal pretraining stage, our models are able to achieve competitive or even superior performance on downstream tasks such as video question answering, video captioning, text-video retrieval, and audio-video dialogue. We also examine the possible limitations of concept representations such as when the text quality of a dataset is poor. We believe we show a potential path towards upgradable multimodal intelligence, whose components can be easily updated towards new models or new modalities of data.

This dissertation contains two collections of analyses, both broadly focused on energy system planning, but motivated by different research objectives in distinct geographic settings. Part I – Chapters I-III – evaluates decarbonization strategies in New York. These studies are characteristic of the primary energy-related challenge faced by the Global North: How can states cost-effectively meet time-bound emissions reduction targets? A series of linear programs are developed to answer this question, culminating in the System Electrification and Capacity TRansition (SECTR) model, a high-fidelity representation of the New York State energy system that characterizes statewide emissions and allows for comparative study of various decarbonization pathways. SECTR simulations indicate that prioritizing heating and vehicle electrification alongside an expansion of instate wind and solar generation capacity allows New York to meet recently legislated climate goals more affordably than through approaches that mandate substantial low-carbon electricity targets. Additional work also explores the optimal distribution of energy infrastructure within New York to meet specified decarbonization targets, along with the value of supply-side, demand-side, and bidirectional methods of system flexibility. Part II of this dissertation – Chapters IV-VII – is concerned with the energy system challenges faced by the lowest income countries. Set in the Ethiopian Highlands, this work first aims to locate smallholder irrigated areas, as irrigation has attendant energy requirements that are larger and more likely to generate supplementary sources of revenue compared to residential demands. Here, a novel classification methodology is developed to collect labeled data, train a machine learning-based irrigation detection model, and understand the spatial extent of model applicability. Across isolated plots of land as small as 30m by 30m, the resulting model achieves >95% prediction accuracy. Further studies then explore the system planning implications of simulated electricity demands associated with these irrigated areas.

Artificial Intelligence (AI) promises to deeply alter the structure of organizations and work. This dissertation explores how firms and their human workers interact with the diffusion of automation and related technologies in the workplace, and how this informs our general understanding of organizations. I use three experiments to examine the consequences and implications of human-machine interaction in organizations. Chapter 1 studies the introduction of AI agents and human new hires into "laboratory firms" as they engage in a coordination-based game. Chapter 2 focuses on the sources of AI bias and offers practical solutions managers can adopt to limit bias. Finally, Chapter 3 studies how organizations can enjoy the benefits of AI and ensure that human collaborators remain engaged and exert effort. Overall, my dissertation develops an organizational and team perspective on the impact of workplace automation. Successful human/AI collaboration requires going beyond the technical capabilities of AI and developing a human-centered approach that incorporates firm strategies, behavioral responses, and managerial choices.

Silicon photonics is leading the advent of very-large-scale photonic integrated circuits (PICs) in which lasers, modulators, photodetectors, and multiplexers are integrated on a single chip and synchronized to enable faster data transfer both between and within highly integrated chips. Silicon photonics now extends beyond communication applications, paving new paths for many emerging applications and holding great potential in creating a compact beam projector. Compact beam steering in the visible and near-infrared spectral range is required for emerging applications such as augmented reality (AR) and virtual reality (VR) displays, optical traps for quantum information processing, biosensing, light detection and ranging (LiDAR), and free-space optical communications (FSO). Here we discuss two novel integrated beam steering platforms in the visible and near-infrared wavelengths, optical phased array (OPA) and focal plane switch array (FPSA), that can shape and steer a light beam. Previous OPA demonstrations have been mainly limited to the near-infrared spectral range due to the fabrication and material challenges imposed by the smaller wavelengths. Here we present the first active blue light phased array at the wavelength of 488 nm, leveraging a high confinement silicon nitride (Si₃N₄) platform. We randomly and sparsely place the emitters to remove grating lobes, alleviate fabrication constraints at this short wavelength and achieve a wide-angle 1D beam steering over a 50° field of view (FoV) with a full width at half maximum (FWHM) beam size of 0.17°. This demonstration is a crucial first step in realizing a non-mechanical fully-integrated beam steering device for many emerging applications. Unlike 1D steering OPA, designing 2D OPA impose a different challenge. Numerous issues arise, including complicated waveguide routing and optical crosstalk between channels. Also, creating a highly directional beam without ghost images is required to deploy visible OPAs in emerging applications. However, current demonstrations of visible OPAs, including our first demonstration, suffer from the issue of low directionality due to the presence of grating lobes, high background noise and a low percentage of power in the main beam. We demonstrate an integrated OPA that generates a highly directional beam at blue wavelengths (488 nm) by leveraging a disordered hyperuniform distribution of emitters. This exotic distribution is found in birds’ cone photoreceptor arrangements, the most uniform sampling given intrinsic packing constraints. Such unique distribution allows us to mitigate fabrication and waveguide routing constraints and achieve a beam with low background noise, high percentage of power and no grating lobes. Large-scale integration of the platform enables fully reconfigurable high-efficiency light projection across the entire visible spectrum. The novel platform offers a viable platform for next-generation applications in visible-spectrum addressing, imaging, and scanning displays. Although OPA is an invaluable device for creating a highly directional beam on a chip-scale, OPA has an inherent power consumption issue. Its architecture requires simultaneous control of all the phase shifters in the system for operation. We propose a novel silicon photonics FPSA system for beam steering with orders of magnitude lower electrical power consumption than other state-of-the-art platforms. The demonstrated system operates in the near-infrared wavelength regime; however, this can be extended into different wavelengths. Our demonstration enables low-size, weight, and power (SWaP) LiDAR for precision and autonomous robotics and optical scanners for mobile devices.

Nowadays, damage detection strategies built on the application of Artificial Neural Network tools to define models that mimic the dynamic behavior of structural systems are viral. However, a fundamental issue in developing these strategies for damage assessment is given by the unbalanced nature of the available databases for civil, mechanical, or aerospace applications, which commonly do not contain sufficient information from all the different classes that need to be identified. Unfortunately, when the aim is to classify between the healthy and damaged conditions in a structure or a generic dynamic system, it is extremely rare to have sufficient data for the unhealthy state since the system has already failed. At the same time, it is common to have plenty of data coming from the system under operational conditions. Consequently, the learning task, carried on with deep learning approaches, becomes case-dependent and tends to be specialized for a particular case and a very limited number of damage scenarios. This doctoral research presents a framework for damage classification in dynamic systems intended to overcome the limitations imposed by unbalanced datasets. In this methodology, the model's classification ability is enriched by using lower-level features derived through an improved extraction strategy that learns from a rich audio dataset how to characterize vibration traits starting from human voice recordings. This knowledge is then transferred to a target domain with much less data points, such as a structural system where the same discrimination approach is employed to classify and differentiate different health conditions. The goal is to enrich the model's ability to discriminate between classes on the audio records, presenting multiple different categories with more information to learn. The proposed methodology is validated both numerically and experimentally.

This thesis pertains to the uses of Functional Data Analysis and Machine Learning when analyzing high-dimensional structured datasets. The theme that motivates the first two chapters is the development of dimension-reduction methods in the context of functional data to advance the understanding of in-vivo measurements of neural-spike data. The last chapter addresses the analysis of survey data using machine learning techniques to identify novel risk factors for suicide in the general population. The first chapter of this thesis, "Adaptive Functional Principal Component Analysis," provides a novel method for adequately capturing modes of variation in data exhibiting sharp changes in smoothness. Our work integrates a novel scatterplot technique that adaptively smooths latent functions estimated in an FPCA framework. We are motivated to identify coordinated patterns of brain activity across multiple simultaneously-recorded neurons during motor behavior to understand the dynamics between the brain and dexterous movement. Our proposed method adequately captures the underlying biological mechanisms in our experiment, offering interpretable activation patterns when compared to standard approaches. The second chapter of our dissertation develops statistical procedures to compare the eigendecomposition from two samples of functional data. We first introduce appropriate tests for both independent and paired functions. We are motivated to test whether activation patterns in the motor cortex hold constant when a mouse performs a reaching movement repeatedly. We test all pairwise comparisons across trials and compare the distribution of the p-values against the distribution under the null. Our results suggest trial-to-trial variation in the latent activation patterns that can't be attributed to sampling noise. Our results can inform future methodology for deriving activation patterns from noisy neural spikes. The last chapter of this dissertation dives into applying Machine Learning Techniques to analyze survey data. We use the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) survey to identify novel risk factors for suicide attempts in the general population. Our analysis uses a Balanced Random Forest (BRF) approach and incorporates extreme class imbalance and survey architecture into the algorithm. We extend prior research focusing on high-risk clinical samples by identifying risk factors for suicide attempts in the general population. Our work identifies risk variables that can help guide clinical assessment and the development of suicide risk scales.

Advances in computer automation and control, compact and portable x-ray sources, and reliable and efficient detector systems over the last ten years have allowed X-Ray Determination of Residual Stress (XRDRS) measurements to become a viable method of evaluating the state of stress in metals, alloys, and ceramics. However, problems associated with incorrect XRDRS equipment operation and poor experimental technique are prevalent, necessitating better operator training and education. Therefore, an interactive computer workstation, called RS/hyper, was developed to lead the operator towards correct operating procedures and reliable experimental technique. RS/hyper teaches the operator proper machine setup, machine maintenance, radiation safety, experimental technique, theoretical understanding, and provides limited data evaluation. Graphical aids are used extensively to avoid confusion and misinterpretation during the learning process. Since RS/hyper is interactive, the operator may select the desired level of understanding on a particular topic. These topics are linked, through a hypertext interface, so that the information is presented coherently and efficiently. Compared to written texts and references, RS/hyper has been shown in preliminary tests to reduce XRDRS training and problem solving time by a factor of 16. RS/hyper will train novice users of XRDRS equipment so that the data acquired from such machines will be reliable in an industrial environment. Since the software educates the user, the results of the data will be more accurately represented before interpretation. The experienced user should find RS/hyper useful as a reference of XRDRS and related information. / Master of Science

The object of this research investigation was to establish whether or not a correlation exists between bottle B.O.D. and CO₂ liberated from a highly aerobic sewage-sludge mixture. The first part of the project consisted of reducing the number of variables in the operation to a minimum. To this end, considerable time and effort was spent finding the relationship existing between the condition of the sludge and its oxidizing potentiality. A definite strong statistical correlation was found to exist between the CO₂ liberated and the suspended solids or organic matter in the sludge. The apparatus functioned very well as could be readily seen from the consistent results obtained with the four units operating simultaneously. / Master of Science

A debate currently exists concerning whether or not oxygen uptake rate is a valid control parameter for monitoring the activated sludge process. A laboratory study was conducted to attempt to shed light on the controversy. Two bench-scale reactors were operated at steady state and under shock load. Oxygen uptake rate (OUR) was measured with the BOD bottle technique and with an on-line respirometer. The reliability of the results obtained from the BOD bottle technique was also of interest. No relationship could be deduced between effluent quality and oxygen uptake rate thereby suggesting that the latter would not be useful as a control parameter. As was concluded from the shock load data, the oxygen uptake rate varies very inconsistently at high organic loadings. It was found that the BOD bottle technique completely failed at very high organic loadings and gave meaningless results. The on-line respirometer, in spite of its high sensitivity, gave more realistic and consistent results. / M.S.